diff options
author | nobody <> | 2002-09-20 15:07:16 +0000 |
---|---|---|
committer | nobody <> | 2002-09-20 15:07:16 +0000 |
commit | 9dcc0b141cc6b94354177de13359f44691713b19 (patch) | |
tree | 4bf74b436a04aab8dafb5645a95ce289fc8c3173 /gdb/hpread.c | |
parent | 148ea30ef6e624235f31f27ee7a285c7529eb68a (diff) | |
download | gdb-9dcc0b141cc6b94354177de13359f44691713b19.zip gdb-9dcc0b141cc6b94354177de13359f44691713b19.tar.gz gdb-9dcc0b141cc6b94354177de13359f44691713b19.tar.bz2 |
This commit was manufactured by cvs2svn to create branch
'carlton_dictionary-branch'.
Cherrypick from master 2002-09-20 15:07:15 UTC Fernando Nasser <fnasser@redhat.com> ' From 2002-07-02 George Helffrich <george@gly.bris.ac.uk>':
gdb/ChangeLog
gdb/Makefile.in
gdb/ada-lang.c
gdb/c-lang.c
gdb/dbxread.c
gdb/mdebugread.c
gdb/symfile.c
gdb/symtab.c
gdb/symtab.h
Cherrypick from gdb_5_3-branch 2002-09-03 22:29:15 UTC nobody 'This commit was manufactured by cvs2svn to create branch 'gdb_5_3-branch'.':
gdb/coffread.c
gdb/dwarf2read.c
gdb/dwarfread.c
gdb/hpread.c
gdb/symfile.h
gdb/xcoffread.c
Diffstat (limited to 'gdb/hpread.c')
-rw-r--r-- | gdb/hpread.c | 6310 |
1 files changed, 6310 insertions, 0 deletions
diff --git a/gdb/hpread.c b/gdb/hpread.c new file mode 100644 index 0000000..4cc5f18 --- /dev/null +++ b/gdb/hpread.c @@ -0,0 +1,6310 @@ +/* Read hp debug symbols and convert to internal format, for GDB. + Copyright 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002 + Free Software Foundation, Inc. + + This file is part of GDB. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 59 Temple Place - Suite 330, + Boston, MA 02111-1307, USA. + + Written by the Center for Software Science at the University of Utah + and by Cygnus Support. */ + +#include "defs.h" +#include "bfd.h" +#include "gdb_string.h" +#include "hp-symtab.h" +#include "syms.h" +#include "symtab.h" +#include "symfile.h" +#include "objfiles.h" +#include "buildsym.h" +#include "complaints.h" +#include "gdb-stabs.h" +#include "gdbtypes.h" +#include "demangle.h" + +/* Private information attached to an objfile which we use to find + and internalize the HP C debug symbols within that objfile. */ + +struct hpread_symfile_info + { + /* The contents of each of the debug sections (there are 4 of them). */ + char *gntt; + char *lntt; + char *slt; + char *vt; + + /* We keep the size of the $VT$ section for range checking. */ + unsigned int vt_size; + + /* Some routines still need to know the number of symbols in the + main debug sections ($LNTT$ and $GNTT$). */ + unsigned int lntt_symcount; + unsigned int gntt_symcount; + + /* To keep track of all the types we've processed. */ + struct type **dntt_type_vector; + int dntt_type_vector_length; + + /* Keeps track of the beginning of a range of source lines. */ + sltpointer sl_index; + + /* Some state variables we'll need. */ + int within_function; + + /* Keep track of the current function's address. We may need to look + up something based on this address. */ + unsigned int current_function_value; + }; + +/* Accessor macros to get at the fields. */ +#define HPUX_SYMFILE_INFO(o) \ + ((struct hpread_symfile_info *)((o)->sym_private)) +#define GNTT(o) (HPUX_SYMFILE_INFO(o)->gntt) +#define LNTT(o) (HPUX_SYMFILE_INFO(o)->lntt) +#define SLT(o) (HPUX_SYMFILE_INFO(o)->slt) +#define VT(o) (HPUX_SYMFILE_INFO(o)->vt) +#define VT_SIZE(o) (HPUX_SYMFILE_INFO(o)->vt_size) +#define LNTT_SYMCOUNT(o) (HPUX_SYMFILE_INFO(o)->lntt_symcount) +#define GNTT_SYMCOUNT(o) (HPUX_SYMFILE_INFO(o)->gntt_symcount) +#define DNTT_TYPE_VECTOR(o) (HPUX_SYMFILE_INFO(o)->dntt_type_vector) +#define DNTT_TYPE_VECTOR_LENGTH(o) \ + (HPUX_SYMFILE_INFO(o)->dntt_type_vector_length) +#define SL_INDEX(o) (HPUX_SYMFILE_INFO(o)->sl_index) +#define WITHIN_FUNCTION(o) (HPUX_SYMFILE_INFO(o)->within_function) +#define CURRENT_FUNCTION_VALUE(o) (HPUX_SYMFILE_INFO(o)->current_function_value) + +/* Given the native debug symbol SYM, set NAMEP to the name associated + with the debug symbol. Note we may be called with a debug symbol which + has no associated name, in that case we return an empty string. + + Also note we "know" that the name for any symbol is always in the + same place. Hence we don't have to conditionalize on the symbol type. */ +#define SET_NAMESTRING(SYM, NAMEP, OBJFILE) \ + if (! hpread_has_name ((SYM)->dblock.kind)) \ + *NAMEP = ""; \ + else if (((unsigned)(SYM)->dsfile.name) >= VT_SIZE (OBJFILE)) \ + { \ + complain (&string_table_offset_complaint, (char *) symnum); \ + *NAMEP = ""; \ + } \ + else \ + *NAMEP = (SYM)->dsfile.name + VT (OBJFILE) + +/* We put a pointer to this structure in the read_symtab_private field + of the psymtab. */ + +struct symloc + { + /* The offset within the file symbol table of first local symbol for + this file. */ + + int ldsymoff; + + /* Length (in bytes) of the section of the symbol table devoted to + this file's symbols (actually, the section bracketed may contain + more than just this file's symbols). If ldsymlen is 0, the only + reason for this thing's existence is the dependency list. + Nothing else will happen when it is read in. */ + + int ldsymlen; + }; + +#define LDSYMOFF(p) (((struct symloc *)((p)->read_symtab_private))->ldsymoff) +#define LDSYMLEN(p) (((struct symloc *)((p)->read_symtab_private))->ldsymlen) +#define SYMLOC(p) ((struct symloc *)((p)->read_symtab_private)) + +/* FIXME: Shouldn't this stuff be in a .h file somewhere? */ +/* Complaints about the symbols we have encountered. */ +extern struct complaint string_table_offset_complaint; +extern struct complaint lbrac_unmatched_complaint; +extern struct complaint lbrac_mismatch_complaint; + +static struct complaint hpread_unhandled_end_common_complaint = +{ + "unhandled symbol in hp-symtab-read.c: DNTT_TYPE_COMMON/DNTT_TYPE_END.\n", 0, 0 +}; + +static struct complaint hpread_unhandled_type_complaint = +{ + "hpread_type_translate: unhandled type code.", 0, 0 +}; + +static struct complaint hpread_struct_complaint = +{ + "hpread_read_struct_type: expected SVAR type...", 0, 0 +}; + +static struct complaint hpread_array_complaint = +{ + "error in hpread_array_type.", 0, 0 +}; + +static struct complaint hpread_type_lookup_complaint = +{ + "error in hpread_type_lookup().", 0, 0 +}; + + +static struct complaint hpread_unexpected_end_complaint = +{ + "internal error in hp-symtab-read.c: Unexpected DNTT_TYPE_END kind.", 0, 0 +}; + +static struct complaint hpread_tagdef_complaint = +{ + "error processing class tagdef", 0, 0 +}; + +static struct complaint hpread_unhandled_common_complaint = +{ + "unhandled symbol in hp-symtab-read.c: DNTT_TYPE_COMMON.", 0, 0 +}; + +static struct complaint hpread_unhandled_blockdata_complaint = +{ + "unhandled symbol in hp-symtab-read.c: DNTT_TYPE_BLOCKDATA.", 0, 0 +}; + +/* To generate dumping code, uncomment this define. The dumping + itself is controlled by routine-local statics called "dumping". */ +/* #define DUMPING 1 */ + +/* To use the quick look-up tables, uncomment this define. */ +#define QUICK_LOOK_UP 1 + +/* To call PXDB to process un-processed files, uncomment this define. */ +#define USE_PXDB 1 + +/* Forward procedure declarations */ + +void hpread_symfile_init (struct objfile *); + +void do_pxdb (bfd *); + +void hpread_build_psymtabs (struct objfile *, int); + +void hpread_symfile_finish (struct objfile *); + +static union dnttentry *hpread_get_gntt (int, struct objfile *); + +static union dnttentry *hpread_get_lntt (int index, struct objfile *objfile); + + +static unsigned long hpread_get_textlow (int, int, struct objfile *, int); + +static struct partial_symtab *hpread_start_psymtab + (struct objfile *, char *, CORE_ADDR, int, + struct partial_symbol **, struct partial_symbol **); + +static struct partial_symtab *hpread_end_psymtab + (struct partial_symtab *, char **, int, int, CORE_ADDR, + struct partial_symtab **, int); + +static unsigned long hpread_get_scope_start (sltpointer, struct objfile *); + +static unsigned long hpread_get_line (sltpointer, struct objfile *); + +static CORE_ADDR hpread_get_location (sltpointer, struct objfile *); + +static void hpread_psymtab_to_symtab_1 (struct partial_symtab *); + +void hpread_psymtab_to_symtab (struct partial_symtab *); + +static struct symtab *hpread_expand_symtab + (struct objfile *, int, int, CORE_ADDR, int, + struct section_offsets *, char *); + +static int hpread_type_translate (dnttpointer); + +static struct type **hpread_lookup_type (dnttpointer, struct objfile *); + +static struct type *hpread_alloc_type (dnttpointer, struct objfile *); + +static struct type *hpread_read_enum_type + (dnttpointer, union dnttentry *, struct objfile *); + +static struct type *hpread_read_function_type + (dnttpointer, union dnttentry *, struct objfile *, int); + +static struct type *hpread_read_doc_function_type + (dnttpointer, union dnttentry *, struct objfile *, int); + +static struct type *hpread_read_struct_type + (dnttpointer, union dnttentry *, struct objfile *); + +static struct type *hpread_get_nth_template_arg (struct objfile *, int); + +static struct type *hpread_read_templ_arg_type + (dnttpointer, union dnttentry *, struct objfile *, char *); + +static struct type *hpread_read_set_type + (dnttpointer, union dnttentry *, struct objfile *); + +static struct type *hpread_read_array_type + (dnttpointer, union dnttentry *dn_bufp, struct objfile *objfile); + +static struct type *hpread_read_subrange_type + (dnttpointer, union dnttentry *, struct objfile *); + +static struct type *hpread_type_lookup (dnttpointer, struct objfile *); + +static sltpointer hpread_record_lines + (struct subfile *, sltpointer, sltpointer, struct objfile *, CORE_ADDR); + +static void hpread_process_one_debug_symbol + (union dnttentry *, char *, struct section_offsets *, + struct objfile *, CORE_ADDR, int, char *, int, int *); + +static int hpread_get_scope_depth (union dnttentry *, struct objfile *, int); + +static void fix_static_member_physnames + (struct type *, char *, struct objfile *); + +static void fixup_class_method_type + (struct type *, struct type *, struct objfile *); + +static void hpread_adjust_bitoffsets (struct type *, int); + +static dnttpointer hpread_get_next_skip_over_anon_unions + (int, dnttpointer, union dnttentry **, struct objfile *); + + +/* Global to indicate presence of HP-compiled objects, + in particular, SOM executable file with SOM debug info + Defined in symtab.c, used in hppa-tdep.c. */ +extern int hp_som_som_object_present; + +/* Static used to indicate a class type that requires a + fix-up of one of its method types */ +static struct type *fixup_class = NULL; + +/* Static used to indicate the method type that is to be + used to fix-up the type for fixup_class */ +static struct type *fixup_method = NULL; + +#ifdef USE_PXDB + +/* NOTE use of system files! May not be portable. */ + +#define PXDB_SVR4 "/opt/langtools/bin/pxdb" +#define PXDB_BSD "/usr/bin/pxdb" + +#include <stdlib.h> +#include "gdb_string.h" + +/* check for the existence of a file, given its full pathname */ +int +file_exists (char *filename) +{ + if (filename) + return (access (filename, F_OK) == 0); + return 0; +} + + +/* Translate from the "hp_language" enumeration in hp-symtab.h + used in the debug info to gdb's generic enumeration in defs.h. */ +static enum language +trans_lang (enum hp_language in_lang) +{ + if (in_lang == HP_LANGUAGE_C) + return language_c; + + else if (in_lang == HP_LANGUAGE_CPLUSPLUS) + return language_cplus; + + else if (in_lang == HP_LANGUAGE_FORTRAN) + return language_fortran; + + else + return language_unknown; +} + +static char main_string[] = "main"; + +/* Call PXDB to process our file. + + Approach copied from DDE's "dbgk_run_pxdb". Note: we + don't check for BSD location of pxdb, nor for existence + of pxdb itself, etc. + + NOTE: uses system function and string functions directly. + + Return value: 1 if ok, 0 if not */ +int +hpread_call_pxdb (const char *file_name) +{ + char *p; + int status; + int retval; + + if (file_exists (PXDB_SVR4)) + { + p = xmalloc (strlen (PXDB_SVR4) + strlen (file_name) + 2); + strcpy (p, PXDB_SVR4); + strcat (p, " "); + strcat (p, file_name); + + warning ("File not processed by pxdb--about to process now.\n"); + status = system (p); + + retval = (status == 0); + } + else + { + warning ("pxdb not found at standard location: /opt/langtools/bin\ngdb will not be able to debug %s.\nPlease install pxdb at the above location and then restart gdb.\nYou can also run pxdb on %s with the command\n\"pxdb %s\" and then restart gdb.", file_name, file_name, file_name); + + retval = 0; + } + return retval; +} /* hpread_call_pxdb */ + + +/* Return 1 if the file turns out to need pre-processing + by PXDB, and we have thus called PXDB to do this processing + and the file therefore needs to be re-loaded. Otherwise + return 0. */ +int +hpread_pxdb_needed (bfd *sym_bfd) +{ + asection *pinfo_section, *debug_section, *header_section; + unsigned int do_pxdb; + char *buf; + bfd_size_type header_section_size; + + unsigned long tmp; + unsigned int pxdbed; + + header_section = bfd_get_section_by_name (sym_bfd, "$HEADER$"); + if (!header_section) + { + return 0; /* No header at all, can't recover... */ + } + + debug_section = bfd_get_section_by_name (sym_bfd, "$DEBUG$"); + pinfo_section = bfd_get_section_by_name (sym_bfd, "$PINFO$"); + + if (pinfo_section && !debug_section) + { + /* Debug info with DOC, has different header format. + this only happens if the file was pxdbed and compiled optimized + otherwise the PINFO section is not there. */ + header_section_size = bfd_section_size (objfile->obfd, header_section); + + if (header_section_size == (bfd_size_type) sizeof (DOC_info_PXDB_header)) + { + buf = alloca (sizeof (DOC_info_PXDB_header)); + + if (!bfd_get_section_contents (sym_bfd, + header_section, + buf, 0, + header_section_size)) + error ("bfd_get_section_contents\n"); + + tmp = bfd_get_32 (sym_bfd, (bfd_byte *) (buf + sizeof (int) * 4)); + pxdbed = (tmp >> 31) & 0x1; + + if (!pxdbed) + error ("file debug header info invalid\n"); + do_pxdb = 0; + } + + else + error ("invalid $HEADER$ size in executable \n"); + } + + else + { + + /* this can be three different cases: + 1. pxdbed and not doc + - DEBUG and HEADER sections are there + - header is PXDB_header type + - pxdbed flag is set to 1 + + 2. not pxdbed and doc + - DEBUG and HEADER sections are there + - header is DOC_info_header type + - pxdbed flag is set to 0 + + 3. not pxdbed and not doc + - DEBUG and HEADER sections are there + - header is XDB_header type + - pxdbed flag is set to 0 + + NOTE: the pxdbed flag is meaningful also in the not + already pxdb processed version of the header, + because in case on non-already processed by pxdb files + that same bit in the header would be always zero. + Why? Because the bit is the leftmost bit of a word + which contains a 'length' which is always a positive value + so that bit is never set to 1 (otherwise it would be negative) + + Given the above, we have two choices : either we ignore the + size of the header itself and just look at the pxdbed field, + or we check the size and then we (for safety and paranoia related + issues) check the bit. + The first solution is used by DDE, the second by PXDB itself. + I am using the second one here, because I already wrote it, + and it is the end of a long day. + Also, using the first approach would still involve size issues + because we need to read in the contents of the header section, and + give the correct amount of stuff we want to read to the + get_bfd_section_contents function. */ + + /* decide which case depending on the size of the header section. + The size is as defined in hp-symtab.h */ + + header_section_size = bfd_section_size (objfile->obfd, header_section); + + if (header_section_size == (bfd_size_type) sizeof (PXDB_header)) /* pxdb and not doc */ + { + + buf = alloca (sizeof (PXDB_header)); + if (!bfd_get_section_contents (sym_bfd, + header_section, + buf, 0, + header_section_size)) + error ("bfd_get_section_contents\n"); + + tmp = bfd_get_32 (sym_bfd, (bfd_byte *) (buf + sizeof (int) * 3)); + pxdbed = (tmp >> 31) & 0x1; + + if (pxdbed) + do_pxdb = 0; + else + error ("file debug header invalid\n"); + } + else /*not pxdbed and doc OR not pxdbed and non doc */ + do_pxdb = 1; + } + + if (do_pxdb) + { + return 1; + } + else + { + return 0; + } +} /* hpread_pxdb_needed */ + +#endif + +/* Check whether the file needs to be preprocessed by pxdb. + If so, call pxdb. */ + +void +do_pxdb (bfd *sym_bfd) +{ + /* The following code is HP-specific. The "right" way of + doing this is unknown, but we bet would involve a target- + specific pre-file-load check using a generic mechanism. */ + + /* This code will not be executed if the file is not in SOM + format (i.e. if compiled with gcc) */ + if (hpread_pxdb_needed (sym_bfd)) + { + /*This file has not been pre-processed. Preprocess now */ + + if (hpread_call_pxdb (sym_bfd->filename)) + { + /* The call above has changed the on-disk file, + we can close the file anyway, because the + symbols will be reread in when the target is run */ + bfd_close (sym_bfd); + } + } +} + + + +#ifdef QUICK_LOOK_UP + +/* Code to handle quick lookup-tables follows. */ + + +/* Some useful macros */ +#define VALID_FILE(i) ((i) < pxdb_header_p->fd_entries) +#define VALID_MODULE(i) ((i) < pxdb_header_p->md_entries) +#define VALID_PROC(i) ((i) < pxdb_header_p->pd_entries) +#define VALID_CLASS(i) ((i) < pxdb_header_p->cd_entries) + +#define FILE_START(i) (qFD[i].adrStart) +#define MODULE_START(i) (qMD[i].adrStart) +#define PROC_START(i) (qPD[i].adrStart) + +#define FILE_END(i) (qFD[i].adrEnd) +#define MODULE_END(i) (qMD[i].adrEnd) +#define PROC_END(i) (qPD[i].adrEnd) + +#define FILE_ISYM(i) (qFD[i].isym) +#define MODULE_ISYM(i) (qMD[i].isym) +#define PROC_ISYM(i) (qPD[i].isym) + +#define VALID_CURR_FILE (curr_fd < pxdb_header_p->fd_entries) +#define VALID_CURR_MODULE (curr_md < pxdb_header_p->md_entries) +#define VALID_CURR_PROC (curr_pd < pxdb_header_p->pd_entries) +#define VALID_CURR_CLASS (curr_cd < pxdb_header_p->cd_entries) + +#define CURR_FILE_START (qFD[curr_fd].adrStart) +#define CURR_MODULE_START (qMD[curr_md].adrStart) +#define CURR_PROC_START (qPD[curr_pd].adrStart) + +#define CURR_FILE_END (qFD[curr_fd].adrEnd) +#define CURR_MODULE_END (qMD[curr_md].adrEnd) +#define CURR_PROC_END (qPD[curr_pd].adrEnd) + +#define CURR_FILE_ISYM (qFD[curr_fd].isym) +#define CURR_MODULE_ISYM (qMD[curr_md].isym) +#define CURR_PROC_ISYM (qPD[curr_pd].isym) + +#define TELL_OBJFILE \ + do { \ + if( !told_objfile ) { \ + told_objfile = 1; \ + warning ("\nIn object file \"%s\":\n", \ + objfile->name); \ + } \ + } while (0) + + + +/* Keeping track of the start/end symbol table (LNTT) indices of + psymtabs created so far */ + +typedef struct +{ + int start; + int end; +} +pst_syms_struct; + +static pst_syms_struct *pst_syms_array = 0; + +static pst_syms_count = 0; +static pst_syms_size = 0; + +/* used by the TELL_OBJFILE macro */ +static boolean told_objfile = 0; + +/* Set up psymtab symbol index stuff */ +static void +init_pst_syms (void) +{ + pst_syms_count = 0; + pst_syms_size = 20; + pst_syms_array = (pst_syms_struct *) xmalloc (20 * sizeof (pst_syms_struct)); +} + +/* Clean up psymtab symbol index stuff */ +static void +clear_pst_syms (void) +{ + pst_syms_count = 0; + pst_syms_size = 0; + xfree (pst_syms_array); + pst_syms_array = 0; +} + +/* Add information about latest psymtab to symbol index table */ +static void +record_pst_syms (int start_sym, int end_sym) +{ + if (++pst_syms_count > pst_syms_size) + { + pst_syms_array = (pst_syms_struct *) xrealloc (pst_syms_array, + 2 * pst_syms_size * sizeof (pst_syms_struct)); + pst_syms_size *= 2; + } + pst_syms_array[pst_syms_count - 1].start = start_sym; + pst_syms_array[pst_syms_count - 1].end = end_sym; +} + +/* Find a suitable symbol table index which can serve as the upper + bound of a psymtab that starts at INDEX + + This scans backwards in the psymtab symbol index table to find a + "hole" in which the given index can fit. This is a heuristic!! + We don't search the entire table to check for multiple holes, + we don't care about overlaps, etc. + + Return 0 => not found */ +static int +find_next_pst_start (int index) +{ + int i; + + for (i = pst_syms_count - 1; i >= 0; i--) + if (pst_syms_array[i].end <= index) + return (i == pst_syms_count - 1) ? 0 : pst_syms_array[i + 1].start - 1; + + if (pst_syms_array[0].start > index) + return pst_syms_array[0].start - 1; + + return 0; +} + + + +/* Utility functions to find the ending symbol index for a psymtab */ + +/* Find the next file entry that begins beyond INDEX, and return + its starting symbol index - 1. + QFD is the file table, CURR_FD is the file entry from where to start, + PXDB_HEADER_P as in hpread_quick_traverse (to allow macros to work). + + Return 0 => not found */ +static int +find_next_file_isym (int index, quick_file_entry *qFD, int curr_fd, + PXDB_header_ptr pxdb_header_p) +{ + while (VALID_CURR_FILE) + { + if (CURR_FILE_ISYM >= index) + return CURR_FILE_ISYM - 1; + curr_fd++; + } + return 0; +} + +/* Find the next procedure entry that begins beyond INDEX, and return + its starting symbol index - 1. + QPD is the procedure table, CURR_PD is the proc entry from where to start, + PXDB_HEADER_P as in hpread_quick_traverse (to allow macros to work). + + Return 0 => not found */ +static int +find_next_proc_isym (int index, quick_procedure_entry *qPD, int curr_pd, + PXDB_header_ptr pxdb_header_p) +{ + while (VALID_CURR_PROC) + { + if (CURR_PROC_ISYM >= index) + return CURR_PROC_ISYM - 1; + curr_pd++; + } + return 0; +} + +/* Find the next module entry that begins beyond INDEX, and return + its starting symbol index - 1. + QMD is the module table, CURR_MD is the modue entry from where to start, + PXDB_HEADER_P as in hpread_quick_traverse (to allow macros to work). + + Return 0 => not found */ +static int +find_next_module_isym (int index, quick_module_entry *qMD, int curr_md, + PXDB_header_ptr pxdb_header_p) +{ + while (VALID_CURR_MODULE) + { + if (CURR_MODULE_ISYM >= index) + return CURR_MODULE_ISYM - 1; + curr_md++; + } + return 0; +} + +/* Scan and record partial symbols for all functions starting from index + pointed to by CURR_PD_P, and between code addresses START_ADR and END_ADR. + Other parameters are explained in comments below. */ + +/* This used to be inline in hpread_quick_traverse, but now that we do + essentially the same thing for two different cases (modules and + module-less files), it's better organized in a separate routine, + although it does take lots of arguments. pai/1997-10-08 + + CURR_PD_P is the pointer to the current proc index. QPD is the + procedure quick lookup table. MAX_PROCS is the number of entries + in the proc. table. START_ADR is the beginning of the code range + for the current psymtab. end_adr is the end of the code range for + the current psymtab. PST is the current psymtab. VT_bits is + a pointer to the strings table of SOM debug space. OBJFILE is + the current object file. */ + +static int +scan_procs (int *curr_pd_p, quick_procedure_entry *qPD, int max_procs, + CORE_ADDR start_adr, CORE_ADDR end_adr, struct partial_symtab *pst, + char *vt_bits, struct objfile *objfile) +{ + union dnttentry *dn_bufp; + int symbol_count = 0; /* Total number of symbols in this psymtab */ + int curr_pd = *curr_pd_p; /* Convenience variable -- avoid dereferencing pointer all the time */ + +#ifdef DUMPING + /* Turn this on for lots of debugging information in this routine */ + static int dumping = 0; +#endif + +#ifdef DUMPING + if (dumping) + { + printf ("Scan_procs called, addresses %x to %x, proc %x\n", start_adr, end_adr, curr_pd); + } +#endif + + while ((CURR_PROC_START <= end_adr) && (curr_pd < max_procs)) + { + + char *rtn_name; /* mangled name */ + char *rtn_dem_name; /* qualified demangled name */ + char *class_name; + int class; + + if ((trans_lang ((enum hp_language) qPD[curr_pd].language) == language_cplus) && + vt_bits[(long) qPD[curr_pd].sbAlias]) /* not a null string */ + { + /* Get mangled name for the procedure, and demangle it */ + rtn_name = &vt_bits[(long) qPD[curr_pd].sbAlias]; + rtn_dem_name = cplus_demangle (rtn_name, DMGL_ANSI | DMGL_PARAMS); + } + else + { + rtn_name = &vt_bits[(long) qPD[curr_pd].sbProc]; + rtn_dem_name = NULL; + } + + /* Hack to get around HP C/C++ compilers' insistence on providing + "_MAIN_" as an alternate name for "main" */ + if ((strcmp (rtn_name, "_MAIN_") == 0) && + (strcmp (&vt_bits[(long) qPD[curr_pd].sbProc], "main") == 0)) + rtn_dem_name = rtn_name = main_string; + +#ifdef DUMPING + if (dumping) + { + printf ("..add %s (demangled %s), index %x to this psymtab\n", rtn_name, rtn_dem_name, curr_pd); + } +#endif + + /* Check for module-spanning routines. */ + if (CURR_PROC_END > end_adr) + { + TELL_OBJFILE; + warning ("Procedure \"%s\" [0x%x] spans file or module boundaries.", rtn_name, curr_pd); + } + + /* Add this routine symbol to the list in the objfile. + Unfortunately we have to go to the LNTT to determine the + correct list to put it on. An alternative (which the + code used to do) would be to not check and always throw + it on the "static" list. But if we go that route, then + symbol_lookup() needs to be tweaked a bit to account + for the fact that the function might not be found on + the correct list in the psymtab. - RT */ + dn_bufp = hpread_get_lntt (qPD[curr_pd].isym, objfile); + if (dn_bufp->dfunc.global) + add_psymbol_with_dem_name_to_list (rtn_name, + strlen (rtn_name), + rtn_dem_name, + strlen (rtn_dem_name), + VAR_NAMESPACE, + LOC_BLOCK, /* "I am a routine" */ + &objfile->global_psymbols, + (qPD[curr_pd].adrStart + /* Starting address of rtn */ + ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile))), + 0, /* core addr?? */ + trans_lang ((enum hp_language) qPD[curr_pd].language), + objfile); + else + add_psymbol_with_dem_name_to_list (rtn_name, + strlen (rtn_name), + rtn_dem_name, + strlen (rtn_dem_name), + VAR_NAMESPACE, + LOC_BLOCK, /* "I am a routine" */ + &objfile->static_psymbols, + (qPD[curr_pd].adrStart + /* Starting address of rtn */ + ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile))), + 0, /* core addr?? */ + trans_lang ((enum hp_language) qPD[curr_pd].language), + objfile); + + symbol_count++; + *curr_pd_p = ++curr_pd; /* bump up count & reflect in caller */ + } /* loop over procedures */ + +#ifdef DUMPING + if (dumping) + { + if (symbol_count == 0) + printf ("Scan_procs: no symbols found!\n"); + } +#endif + + return symbol_count; +} + + +/* Traverse the quick look-up tables, building a set of psymtabs. + + This constructs a psymtab for modules and files in the quick lookup + tables. + + Mostly, modules correspond to compilation units, so we try to + create psymtabs that correspond to modules; however, in some cases + a file can result in a compiled object which does not have a module + entry for it, so in such cases we create a psymtab for the file. */ + +int +hpread_quick_traverse (struct objfile *objfile, char *gntt_bits, + char *vt_bits, PXDB_header_ptr pxdb_header_p) +{ + struct partial_symtab *pst; + + char *addr; + + quick_procedure_entry *qPD; + quick_file_entry *qFD; + quick_module_entry *qMD; + quick_class_entry *qCD; + + int idx; + int i; + CORE_ADDR start_adr; /* current psymtab's starting code addr */ + CORE_ADDR end_adr; /* current psymtab's ending code addr */ + CORE_ADDR next_mod_adr; /* next module's starting code addr */ + int curr_pd; /* current procedure */ + int curr_fd; /* current file */ + int curr_md; /* current module */ + int start_sym; /* current psymtab's starting symbol index */ + int end_sym; /* current psymtab's ending symbol index */ + int max_LNTT_sym_index; + int syms_in_pst; + B_TYPE *class_entered; + + struct partial_symbol **global_syms; /* We'll be filling in the "global" */ + struct partial_symbol **static_syms; /* and "static" tables in the objfile + as we go, so we need a pair of + current pointers. */ + +#ifdef DUMPING + /* Turn this on for lots of debugging information in this routine. + You get a blow-by-blow account of quick lookup table reading */ + static int dumping = 0; +#endif + + pst = (struct partial_symtab *) 0; + + /* Clear out some globals */ + init_pst_syms (); + told_objfile = 0; + + /* Demangling style -- if EDG style already set, don't change it, + as HP style causes some problems with the KAI EDG compiler */ + if (current_demangling_style != edg_demangling) + { + /* Otherwise, ensure that we are using HP style demangling */ + set_demangling_style (HP_DEMANGLING_STYLE_STRING); + } + + /* First we need to find the starting points of the quick + look-up tables in the GNTT. */ + + addr = gntt_bits; + + qPD = (quick_procedure_entry_ptr) addr; + addr += pxdb_header_p->pd_entries * sizeof (quick_procedure_entry); + +#ifdef DUMPING + if (dumping) + { + printf ("\n Printing routines as we see them\n"); + for (i = 0; VALID_PROC (i); i++) + { + idx = (long) qPD[i].sbProc; + printf ("%s %x..%x\n", &vt_bits[idx], + (int) PROC_START (i), + (int) PROC_END (i)); + } + } +#endif + + qFD = (quick_file_entry_ptr) addr; + addr += pxdb_header_p->fd_entries * sizeof (quick_file_entry); + +#ifdef DUMPING + if (dumping) + { + printf ("\n Printing files as we see them\n"); + for (i = 0; VALID_FILE (i); i++) + { + idx = (long) qFD[i].sbFile; + printf ("%s %x..%x\n", &vt_bits[idx], + (int) FILE_START (i), + (int) FILE_END (i)); + } + } +#endif + + qMD = (quick_module_entry_ptr) addr; + addr += pxdb_header_p->md_entries * sizeof (quick_module_entry); + +#ifdef DUMPING + if (dumping) + { + printf ("\n Printing modules as we see them\n"); + for (i = 0; i < pxdb_header_p->md_entries; i++) + { + idx = (long) qMD[i].sbMod; + printf ("%s\n", &vt_bits[idx]); + } + } +#endif + + qCD = (quick_class_entry_ptr) addr; + addr += pxdb_header_p->cd_entries * sizeof (quick_class_entry); + +#ifdef DUMPING + if (dumping) + { + printf ("\n Printing classes as we see them\n"); + for (i = 0; VALID_CLASS (i); i++) + { + idx = (long) qCD[i].sbClass; + printf ("%s\n", &vt_bits[idx]); + } + + printf ("\n Done with dump, on to build!\n"); + } +#endif + + /* We need this index only while hp-symtab-read.c expects + a byte offset to the end of the LNTT entries for a given + psymtab. Thus the need for it should go away someday. + + When it goes away, then we won't have any need to load the + LNTT from the objfile at psymtab-time, and start-up will be + faster. To make that work, we'll need some way to create + a null pst for the "globals" pseudo-module. */ + max_LNTT_sym_index = LNTT_SYMCOUNT (objfile); + + /* Scan the module descriptors and make a psymtab for each. + + We know the MDs, FDs and the PDs are in order by starting + address. We use that fact to traverse all three arrays in + parallel, knowing when the next PD is in a new file + and we need to create a new psymtab. */ + curr_pd = 0; /* Current procedure entry */ + curr_fd = 0; /* Current file entry */ + curr_md = 0; /* Current module entry */ + + start_adr = 0; /* Current psymtab code range */ + end_adr = 0; + + start_sym = 0; /* Current psymtab symbol range */ + end_sym = 0; + + syms_in_pst = 0; /* Symbol count for psymtab */ + + /* Psts actually just have pointers into the objfile's + symbol table, not their own symbol tables. */ + global_syms = objfile->global_psymbols.list; + static_syms = objfile->static_psymbols.list; + + + /* First skip over pseudo-entries with address 0. These represent inlined + routines and abstract (uninstantiated) template routines. + FIXME: These should be read in and available -- even if we can't set + breakpoints, etc., there's some information that can be presented + to the user. pai/1997-10-08 */ + + while (VALID_CURR_PROC && (CURR_PROC_START == 0)) + curr_pd++; + + /* Loop over files, modules, and procedures in code address order. Each + time we enter an iteration of this loop, curr_pd points to the first + unprocessed procedure, curr_fd points to the first unprocessed file, and + curr_md to the first unprocessed module. Each iteration of this loop + updates these as required -- any or all of them may be bumpd up + each time around. When we exit this loop, we are done with all files + and modules in the tables -- there may still be some procedures, however. + + Note: This code used to loop only over module entries, under the assumption + that files can occur via inclusions and are thus unreliable, while a + compiled object always corresponds to a module. With CTTI in the HP aCC + compiler, it turns out that compiled objects may have only files and no + modules; so we have to loop over files and modules, creating psymtabs for + either as appropriate. Unfortunately there are some problems (notably: + 1. the lack of "SRC_FILE_END" entries in the LNTT, 2. the lack of pointers + to the ending symbol indices of a module or a file) which make it quite hard + to do this correctly. Currently it uses a bunch of heuristics to start and + end psymtabs; they seem to work well with most objects generated by aCC, but + who knows when that will change... */ + + while (VALID_CURR_FILE || VALID_CURR_MODULE) + { + + char *mod_name_string; + char *full_name_string; + + /* First check for modules like "version.c", which have no code + in them but still have qMD entries. They also have no qFD or + qPD entries. Their start address is -1 and their end address + is 0. */ + if (VALID_CURR_MODULE && (CURR_MODULE_START == -1) && (CURR_MODULE_END == 0)) + { + + mod_name_string = &vt_bits[(long) qMD[curr_md].sbMod]; + +#ifdef DUMPING + if (dumping) + printf ("Module with data only %s\n", mod_name_string); +#endif + + /* We'll skip the rest (it makes error-checking easier), and + just make an empty pst. Right now empty psts are not put + in the pst chain, so all this is for naught, but later it + might help. */ + + pst = hpread_start_psymtab (objfile, + mod_name_string, + CURR_MODULE_START, /* Low text address: bogus! */ + (CURR_MODULE_ISYM * sizeof (struct dntt_type_block)), + /* ldsymoff */ + global_syms, + static_syms); + + pst = hpread_end_psymtab (pst, + NULL, /* psymtab_include_list */ + 0, /* includes_used */ + end_sym * sizeof (struct dntt_type_block), + /* byte index in LNTT of end + = capping symbol offset + = LDSYMOFF of nextfile */ + 0, /* text high */ + NULL, /* dependency_list */ + 0); /* dependencies_used */ + + global_syms = objfile->global_psymbols.next; + static_syms = objfile->static_psymbols.next; + + curr_md++; + } + else if (VALID_CURR_MODULE && + ((CURR_MODULE_START == 0) || (CURR_MODULE_START == -1) || + (CURR_MODULE_END == 0) || (CURR_MODULE_END == -1))) + { + TELL_OBJFILE; + warning ("Module \"%s\" [0x%s] has non-standard addresses. It starts at 0x%s, ends at 0x%s, and will be skipped.", + mod_name_string, paddr_nz (curr_md), paddr_nz (start_adr), paddr_nz (end_adr)); + /* On to next module */ + curr_md++; + } + else + { + /* First check if we are looking at a file with code in it + that does not overlap the current module's code range */ + + if (VALID_CURR_FILE ? (VALID_CURR_MODULE ? (CURR_FILE_END < CURR_MODULE_START) : 1) : 0) + { + + /* Looking at file not corresponding to any module, + create a psymtab for it */ + full_name_string = &vt_bits[(long) qFD[curr_fd].sbFile]; + start_adr = CURR_FILE_START; + end_adr = CURR_FILE_END; + start_sym = CURR_FILE_ISYM; + + /* Check if there are any procedures not handled until now, that + begin before the start address of this file, and if so, adjust + this module's start address to include them. This handles routines that + are in between file or module ranges for some reason (probably + indicates a compiler bug */ + + if (CURR_PROC_START < start_adr) + { + TELL_OBJFILE; + warning ("Found procedure \"%s\" [0x%x] that is not in any file or module.", + &vt_bits[(long) qPD[curr_pd].sbProc], curr_pd); + start_adr = CURR_PROC_START; + if (CURR_PROC_ISYM < start_sym) + start_sym = CURR_PROC_ISYM; + } + + /* Sometimes (compiler bug -- COBOL) the module end address is higher + than the start address of the next module, so check for that and + adjust accordingly */ + + if (VALID_FILE (curr_fd + 1) && (FILE_START (curr_fd + 1) <= end_adr)) + { + TELL_OBJFILE; + warning ("File \"%s\" [0x%x] has ending address after starting address of next file; adjusting ending address down.", + full_name_string, curr_fd); + end_adr = FILE_START (curr_fd + 1) - 1; /* Is -4 (or -8 for 64-bit) better? */ + } + if (VALID_MODULE (curr_md) && (CURR_MODULE_START <= end_adr)) + { + TELL_OBJFILE; + warning ("File \"%s\" [0x%x] has ending address after starting address of next module; adjusting ending address down.", + full_name_string, curr_fd); + end_adr = CURR_MODULE_START - 1; /* Is -4 (or -8 for 64-bit) better? */ + } + + +#ifdef DUMPING + if (dumping) + { + printf ("Make new psymtab for file %s (%x to %x).\n", + full_name_string, start_adr, end_adr); + } +#endif + /* Create the basic psymtab, connecting it in the list + for this objfile and pointing its symbol entries + to the current end of the symbol areas in the objfile. + + The "ldsymoff" parameter is the byte offset in the LNTT + of the first symbol in this file. Some day we should + turn this into an index (fix in hp-symtab-read.c as well). + And it's not even the right byte offset, as we're using + the size of a union! FIXME! */ + pst = hpread_start_psymtab (objfile, + full_name_string, + start_adr, /* Low text address */ + (start_sym * sizeof (struct dntt_type_block)), + /* ldsymoff */ + global_syms, + static_syms); + + /* Set up to only enter each class referenced in this module once. */ + class_entered = xmalloc (B_BYTES (pxdb_header_p->cd_entries)); + B_CLRALL (class_entered, pxdb_header_p->cd_entries); + + /* Scan the procedure descriptors for procedures in the current + file, based on the starting addresses. */ + + syms_in_pst = scan_procs (&curr_pd, qPD, pxdb_header_p->pd_entries, + start_adr, end_adr, pst, vt_bits, objfile); + + /* Get ending symbol offset */ + + end_sym = 0; + /* First check for starting index before previous psymtab */ + if (pst_syms_count && start_sym < pst_syms_array[pst_syms_count - 1].end) + { + end_sym = find_next_pst_start (start_sym); + } + /* Look for next start index of a file or module, or procedure */ + if (!end_sym) + { + int next_file_isym = find_next_file_isym (start_sym, qFD, curr_fd + 1, pxdb_header_p); + int next_module_isym = find_next_module_isym (start_sym, qMD, curr_md, pxdb_header_p); + int next_proc_isym = find_next_proc_isym (start_sym, qPD, curr_pd, pxdb_header_p); + + if (next_file_isym && next_module_isym) + { + /* pick lower of next file or module start index */ + end_sym = min (next_file_isym, next_module_isym); + } + else + { + /* one of them is zero, pick the other */ + end_sym = max (next_file_isym, next_module_isym); + } + + /* As a precaution, check next procedure index too */ + if (!end_sym) + end_sym = next_proc_isym; + else + end_sym = min (end_sym, next_proc_isym); + } + + /* Couldn't find procedure, file, or module, use globals as default */ + if (!end_sym) + end_sym = pxdb_header_p->globals; + +#ifdef DUMPING + if (dumping) + { + printf ("File psymtab indices: %x to %x\n", start_sym, end_sym); + } +#endif + + pst = hpread_end_psymtab (pst, + NULL, /* psymtab_include_list */ + 0, /* includes_used */ + end_sym * sizeof (struct dntt_type_block), + /* byte index in LNTT of end + = capping symbol offset + = LDSYMOFF of nextfile */ + end_adr, /* text high */ + NULL, /* dependency_list */ + 0); /* dependencies_used */ + + record_pst_syms (start_sym, end_sym); + + if (NULL == pst) + warning ("No symbols in psymtab for file \"%s\" [0x%x].", full_name_string, curr_fd); + +#ifdef DUMPING + if (dumping) + { + printf ("Made new psymtab for file %s (%x to %x), sym %x to %x.\n", + full_name_string, start_adr, end_adr, CURR_FILE_ISYM, end_sym); + } +#endif + /* Prepare for the next psymtab. */ + global_syms = objfile->global_psymbols.next; + static_syms = objfile->static_psymbols.next; + xfree (class_entered); + + curr_fd++; + } /* Psymtab for file */ + else + { + /* We have a module for which we create a psymtab */ + + mod_name_string = &vt_bits[(long) qMD[curr_md].sbMod]; + + /* We will include the code ranges of any files that happen to + overlap with this module */ + + /* So, first pick the lower of the file's and module's start addresses */ + start_adr = CURR_MODULE_START; + if (VALID_CURR_FILE) + { + if (CURR_FILE_START < CURR_MODULE_START) + { + TELL_OBJFILE; + warning ("File \"%s\" [0x%x] crosses beginning of module \"%s\".", + &vt_bits[(long) qFD[curr_fd].sbFile], + curr_fd, mod_name_string); + + start_adr = CURR_FILE_START; + } + } + + /* Also pick the lower of the file's and the module's start symbol indices */ + start_sym = CURR_MODULE_ISYM; + if (VALID_CURR_FILE && (CURR_FILE_ISYM < CURR_MODULE_ISYM)) + start_sym = CURR_FILE_ISYM; + + /* For the end address, we scan through the files till we find one + that overlaps the current module but ends beyond it; if no such file exists we + simply use the module's start address. + (Note, if file entries themselves overlap + we take the longest overlapping extension beyond the end of the module...) + We assume that modules never overlap. */ + + end_adr = CURR_MODULE_END; + + if (VALID_CURR_FILE) + { + while (VALID_CURR_FILE && (CURR_FILE_START < end_adr)) + { + +#ifdef DUMPING + if (dumping) + printf ("Maybe skipping file %s which overlaps with module %s\n", + &vt_bits[(long) qFD[curr_fd].sbFile], mod_name_string); +#endif + if (CURR_FILE_END > end_adr) + { + TELL_OBJFILE; + warning ("File \"%s\" [0x%x] crosses end of module \"%s\".", + &vt_bits[(long) qFD[curr_fd].sbFile], + curr_fd, mod_name_string); + end_adr = CURR_FILE_END; + } + curr_fd++; + } + curr_fd--; /* back up after going too far */ + } + + /* Sometimes (compiler bug -- COBOL) the module end address is higher + than the start address of the next module, so check for that and + adjust accordingly */ + + if (VALID_MODULE (curr_md + 1) && (MODULE_START (curr_md + 1) <= end_adr)) + { + TELL_OBJFILE; + warning ("Module \"%s\" [0x%x] has ending address after starting address of next module; adjusting ending address down.", + mod_name_string, curr_md); + end_adr = MODULE_START (curr_md + 1) - 1; /* Is -4 (or -8 for 64-bit) better? */ + } + if (VALID_FILE (curr_fd + 1) && (FILE_START (curr_fd + 1) <= end_adr)) + { + TELL_OBJFILE; + warning ("Module \"%s\" [0x%x] has ending address after starting address of next file; adjusting ending address down.", + mod_name_string, curr_md); + end_adr = FILE_START (curr_fd + 1) - 1; /* Is -4 (or -8 for 64-bit) better? */ + } + + /* Use one file to get the full name for the module. This + situation can arise if there is executable code in a #include + file. Each file with code in it gets a qFD. Files which don't + contribute code don't get a qFD, even if they include files + which do, e.g.: + + body.c: rtn.h: + int x; int main() { + #include "rtn.h" return x; + } + + There will a qFD for "rtn.h",and a qMD for "body.c", + but no qMD for "rtn.h" or qFD for "body.c"! + + We pick the name of the last file to overlap with this + module. C convention is to put include files first. In a + perfect world, we could check names and use the file whose full + path name ends with the module name. */ + + if (VALID_CURR_FILE) + full_name_string = &vt_bits[(long) qFD[curr_fd].sbFile]; + else + full_name_string = mod_name_string; + + /* Check if there are any procedures not handled until now, that + begin before the start address we have now, and if so, adjust + this psymtab's start address to include them. This handles routines that + are in between file or module ranges for some reason (probably + indicates a compiler bug */ + + if (CURR_PROC_START < start_adr) + { + TELL_OBJFILE; + warning ("Found procedure \"%s\" [0x%x] that is not in any file or module.", + &vt_bits[(long) qPD[curr_pd].sbProc], curr_pd); + start_adr = CURR_PROC_START; + if (CURR_PROC_ISYM < start_sym) + start_sym = CURR_PROC_ISYM; + } + +#ifdef DUMPING + if (dumping) + { + printf ("Make new psymtab for module %s (%x to %x), using file %s\n", + mod_name_string, start_adr, end_adr, full_name_string); + } +#endif + /* Create the basic psymtab, connecting it in the list + for this objfile and pointing its symbol entries + to the current end of the symbol areas in the objfile. + + The "ldsymoff" parameter is the byte offset in the LNTT + of the first symbol in this file. Some day we should + turn this into an index (fix in hp-symtab-read.c as well). + And it's not even the right byte offset, as we're using + the size of a union! FIXME! */ + pst = hpread_start_psymtab (objfile, + full_name_string, + start_adr, /* Low text address */ + (start_sym * sizeof (struct dntt_type_block)), + /* ldsymoff */ + global_syms, + static_syms); + + /* Set up to only enter each class referenced in this module once. */ + class_entered = xmalloc (B_BYTES (pxdb_header_p->cd_entries)); + B_CLRALL (class_entered, pxdb_header_p->cd_entries); + + /* Scan the procedure descriptors for procedures in the current + module, based on the starting addresses. */ + + syms_in_pst = scan_procs (&curr_pd, qPD, pxdb_header_p->pd_entries, + start_adr, end_adr, pst, vt_bits, objfile); + + /* Get ending symbol offset */ + + end_sym = 0; + /* First check for starting index before previous psymtab */ + if (pst_syms_count && start_sym < pst_syms_array[pst_syms_count - 1].end) + { + end_sym = find_next_pst_start (start_sym); + } + /* Look for next start index of a file or module, or procedure */ + if (!end_sym) + { + int next_file_isym = find_next_file_isym (start_sym, qFD, curr_fd + 1, pxdb_header_p); + int next_module_isym = find_next_module_isym (start_sym, qMD, curr_md + 1, pxdb_header_p); + int next_proc_isym = find_next_proc_isym (start_sym, qPD, curr_pd, pxdb_header_p); + + if (next_file_isym && next_module_isym) + { + /* pick lower of next file or module start index */ + end_sym = min (next_file_isym, next_module_isym); + } + else + { + /* one of them is zero, pick the other */ + end_sym = max (next_file_isym, next_module_isym); + } + + /* As a precaution, check next procedure index too */ + if (!end_sym) + end_sym = next_proc_isym; + else + end_sym = min (end_sym, next_proc_isym); + } + + /* Couldn't find procedure, file, or module, use globals as default */ + if (!end_sym) + end_sym = pxdb_header_p->globals; + +#ifdef DUMPING + if (dumping) + { + printf ("Module psymtab indices: %x to %x\n", start_sym, end_sym); + } +#endif + + pst = hpread_end_psymtab (pst, + NULL, /* psymtab_include_list */ + 0, /* includes_used */ + end_sym * sizeof (struct dntt_type_block), + /* byte index in LNTT of end + = capping symbol offset + = LDSYMOFF of nextfile */ + end_adr, /* text high */ + NULL, /* dependency_list */ + 0); /* dependencies_used */ + + record_pst_syms (start_sym, end_sym); + + if (NULL == pst) + warning ("No symbols in psymtab for module \"%s\" [0x%x].", mod_name_string, curr_md); + +#ifdef DUMPING + if (dumping) + { + printf ("Made new psymtab for module %s (%x to %x), sym %x to %x.\n", + mod_name_string, start_adr, end_adr, CURR_MODULE_ISYM, end_sym); + } +#endif + + /* Prepare for the next psymtab. */ + global_syms = objfile->global_psymbols.next; + static_syms = objfile->static_psymbols.next; + xfree (class_entered); + + curr_md++; + curr_fd++; + } /* psymtab for module */ + } /* psymtab for non-bogus file or module */ + } /* End of while loop over all files & modules */ + + /* There may be some routines after all files and modules -- these will get + inserted in a separate new module of their own */ + if (VALID_CURR_PROC) + { + start_adr = CURR_PROC_START; + end_adr = qPD[pxdb_header_p->pd_entries - 1].adrEnd; + TELL_OBJFILE; + warning ("Found functions beyond end of all files and modules [0x%x].", curr_pd); +#ifdef DUMPING + if (dumping) + { + printf ("Orphan functions at end, PD %d and beyond (%x to %x)\n", + curr_pd, start_adr, end_adr); + } +#endif + pst = hpread_start_psymtab (objfile, + "orphans", + start_adr, /* Low text address */ + (CURR_PROC_ISYM * sizeof (struct dntt_type_block)), + /* ldsymoff */ + global_syms, + static_syms); + + scan_procs (&curr_pd, qPD, pxdb_header_p->pd_entries, + start_adr, end_adr, pst, vt_bits, objfile); + + pst = hpread_end_psymtab (pst, + NULL, /* psymtab_include_list */ + 0, /* includes_used */ + pxdb_header_p->globals * sizeof (struct dntt_type_block), + /* byte index in LNTT of end + = capping symbol offset + = LDSYMOFF of nextfile */ + end_adr, /* text high */ + NULL, /* dependency_list */ + 0); /* dependencies_used */ + } + + +#ifdef NEVER_NEVER + /* Now build psts for non-module things (in the tail of + the LNTT, after the last END MODULE entry). + + If null psts were kept on the chain, this would be + a solution. FIXME */ + pst = hpread_start_psymtab (objfile, + "globals", + 0, + (pxdb_header_p->globals + * sizeof (struct dntt_type_block)), + objfile->global_psymbols.next, + objfile->static_psymbols.next); + hpread_end_psymtab (pst, + NULL, 0, + (max_LNTT_sym_index * sizeof (struct dntt_type_block)), + 0, + NULL, 0); +#endif + + clear_pst_syms (); + + return 1; + +} /* End of hpread_quick_traverse. */ + + +/* Get appropriate header, based on pxdb type. + Return value: 1 if ok, 0 if not */ +int +hpread_get_header (struct objfile *objfile, PXDB_header_ptr pxdb_header_p) +{ + asection *pinfo_section, *debug_section, *header_section; + +#ifdef DUMPING + /* Turn on for debugging information */ + static int dumping = 0; +#endif + + header_section = bfd_get_section_by_name (objfile->obfd, "$HEADER$"); + if (!header_section) + { + /* We don't have either PINFO or DEBUG sections. But + stuff like "libc.sl" has no debug info. There's no + need to warn the user of this, as it may be ok. The + caller will figure it out and issue any needed + messages. */ +#ifdef DUMPING + if (dumping) + printf ("==No debug info at all for %s.\n", objfile->name); +#endif + + return 0; + } + + /* We would like either a $DEBUG$ or $PINFO$ section. + Once we know which, we can understand the header + data (which we have defined to suit the more common + $DEBUG$ case). */ + debug_section = bfd_get_section_by_name (objfile->obfd, "$DEBUG$"); + pinfo_section = bfd_get_section_by_name (objfile->obfd, "$PINFO$"); + if (debug_section) + { + /* The expected case: normal pxdb header. */ + bfd_get_section_contents (objfile->obfd, header_section, + pxdb_header_p, 0, sizeof (PXDB_header)); + + if (!pxdb_header_p->pxdbed) + { + /* This shouldn't happen if we check in "symfile.c". */ + return 0; + } /* DEBUG section */ + } + + else if (pinfo_section) + { + /* The DOC case; we need to translate this into a + regular header. */ + DOC_info_PXDB_header doc_header; + +#ifdef DUMPING + if (dumping) + { + printf ("==OOps, PINFO, let's try to handle this, %s.\n", objfile->name); + } +#endif + + bfd_get_section_contents (objfile->obfd, + header_section, + &doc_header, 0, + sizeof (DOC_info_PXDB_header)); + + if (!doc_header.pxdbed) + { + /* This shouldn't happen if we check in "symfile.c". */ + warning ("File \"%s\" not processed by pxdb!", objfile->name); + return 0; + } + + /* Copy relevent fields to standard header passed in. */ + pxdb_header_p->pd_entries = doc_header.pd_entries; + pxdb_header_p->fd_entries = doc_header.fd_entries; + pxdb_header_p->md_entries = doc_header.md_entries; + pxdb_header_p->pxdbed = doc_header.pxdbed; + pxdb_header_p->bighdr = doc_header.bighdr; + pxdb_header_p->sa_header = doc_header.sa_header; + pxdb_header_p->inlined = doc_header.inlined; + pxdb_header_p->globals = doc_header.globals; + pxdb_header_p->time = doc_header.time; + pxdb_header_p->pg_entries = doc_header.pg_entries; + pxdb_header_p->functions = doc_header.functions; + pxdb_header_p->files = doc_header.files; + pxdb_header_p->cd_entries = doc_header.cd_entries; + pxdb_header_p->aa_entries = doc_header.aa_entries; + pxdb_header_p->oi_entries = doc_header.oi_entries; + pxdb_header_p->version = doc_header.version; + } /* PINFO section */ + + else + { +#ifdef DUMPING + if (dumping) + printf ("==No debug info at all for %s.\n", objfile->name); +#endif + + return 0; + + } + + return 1; +} /* End of hpread_get_header */ +#endif /* QUICK_LOOK_UP */ + + +/* Initialization for reading native HP C debug symbols from OBJFILE. + + Its only purpose in life is to set up the symbol reader's private + per-objfile data structures, and read in the raw contents of the debug + sections (attaching pointers to the debug info into the private data + structures). + + Since BFD doesn't know how to read debug symbols in a format-independent + way (and may never do so...), we have to do it ourselves. Note we may + be called on a file without native HP C debugging symbols. + + FIXME, there should be a cleaner peephole into the BFD environment + here. */ +void +hpread_symfile_init (struct objfile *objfile) +{ + asection *vt_section, *slt_section, *lntt_section, *gntt_section; + + /* Allocate struct to keep track of the symfile */ + objfile->sym_private = (PTR) + xmmalloc (objfile->md, sizeof (struct hpread_symfile_info)); + memset (objfile->sym_private, 0, sizeof (struct hpread_symfile_info)); + + /* We haven't read in any types yet. */ + DNTT_TYPE_VECTOR (objfile) = 0; + + /* Read in data from the $GNTT$ subspace. */ + gntt_section = bfd_get_section_by_name (objfile->obfd, "$GNTT$"); + if (!gntt_section) + return; + + GNTT (objfile) + = obstack_alloc (&objfile->symbol_obstack, + bfd_section_size (objfile->obfd, gntt_section)); + + bfd_get_section_contents (objfile->obfd, gntt_section, GNTT (objfile), + 0, bfd_section_size (objfile->obfd, gntt_section)); + + GNTT_SYMCOUNT (objfile) + = bfd_section_size (objfile->obfd, gntt_section) + / sizeof (struct dntt_type_block); + + /* Read in data from the $LNTT$ subspace. Also keep track of the number + of LNTT symbols. + + FIXME: this could be moved into the psymtab-to-symtab expansion + code, and save startup time. At the moment this data is + still used, though. We'd need a way to tell hp-symtab-read.c + whether or not to load the LNTT. */ + lntt_section = bfd_get_section_by_name (objfile->obfd, "$LNTT$"); + if (!lntt_section) + return; + + LNTT (objfile) + = obstack_alloc (&objfile->symbol_obstack, + bfd_section_size (objfile->obfd, lntt_section)); + + bfd_get_section_contents (objfile->obfd, lntt_section, LNTT (objfile), + 0, bfd_section_size (objfile->obfd, lntt_section)); + + LNTT_SYMCOUNT (objfile) + = bfd_section_size (objfile->obfd, lntt_section) + / sizeof (struct dntt_type_block); + + /* Read in data from the $SLT$ subspace. $SLT$ contains information + on source line numbers. */ + slt_section = bfd_get_section_by_name (objfile->obfd, "$SLT$"); + if (!slt_section) + return; + + SLT (objfile) = + obstack_alloc (&objfile->symbol_obstack, + bfd_section_size (objfile->obfd, slt_section)); + + bfd_get_section_contents (objfile->obfd, slt_section, SLT (objfile), + 0, bfd_section_size (objfile->obfd, slt_section)); + + /* Read in data from the $VT$ subspace. $VT$ contains things like + names and constants. Keep track of the number of symbols in the VT. */ + vt_section = bfd_get_section_by_name (objfile->obfd, "$VT$"); + if (!vt_section) + return; + + VT_SIZE (objfile) = bfd_section_size (objfile->obfd, vt_section); + + VT (objfile) = + (char *) obstack_alloc (&objfile->symbol_obstack, + VT_SIZE (objfile)); + + bfd_get_section_contents (objfile->obfd, vt_section, VT (objfile), + 0, VT_SIZE (objfile)); +} + +/* Scan and build partial symbols for a symbol file. + + The minimal symbol table (either SOM or HP a.out) has already been + read in; all we need to do is setup partial symbols based on the + native debugging information. + + Note that the minimal table is produced by the linker, and has + only global routines in it; the psymtab is based on compiler- + generated debug information and has non-global + routines in it as well as files and class information. + + We assume hpread_symfile_init has been called to initialize the + symbol reader's private data structures. + + MAINLINE is true if we are reading the main symbol table (as + opposed to a shared lib or dynamically loaded file). */ + +void +hpread_build_psymtabs (struct objfile *objfile, int mainline) +{ + +#ifdef DUMPING + /* Turn this on to get debugging output. */ + static int dumping = 0; +#endif + + char *namestring; + int past_first_source_file = 0; + struct cleanup *old_chain; + + int hp_symnum, symcount, i; + int scan_start = 0; + + union dnttentry *dn_bufp; + unsigned long valu; + char *p; + int texthigh = 0; + int have_name = 0; + + /* Current partial symtab */ + struct partial_symtab *pst; + + /* List of current psymtab's include files */ + char **psymtab_include_list; + int includes_allocated; + int includes_used; + + /* Index within current psymtab dependency list */ + struct partial_symtab **dependency_list; + int dependencies_used, dependencies_allocated; + + /* Just in case the stabs reader left turds lying around. */ + free_pending_blocks (); + make_cleanup (really_free_pendings, 0); + + pst = (struct partial_symtab *) 0; + + /* We shouldn't use alloca, instead use malloc/free. Doing so avoids + a number of problems with cross compilation and creating useless holes + in the stack when we have to allocate new entries. FIXME. */ + + includes_allocated = 30; + includes_used = 0; + psymtab_include_list = (char **) alloca (includes_allocated * + sizeof (char *)); + + dependencies_allocated = 30; + dependencies_used = 0; + dependency_list = + (struct partial_symtab **) alloca (dependencies_allocated * + sizeof (struct partial_symtab *)); + + old_chain = make_cleanup_free_objfile (objfile); + + last_source_file = 0; + +#ifdef QUICK_LOOK_UP + { + /* Begin code for new-style loading of quick look-up tables. */ + + /* elz: this checks whether the file has beeen processed by pxdb. + If not we would like to try to read the psymbols in + anyway, but it turns out to be not so easy. So this could + actually be commented out, but I leave it in, just in case + we decide to add support for non-pxdb-ed stuff in the future. */ + PXDB_header pxdb_header; + int found_modules_in_program; + + if (hpread_get_header (objfile, &pxdb_header)) + { + /* Build a minimal table. No types, no global variables, + no include files.... */ +#ifdef DUMPING + if (dumping) + printf ("\nNew method for %s\n", objfile->name); +#endif + + /* elz: quick_traverse returns true if it found + some modules in the main source file, other + than those in end.c + In C and C++, all the files have MODULES entries + in the LNTT, and the quick table traverse is all + based on finding these MODULES entries. Without + those it cannot work. + It happens that F77 programs don't have MODULES + so the quick traverse gets confused. F90 programs + have modules, and the quick method still works. + So, if modules (other than those in end.c) are + not found we give up on the quick table stuff, + and fall back on the slower method */ + found_modules_in_program = hpread_quick_traverse (objfile, + GNTT (objfile), + VT (objfile), + &pxdb_header); + + discard_cleanups (old_chain); + + /* Set up to scan the global section of the LNTT. + + This field is not always correct: if there are + no globals, it will point to the last record in + the regular LNTT, which is usually an END MODULE. + + Since it might happen that there could be a file + with just one global record, there's no way to + tell other than by looking at the record, so that's + done below. */ + if (found_modules_in_program) + scan_start = pxdb_header.globals; + } +#ifdef DUMPING + else + { + if (dumping) + printf ("\nGoing on to old method for %s\n", objfile->name); + } +#endif + } +#endif /* QUICK_LOOK_UP */ + + /* Make two passes, one over the GNTT symbols, the other for the + LNTT symbols. + + JB comment: above isn't true--they only make one pass, over + the LNTT. */ + for (i = 0; i < 1; i++) + { + int within_function = 0; + + if (i) + symcount = GNTT_SYMCOUNT (objfile); + else + symcount = LNTT_SYMCOUNT (objfile); + + + for (hp_symnum = scan_start; hp_symnum < symcount; hp_symnum++) + { + QUIT; + if (i) + dn_bufp = hpread_get_gntt (hp_symnum, objfile); + else + dn_bufp = hpread_get_lntt (hp_symnum, objfile); + + if (dn_bufp->dblock.extension) + continue; + + /* Only handle things which are necessary for minimal symbols. + everything else is ignored. */ + switch (dn_bufp->dblock.kind) + { + case DNTT_TYPE_SRCFILE: + { +#ifdef QUICK_LOOK_UP + if (scan_start == hp_symnum + && symcount == hp_symnum + 1) + { + /* If there are NO globals in an executable, + PXDB's index to the globals will point to + the last record in the file, which + could be this record. (this happened for F77 libraries) + ignore it and be done! */ + continue; + } +#endif /* QUICK_LOOK_UP */ + + /* A source file of some kind. Note this may simply + be an included file. */ + SET_NAMESTRING (dn_bufp, &namestring, objfile); + + /* Check if this is the source file we are already working + with. */ + if (pst && !strcmp (namestring, pst->filename)) + continue; + + /* Check if this is an include file, if so check if we have + already seen it. Add it to the include list */ + p = strrchr (namestring, '.'); + if (!strcmp (p, ".h")) + { + int j, found; + + found = 0; + for (j = 0; j < includes_used; j++) + if (!strcmp (namestring, psymtab_include_list[j])) + { + found = 1; + break; + } + if (found) + continue; + + /* Add it to the list of includes seen so far and + allocate more include space if necessary. */ + psymtab_include_list[includes_used++] = namestring; + if (includes_used >= includes_allocated) + { + char **orig = psymtab_include_list; + + psymtab_include_list = (char **) + alloca ((includes_allocated *= 2) * + sizeof (char *)); + memcpy ((PTR) psymtab_include_list, (PTR) orig, + includes_used * sizeof (char *)); + } + continue; + } + + if (pst) + { + if (!have_name) + { + pst->filename = (char *) + obstack_alloc (&pst->objfile->psymbol_obstack, + strlen (namestring) + 1); + strcpy (pst->filename, namestring); + have_name = 1; + continue; + } + continue; + } + + /* This is a bonafide new source file. + End the current partial symtab and start a new one. */ + + if (pst && past_first_source_file) + { + hpread_end_psymtab (pst, psymtab_include_list, + includes_used, + (hp_symnum + * sizeof (struct dntt_type_block)), + texthigh, + dependency_list, dependencies_used); + pst = (struct partial_symtab *) 0; + includes_used = 0; + dependencies_used = 0; + } + else + past_first_source_file = 1; + + valu = hpread_get_textlow (i, hp_symnum, objfile, symcount); + valu += ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile)); + pst = hpread_start_psymtab (objfile, + namestring, valu, + (hp_symnum + * sizeof (struct dntt_type_block)), + objfile->global_psymbols.next, + objfile->static_psymbols.next); + texthigh = valu; + have_name = 1; + continue; + } + + case DNTT_TYPE_MODULE: + /* A source file. It's still unclear to me what the + real difference between a DNTT_TYPE_SRCFILE and DNTT_TYPE_MODULE + is supposed to be. */ + + /* First end the previous psymtab */ + if (pst) + { + hpread_end_psymtab (pst, psymtab_include_list, includes_used, + ((hp_symnum - 1) + * sizeof (struct dntt_type_block)), + texthigh, + dependency_list, dependencies_used); + pst = (struct partial_symtab *) 0; + includes_used = 0; + dependencies_used = 0; + have_name = 0; + } + + /* Now begin a new module and a new psymtab for it */ + SET_NAMESTRING (dn_bufp, &namestring, objfile); + valu = hpread_get_textlow (i, hp_symnum, objfile, symcount); + valu += ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile)); + if (!pst) + { + pst = hpread_start_psymtab (objfile, + namestring, valu, + (hp_symnum + * sizeof (struct dntt_type_block)), + objfile->global_psymbols.next, + objfile->static_psymbols.next); + texthigh = valu; + have_name = 0; + } + continue; + + case DNTT_TYPE_FUNCTION: + case DNTT_TYPE_ENTRY: + /* The beginning of a function. DNTT_TYPE_ENTRY may also denote + a secondary entry point. */ + valu = dn_bufp->dfunc.hiaddr + ANOFFSET (objfile->section_offsets, + SECT_OFF_TEXT (objfile)); + if (valu > texthigh) + texthigh = valu; + valu = dn_bufp->dfunc.lowaddr + + ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile)); + SET_NAMESTRING (dn_bufp, &namestring, objfile); + if (dn_bufp->dfunc.global) + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_BLOCK, + &objfile->global_psymbols, valu, + 0, language_unknown, objfile); + else + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_BLOCK, + &objfile->static_psymbols, valu, + 0, language_unknown, objfile); + within_function = 1; + continue; + + case DNTT_TYPE_DOC_FUNCTION: + valu = dn_bufp->ddocfunc.hiaddr + ANOFFSET (objfile->section_offsets, + SECT_OFF_TEXT (objfile)); + if (valu > texthigh) + texthigh = valu; + valu = dn_bufp->ddocfunc.lowaddr + + ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile)); + SET_NAMESTRING (dn_bufp, &namestring, objfile); + if (dn_bufp->ddocfunc.global) + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_BLOCK, + &objfile->global_psymbols, valu, + 0, language_unknown, objfile); + else + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_BLOCK, + &objfile->static_psymbols, valu, + 0, language_unknown, objfile); + within_function = 1; + continue; + + case DNTT_TYPE_BEGIN: + case DNTT_TYPE_END: + /* We don't check MODULE end here, because there can be + symbols beyond the module end which properly belong to the + current psymtab -- so we wait till the next MODULE start */ + + +#ifdef QUICK_LOOK_UP + if (scan_start == hp_symnum + && symcount == hp_symnum + 1) + { + /* If there are NO globals in an executable, + PXDB's index to the globals will point to + the last record in the file, which is + probably an END MODULE, i.e. this record. + ignore it and be done! */ + continue; + } +#endif /* QUICK_LOOK_UP */ + + /* Scope block begin/end. We only care about function + and file blocks right now. */ + + if ((dn_bufp->dend.endkind == DNTT_TYPE_FUNCTION) || + (dn_bufp->dend.endkind == DNTT_TYPE_DOC_FUNCTION)) + within_function = 0; + continue; + + case DNTT_TYPE_SVAR: + case DNTT_TYPE_DVAR: + case DNTT_TYPE_TYPEDEF: + case DNTT_TYPE_TAGDEF: + { + /* Variables, typedefs an the like. */ + enum address_class storage; + namespace_enum namespace; + + /* Don't add locals to the partial symbol table. */ + if (within_function + && (dn_bufp->dblock.kind == DNTT_TYPE_SVAR + || dn_bufp->dblock.kind == DNTT_TYPE_DVAR)) + continue; + + /* TAGDEFs go into the structure namespace. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_TAGDEF) + namespace = STRUCT_NAMESPACE; + else + namespace = VAR_NAMESPACE; + + /* What kind of "storage" does this use? */ + if (dn_bufp->dblock.kind == DNTT_TYPE_SVAR) + storage = LOC_STATIC; + else if (dn_bufp->dblock.kind == DNTT_TYPE_DVAR + && dn_bufp->ddvar.regvar) + storage = LOC_REGISTER; + else if (dn_bufp->dblock.kind == DNTT_TYPE_DVAR) + storage = LOC_LOCAL; + else + storage = LOC_UNDEF; + + SET_NAMESTRING (dn_bufp, &namestring, objfile); + if (!pst) + { + pst = hpread_start_psymtab (objfile, + "globals", 0, + (hp_symnum + * sizeof (struct dntt_type_block)), + objfile->global_psymbols.next, + objfile->static_psymbols.next); + } + + /* Compute address of the data symbol */ + valu = dn_bufp->dsvar.location; + /* Relocate in case it's in a shared library */ + if (storage == LOC_STATIC) + valu += ANOFFSET (objfile->section_offsets, SECT_OFF_DATA (objfile)); + + /* Luckily, dvar, svar, typedef, and tagdef all + have their "global" bit in the same place, so it works + (though it's bad programming practice) to reference + "dsvar.global" even though we may be looking at + any of the above four types. */ + if (dn_bufp->dsvar.global) + { + add_psymbol_to_list (namestring, strlen (namestring), + namespace, storage, + &objfile->global_psymbols, + valu, + 0, language_unknown, objfile); + } + else + { + add_psymbol_to_list (namestring, strlen (namestring), + namespace, storage, + &objfile->static_psymbols, + valu, + 0, language_unknown, objfile); + } + + /* For TAGDEF's, the above code added the tagname to the + struct namespace. This will cause tag "t" to be found + on a reference of the form "(struct t) x". But for + C++ classes, "t" will also be a typename, which we + want to find on a reference of the form "ptype t". + Therefore, we also add "t" to the var namespace. + Do the same for enum's due to the way aCC generates + debug info for these (see more extended comment + in hp-symtab-read.c). + We do the same for templates, so that "ptype t" + where "t" is a template also works. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_TAGDEF && + dn_bufp->dtype.type.dnttp.index < LNTT_SYMCOUNT (objfile)) + { + int global = dn_bufp->dtag.global; + /* Look ahead to see if it's a C++ class */ + dn_bufp = hpread_get_lntt (dn_bufp->dtype.type.dnttp.index, objfile); + if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS || + dn_bufp->dblock.kind == DNTT_TYPE_ENUM || + dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + if (global) + { + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, storage, + &objfile->global_psymbols, + dn_bufp->dsvar.location, + 0, language_unknown, objfile); + } + else + { + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, storage, + &objfile->static_psymbols, + dn_bufp->dsvar.location, + 0, language_unknown, objfile); + } + } + } + } + continue; + + case DNTT_TYPE_MEMENUM: + case DNTT_TYPE_CONST: + /* Constants and members of enumerated types. */ + SET_NAMESTRING (dn_bufp, &namestring, objfile); + if (!pst) + { + pst = hpread_start_psymtab (objfile, + "globals", 0, + (hp_symnum + * sizeof (struct dntt_type_block)), + objfile->global_psymbols.next, + objfile->static_psymbols.next); + } + if (dn_bufp->dconst.global) + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_CONST, + &objfile->global_psymbols, 0, + 0, language_unknown, objfile); + else + add_psymbol_to_list (namestring, strlen (namestring), + VAR_NAMESPACE, LOC_CONST, + &objfile->static_psymbols, 0, + 0, language_unknown, objfile); + continue; + default: + continue; + } + } + } + + /* End any pending partial symbol table. */ + if (pst) + { + hpread_end_psymtab (pst, psymtab_include_list, includes_used, + hp_symnum * sizeof (struct dntt_type_block), + 0, dependency_list, dependencies_used); + } + + discard_cleanups (old_chain); +} + +/* Perform any local cleanups required when we are done with a particular + objfile. I.E, we are in the process of discarding all symbol information + for an objfile, freeing up all memory held for it, and unlinking the + objfile struct from the global list of known objfiles. */ + +void +hpread_symfile_finish (struct objfile *objfile) +{ + if (objfile->sym_private != NULL) + { + xmfree (objfile->md, objfile->sym_private); + } +} + + +/* The remaining functions are all for internal use only. */ + +/* Various small functions to get entries in the debug symbol sections. */ + +union dnttentry * +hpread_get_lntt (int index, struct objfile *objfile) +{ + return (union dnttentry *) + &(LNTT (objfile)[(index * sizeof (struct dntt_type_block))]); +} + +static union dnttentry * +hpread_get_gntt (int index, struct objfile *objfile) +{ + return (union dnttentry *) + &(GNTT (objfile)[(index * sizeof (struct dntt_type_block))]); +} + +union sltentry * +hpread_get_slt (int index, struct objfile *objfile) +{ + return (union sltentry *) &(SLT (objfile)[index * sizeof (union sltentry)]); +} + +/* Get the low address associated with some symbol (typically the start + of a particular source file or module). Since that information is not + stored as part of the DNTT_TYPE_MODULE or DNTT_TYPE_SRCFILE symbol we + must infer it from the existence of DNTT_TYPE_FUNCTION symbols. */ + +static unsigned long +hpread_get_textlow (int global, int index, struct objfile *objfile, + int symcount) +{ + union dnttentry *dn_bufp; + struct minimal_symbol *msymbol; + + /* Look for a DNTT_TYPE_FUNCTION symbol. */ + if (index < symcount) /* symcount is the number of symbols in */ + { /* the dbinfo, LNTT table */ + do + { + if (global) + dn_bufp = hpread_get_gntt (index++, objfile); + else + dn_bufp = hpread_get_lntt (index++, objfile); + } + while (dn_bufp->dblock.kind != DNTT_TYPE_FUNCTION + && dn_bufp->dblock.kind != DNTT_TYPE_DOC_FUNCTION + && dn_bufp->dblock.kind != DNTT_TYPE_END + && index < symcount); + } + + /* Avoid going past a DNTT_TYPE_END when looking for a DNTT_TYPE_FUNCTION. This + might happen when a sourcefile has no functions. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_END) + return 0; + + /* Avoid going past the end of the LNTT file */ + if (index == symcount) + return 0; + + /* The minimal symbols are typically more accurate for some reason. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_FUNCTION) + msymbol = lookup_minimal_symbol (dn_bufp->dfunc.name + VT (objfile), NULL, + objfile); + else /* must be a DNTT_TYPE_DOC_FUNCTION */ + msymbol = lookup_minimal_symbol (dn_bufp->ddocfunc.name + VT (objfile), NULL, + objfile); + + if (msymbol) + return SYMBOL_VALUE_ADDRESS (msymbol); + else + return dn_bufp->dfunc.lowaddr; +} + +/* Allocate and partially fill a partial symtab. It will be + completely filled at the end of the symbol list. + + SYMFILE_NAME is the name of the symbol-file we are reading from, and ADDR + is the address relative to which its symbols are (incremental) or 0 + (normal). */ + +static struct partial_symtab * +hpread_start_psymtab (struct objfile *objfile, char *filename, + CORE_ADDR textlow, int ldsymoff, + struct partial_symbol **global_syms, + struct partial_symbol **static_syms) +{ + int offset = ANOFFSET (objfile->section_offsets, SECT_OFF_TEXT (objfile)); + extern void hpread_psymtab_to_symtab (); + struct partial_symtab *result = + start_psymtab_common (objfile, objfile->section_offsets, + filename, textlow, global_syms, static_syms); + + result->textlow += offset; + result->read_symtab_private = (char *) + obstack_alloc (&objfile->psymbol_obstack, sizeof (struct symloc)); + LDSYMOFF (result) = ldsymoff; + result->read_symtab = hpread_psymtab_to_symtab; + + return result; +} + + +/* Close off the current usage of PST. + Returns PST or NULL if the partial symtab was empty and thrown away. + + capping_symbol_offset --Byte index in LNTT or GNTT of the + last symbol processed during the build + of the previous pst. + + FIXME: List variables and peculiarities of same. */ + +static struct partial_symtab * +hpread_end_psymtab (struct partial_symtab *pst, char **include_list, + int num_includes, int capping_symbol_offset, + CORE_ADDR capping_text, + struct partial_symtab **dependency_list, + int number_dependencies) +{ + int i; + struct objfile *objfile = pst->objfile; + int offset = ANOFFSET (pst->section_offsets, SECT_OFF_TEXT (objfile)); + +#ifdef DUMPING + /* Turn on to see what kind of a psymtab we've built. */ + static int dumping = 0; +#endif + + if (capping_symbol_offset != -1) + LDSYMLEN (pst) = capping_symbol_offset - LDSYMOFF (pst); + else + LDSYMLEN (pst) = 0; + pst->texthigh = capping_text + offset; + + pst->n_global_syms = + objfile->global_psymbols.next - (objfile->global_psymbols.list + pst->globals_offset); + pst->n_static_syms = + objfile->static_psymbols.next - (objfile->static_psymbols.list + pst->statics_offset); + +#ifdef DUMPING + if (dumping) + { + printf ("\nPst %s, LDSYMOFF %x (%x), LDSYMLEN %x (%x), globals %d, statics %d\n", + pst->filename, + LDSYMOFF (pst), + LDSYMOFF (pst) / sizeof (struct dntt_type_block), + LDSYMLEN (pst), + LDSYMLEN (pst) / sizeof (struct dntt_type_block), + pst->n_global_syms, pst->n_static_syms); + } +#endif + + pst->number_of_dependencies = number_dependencies; + if (number_dependencies) + { + pst->dependencies = (struct partial_symtab **) + obstack_alloc (&objfile->psymbol_obstack, + number_dependencies * sizeof (struct partial_symtab *)); + memcpy (pst->dependencies, dependency_list, + number_dependencies * sizeof (struct partial_symtab *)); + } + else + pst->dependencies = 0; + + for (i = 0; i < num_includes; i++) + { + struct partial_symtab *subpst = + allocate_psymtab (include_list[i], objfile); + + subpst->section_offsets = pst->section_offsets; + subpst->read_symtab_private = + (char *) obstack_alloc (&objfile->psymbol_obstack, + sizeof (struct symloc)); + LDSYMOFF (subpst) = + LDSYMLEN (subpst) = + subpst->textlow = + subpst->texthigh = 0; + + /* We could save slight bits of space by only making one of these, + shared by the entire set of include files. FIXME-someday. */ + subpst->dependencies = (struct partial_symtab **) + obstack_alloc (&objfile->psymbol_obstack, + sizeof (struct partial_symtab *)); + subpst->dependencies[0] = pst; + subpst->number_of_dependencies = 1; + + subpst->globals_offset = + subpst->n_global_syms = + subpst->statics_offset = + subpst->n_static_syms = 0; + + subpst->readin = 0; + subpst->symtab = 0; + subpst->read_symtab = pst->read_symtab; + } + + sort_pst_symbols (pst); + + /* If there is already a psymtab or symtab for a file of this name, remove it. + (If there is a symtab, more drastic things also happen.) + This happens in VxWorks. */ + free_named_symtabs (pst->filename); + + if (num_includes == 0 + && number_dependencies == 0 + && pst->n_global_syms == 0 + && pst->n_static_syms == 0) + { + /* Throw away this psymtab, it's empty. We can't deallocate it, since + it is on the obstack, but we can forget to chain it on the list. + Empty psymtabs happen as a result of header files which don't have + any symbols in them. There can be a lot of them. But this check + is wrong, in that a psymtab with N_SLINE entries but nothing else + is not empty, but we don't realize that. Fixing that without slowing + things down might be tricky. + It's also wrong if we're using the quick look-up tables, as + we can get empty psymtabs from modules with no routines in + them. */ + + discard_psymtab (pst); + + /* Indicate that psymtab was thrown away. */ + pst = (struct partial_symtab *) NULL; + + } + return pst; +} + + +/* Get the nesting depth for the source line identified by INDEX. */ + +static unsigned long +hpread_get_scope_start (sltpointer index, struct objfile *objfile) +{ + union sltentry *sl_bufp; + + sl_bufp = hpread_get_slt (index, objfile); + return sl_bufp->sspec.backptr.dnttp.index; +} + +/* Get the source line number the the line identified by INDEX. */ + +static unsigned long +hpread_get_line (sltpointer index, struct objfile *objfile) +{ + union sltentry *sl_bufp; + + sl_bufp = hpread_get_slt (index, objfile); + return sl_bufp->snorm.line; +} + +/* Find the code address associated with a given sltpointer */ + +static CORE_ADDR +hpread_get_location (sltpointer index, struct objfile *objfile) +{ + union sltentry *sl_bufp; + int i; + + /* code location of special sltentrys is determined from context */ + sl_bufp = hpread_get_slt (index, objfile); + + if (sl_bufp->snorm.sltdesc == SLT_END) + { + /* find previous normal sltentry and get address */ + for (i = 0; ((sl_bufp->snorm.sltdesc != SLT_NORMAL) && + (sl_bufp->snorm.sltdesc != SLT_NORMAL_OFFSET) && + (sl_bufp->snorm.sltdesc != SLT_EXIT)); i++) + sl_bufp = hpread_get_slt (index - i, objfile); + if (sl_bufp->snorm.sltdesc == SLT_NORMAL_OFFSET) + return sl_bufp->snormoff.address; + else + return sl_bufp->snorm.address; + } + + /* find next normal sltentry and get address */ + for (i = 0; ((sl_bufp->snorm.sltdesc != SLT_NORMAL) && + (sl_bufp->snorm.sltdesc != SLT_NORMAL_OFFSET) && + (sl_bufp->snorm.sltdesc != SLT_EXIT)); i++) + sl_bufp = hpread_get_slt (index + i, objfile); + if (sl_bufp->snorm.sltdesc == SLT_NORMAL_OFFSET) + return sl_bufp->snormoff.address; + else + return sl_bufp->snorm.address; +} + + +/* Return 1 if an HP debug symbol of type KIND has a name associated with + * it, else return 0. (This function is not currently used, but I'll + * leave it here in case it proves useful later on. - RT). + */ + +int +hpread_has_name (enum dntt_entry_type kind) +{ + switch (kind) + { + case DNTT_TYPE_SRCFILE: + case DNTT_TYPE_MODULE: + case DNTT_TYPE_FUNCTION: + case DNTT_TYPE_DOC_FUNCTION: + case DNTT_TYPE_ENTRY: + case DNTT_TYPE_IMPORT: + case DNTT_TYPE_LABEL: + case DNTT_TYPE_FPARAM: + case DNTT_TYPE_SVAR: + case DNTT_TYPE_DVAR: + case DNTT_TYPE_CONST: + case DNTT_TYPE_TYPEDEF: + case DNTT_TYPE_TAGDEF: + case DNTT_TYPE_MEMENUM: + case DNTT_TYPE_FIELD: + case DNTT_TYPE_SA: + case DNTT_TYPE_BLOCKDATA: + case DNTT_TYPE_MEMFUNC: + case DNTT_TYPE_DOC_MEMFUNC: + return 1; + + case DNTT_TYPE_BEGIN: + case DNTT_TYPE_END: + case DNTT_TYPE_POINTER: + case DNTT_TYPE_ENUM: + case DNTT_TYPE_SET: + case DNTT_TYPE_ARRAY: + case DNTT_TYPE_STRUCT: + case DNTT_TYPE_UNION: + case DNTT_TYPE_VARIANT: + case DNTT_TYPE_FILE: + case DNTT_TYPE_FUNCTYPE: + case DNTT_TYPE_SUBRANGE: + case DNTT_TYPE_WITH: + case DNTT_TYPE_COMMON: + case DNTT_TYPE_COBSTRUCT: + case DNTT_TYPE_XREF: + case DNTT_TYPE_MACRO: + case DNTT_TYPE_CLASS_SCOPE: + case DNTT_TYPE_REFERENCE: + case DNTT_TYPE_PTRMEM: + case DNTT_TYPE_PTRMEMFUNC: + case DNTT_TYPE_CLASS: + case DNTT_TYPE_GENFIELD: + case DNTT_TYPE_VFUNC: + case DNTT_TYPE_MEMACCESS: + case DNTT_TYPE_INHERITANCE: + case DNTT_TYPE_FRIEND_CLASS: + case DNTT_TYPE_FRIEND_FUNC: + case DNTT_TYPE_MODIFIER: + case DNTT_TYPE_OBJECT_ID: + case DNTT_TYPE_TEMPLATE: + case DNTT_TYPE_TEMPLATE_ARG: + case DNTT_TYPE_FUNC_TEMPLATE: + case DNTT_TYPE_LINK: + /* DNTT_TYPE_DYN_ARRAY_DESC ? */ + /* DNTT_TYPE_DESC_SUBRANGE ? */ + /* DNTT_TYPE_BEGIN_EXT ? */ + /* DNTT_TYPE_INLN ? */ + /* DNTT_TYPE_INLN_LIST ? */ + /* DNTT_TYPE_ALIAS ? */ + default: + return 0; + } +} + +/* Do the dirty work of reading in the full symbol from a partial symbol + table. */ + +static void +hpread_psymtab_to_symtab_1 (struct partial_symtab *pst) +{ + struct cleanup *old_chain; + int i; + + /* Get out quick if passed junk. */ + if (!pst) + return; + + /* Complain if we've already read in this symbol table. */ + if (pst->readin) + { + fprintf_unfiltered (gdb_stderr, "Psymtab for %s already read in." + " Shouldn't happen.\n", + pst->filename); + return; + } + + /* Read in all partial symtabs on which this one is dependent */ + for (i = 0; i < pst->number_of_dependencies; i++) + if (!pst->dependencies[i]->readin) + { + /* Inform about additional files that need to be read in. */ + if (info_verbose) + { + fputs_filtered (" ", gdb_stdout); + wrap_here (""); + fputs_filtered ("and ", gdb_stdout); + wrap_here (""); + printf_filtered ("%s...", pst->dependencies[i]->filename); + wrap_here (""); /* Flush output */ + gdb_flush (gdb_stdout); + } + hpread_psymtab_to_symtab_1 (pst->dependencies[i]); + } + + /* If it's real... */ + if (LDSYMLEN (pst)) + { + /* Init stuff necessary for reading in symbols */ + buildsym_init (); + old_chain = make_cleanup (really_free_pendings, 0); + + pst->symtab = + hpread_expand_symtab (pst->objfile, LDSYMOFF (pst), LDSYMLEN (pst), + pst->textlow, pst->texthigh - pst->textlow, + pst->section_offsets, pst->filename); + sort_symtab_syms (pst->symtab); + + do_cleanups (old_chain); + } + + pst->readin = 1; +} + +/* Read in all of the symbols for a given psymtab for real. + Be verbose about it if the user wants that. */ + +void +hpread_psymtab_to_symtab (struct partial_symtab *pst) +{ + /* Get out quick if given junk. */ + if (!pst) + return; + + /* Sanity check. */ + if (pst->readin) + { + fprintf_unfiltered (gdb_stderr, "Psymtab for %s already read in." + " Shouldn't happen.\n", + pst->filename); + return; + } + + /* elz: setting the flag to indicate that the code of the target + was compiled using an HP compiler (aCC, cc) + the processing_acc_compilation variable is declared in the + file buildsym.h, the HP_COMPILED_TARGET is defined to be equal + to 3 in the file tm_hppa.h */ + + processing_gcc_compilation = 0; + + if (LDSYMLEN (pst) || pst->number_of_dependencies) + { + /* Print the message now, before reading the string table, + to avoid disconcerting pauses. */ + if (info_verbose) + { + printf_filtered ("Reading in symbols for %s...", pst->filename); + gdb_flush (gdb_stdout); + } + + hpread_psymtab_to_symtab_1 (pst); + + /* Match with global symbols. This only needs to be done once, + after all of the symtabs and dependencies have been read in. */ + scan_file_globals (pst->objfile); + + /* Finish up the debug error message. */ + if (info_verbose) + printf_filtered ("done.\n"); + } +} + +/* Read in a defined section of a specific object file's symbols. + + DESC is the file descriptor for the file, positioned at the + beginning of the symtab + SYM_OFFSET is the offset within the file of + the beginning of the symbols we want to read + SYM_SIZE is the size of the symbol info to read in. + TEXT_OFFSET is the beginning of the text segment we are reading symbols for + TEXT_SIZE is the size of the text segment read in. + SECTION_OFFSETS are the relocation offsets which get added to each symbol. */ + +static struct symtab * +hpread_expand_symtab (struct objfile *objfile, int sym_offset, int sym_size, + CORE_ADDR text_offset, int text_size, + struct section_offsets *section_offsets, char *filename) +{ + char *namestring; + union dnttentry *dn_bufp; + unsigned max_symnum; + int at_module_boundary = 0; + /* 1 => at end, -1 => at beginning */ + + int sym_index = sym_offset / sizeof (struct dntt_type_block); + + current_objfile = objfile; + subfile_stack = 0; + + last_source_file = 0; + + /* Demangling style -- if EDG style already set, don't change it, + as HP style causes some problems with the KAI EDG compiler */ + if (current_demangling_style != edg_demangling) + { + /* Otherwise, ensure that we are using HP style demangling */ + set_demangling_style (HP_DEMANGLING_STYLE_STRING); + } + + dn_bufp = hpread_get_lntt (sym_index, objfile); + if (!((dn_bufp->dblock.kind == (unsigned char) DNTT_TYPE_SRCFILE) || + (dn_bufp->dblock.kind == (unsigned char) DNTT_TYPE_MODULE))) + { + start_symtab ("globals", NULL, 0); + record_debugformat ("HP"); + } + + /* The psymtab builder (hp-psymtab-read.c) is the one that + * determined the "sym_size" argument (i.e. how many DNTT symbols + * are in this symtab), which we use to compute "max_symnum" + * (point in DNTT to which we read). + * + * Perhaps this should be changed so that + * process_one_debug_symbol() "knows" when + * to stop reading (based on reading from the MODULE to the matching + * END), and take out this reliance on a #-syms being passed in... + * (I'm worried about the reliability of this number). But I'll + * leave it as-is, for now. - RT + * + * The change above has been made. I've left the "for" loop control + * in to prepare for backing this out again. -JB + */ + max_symnum = sym_size / sizeof (struct dntt_type_block); + /* No reason to multiply on pst side and divide on sym side... FIXME */ + + /* Read in and process each debug symbol within the specified range. + */ + for (symnum = 0; + symnum < max_symnum; + symnum++) + { + QUIT; /* Allow this to be interruptable */ + dn_bufp = hpread_get_lntt (sym_index + symnum, objfile); + + if (dn_bufp->dblock.extension) + continue; + + /* Yow! We call SET_NAMESTRING on things without names! */ + SET_NAMESTRING (dn_bufp, &namestring, objfile); + + hpread_process_one_debug_symbol (dn_bufp, namestring, section_offsets, + objfile, text_offset, text_size, + filename, symnum + sym_index, + &at_module_boundary + ); + + /* OLD COMMENTS: This routine is only called for psts. All psts + * correspond to MODULES. If we ever do lazy-reading of globals + * from the LNTT, then there will be a pst which ends when the + * LNTT ends, and not at an END MODULE entry. Then we'll have + * to re-visit this break. + + if( at_end_of_module ) + break; + + */ + + /* We no longer break out of the loop when we reach the end of a + module. The reason is that with CTTI, the compiler can generate + function symbols (for template function instantiations) which are not + in any module; typically they show up beyond a module's end, and + before the next module's start. We include them in the current + module. However, we still don't trust the MAX_SYMNUM value from + the psymtab, so we break out if we enter a new module. */ + + if (at_module_boundary == -1) + break; + } + + current_objfile = NULL; + hp_som_som_object_present = 1; /* Indicate we've processed an HP SOM SOM file */ + + return end_symtab (text_offset + text_size, objfile, SECT_OFF_TEXT (objfile)); +} + + + + +/* Convert basic types from HP debug format into GDB internal format. */ + +static int +hpread_type_translate (dnttpointer typep) +{ + if (!typep.dntti.immediate) + { + error ("error in hpread_type_translate\n."); + return FT_VOID; + } + + switch (typep.dntti.type) + { + case HP_TYPE_BOOLEAN: + case HP_TYPE_BOOLEAN_S300_COMPAT: + case HP_TYPE_BOOLEAN_VAX_COMPAT: + return FT_BOOLEAN; + case HP_TYPE_CHAR: /* C signed char, C++ plain char */ + + case HP_TYPE_WIDE_CHAR: + return FT_CHAR; + case HP_TYPE_INT: + if (typep.dntti.bitlength <= 8) + return FT_SIGNED_CHAR; /* C++ signed char */ + if (typep.dntti.bitlength <= 16) + return FT_SHORT; + if (typep.dntti.bitlength <= 32) + return FT_INTEGER; + return FT_LONG_LONG; + case HP_TYPE_LONG: + if (typep.dntti.bitlength <= 8) + return FT_SIGNED_CHAR; /* C++ signed char. */ + return FT_LONG; + case HP_TYPE_UNSIGNED_LONG: + if (typep.dntti.bitlength <= 8) + return FT_UNSIGNED_CHAR; /* C/C++ unsigned char */ + if (typep.dntti.bitlength <= 16) + return FT_UNSIGNED_SHORT; + if (typep.dntti.bitlength <= 32) + return FT_UNSIGNED_LONG; + return FT_UNSIGNED_LONG_LONG; + case HP_TYPE_UNSIGNED_INT: + if (typep.dntti.bitlength <= 8) + return FT_UNSIGNED_CHAR; + if (typep.dntti.bitlength <= 16) + return FT_UNSIGNED_SHORT; + if (typep.dntti.bitlength <= 32) + return FT_UNSIGNED_INTEGER; + return FT_UNSIGNED_LONG_LONG; + case HP_TYPE_REAL: + case HP_TYPE_REAL_3000: + case HP_TYPE_DOUBLE: + if (typep.dntti.bitlength == 64) + return FT_DBL_PREC_FLOAT; + if (typep.dntti.bitlength == 128) + return FT_EXT_PREC_FLOAT; + return FT_FLOAT; + case HP_TYPE_COMPLEX: + case HP_TYPE_COMPLEXS3000: + if (typep.dntti.bitlength == 128) + return FT_DBL_PREC_COMPLEX; + if (typep.dntti.bitlength == 192) + return FT_EXT_PREC_COMPLEX; + return FT_COMPLEX; + case HP_TYPE_VOID: + return FT_VOID; + case HP_TYPE_STRING200: + case HP_TYPE_LONGSTRING200: + case HP_TYPE_FTN_STRING_SPEC: + case HP_TYPE_MOD_STRING_SPEC: + case HP_TYPE_MOD_STRING_3000: + case HP_TYPE_FTN_STRING_S300_COMPAT: + case HP_TYPE_FTN_STRING_VAX_COMPAT: + return FT_STRING; + case HP_TYPE_TEMPLATE_ARG: + return FT_TEMPLATE_ARG; + case HP_TYPE_TEXT: + case HP_TYPE_FLABEL: + case HP_TYPE_PACKED_DECIMAL: + case HP_TYPE_ANYPOINTER: + case HP_TYPE_GLOBAL_ANYPOINTER: + case HP_TYPE_LOCAL_ANYPOINTER: + default: + warning ("hpread_type_translate: unhandled type code.\n"); + return FT_VOID; + } +} + +/* Given a position in the DNTT, return a pointer to the + * already-built "struct type" (if any), for the type defined + * at that position. + */ + +static struct type ** +hpread_lookup_type (dnttpointer hp_type, struct objfile *objfile) +{ + unsigned old_len; + int index = hp_type.dnttp.index; + int size_changed = 0; + + /* The immediate flag indicates this doesn't actually point to + * a type DNTT. + */ + if (hp_type.dntti.immediate) + return NULL; + + /* For each objfile, we maintain a "type vector". + * This an array of "struct type *"'s with one pointer per DNTT index. + * Given a DNTT index, we look in this array to see if we have + * already processed this DNTT and if it is a type definition. + * If so, then we can locate a pointer to the already-built + * "struct type", and not build it again. + * + * The need for this arises because our DNTT-walking code wanders + * around. In particular, it will encounter the same type multiple + * times (once for each object of that type). We don't want to + * built multiple "struct type"'s for the same thing. + * + * Having said this, I should point out that this type-vector is + * an expensive way to keep track of this. If most DNTT entries are + * 3 words, the type-vector will be 1/3 the size of the DNTT itself. + * Alternative solutions: + * - Keep a compressed or hashed table. Less memory, but more expensive + * to search and update. + * - (Suggested by JB): Overwrite the DNTT entry itself + * with the info. Create a new type code "ALREADY_BUILT", and modify + * the DNTT to have that type code and point to the already-built entry. + * -RT + */ + + if (index < LNTT_SYMCOUNT (objfile)) + { + if (index >= DNTT_TYPE_VECTOR_LENGTH (objfile)) + { + old_len = DNTT_TYPE_VECTOR_LENGTH (objfile); + + /* See if we need to allocate a type-vector. */ + if (old_len == 0) + { + DNTT_TYPE_VECTOR_LENGTH (objfile) = LNTT_SYMCOUNT (objfile) + GNTT_SYMCOUNT (objfile); + DNTT_TYPE_VECTOR (objfile) = (struct type **) + xmmalloc (objfile->md, DNTT_TYPE_VECTOR_LENGTH (objfile) * sizeof (struct type *)); + memset (&DNTT_TYPE_VECTOR (objfile)[old_len], 0, + (DNTT_TYPE_VECTOR_LENGTH (objfile) - old_len) * + sizeof (struct type *)); + } + + /* See if we need to resize type-vector. With my change to + * initially allocate a correct-size type-vector, this code + * should no longer trigger. + */ + while (index >= DNTT_TYPE_VECTOR_LENGTH (objfile)) + { + DNTT_TYPE_VECTOR_LENGTH (objfile) *= 2; + size_changed = 1; + } + if (size_changed) + { + DNTT_TYPE_VECTOR (objfile) = (struct type **) + xmrealloc (objfile->md, + (char *) DNTT_TYPE_VECTOR (objfile), + (DNTT_TYPE_VECTOR_LENGTH (objfile) * sizeof (struct type *))); + + memset (&DNTT_TYPE_VECTOR (objfile)[old_len], 0, + (DNTT_TYPE_VECTOR_LENGTH (objfile) - old_len) * + sizeof (struct type *)); + } + + } + return &DNTT_TYPE_VECTOR (objfile)[index]; + } + else + return NULL; +} + +/* Possibly allocate a GDB internal type so we can internalize HP_TYPE. + Note we'll just return the address of a GDB internal type if we already + have it lying around. */ + +static struct type * +hpread_alloc_type (dnttpointer hp_type, struct objfile *objfile) +{ + struct type **type_addr; + + type_addr = hpread_lookup_type (hp_type, objfile); + if (*type_addr == 0) + { + *type_addr = alloc_type (objfile); + + /* A hack - if we really are a C++ class symbol, then this default + * will get overriden later on. + */ + TYPE_CPLUS_SPECIFIC (*type_addr) + = (struct cplus_struct_type *) &cplus_struct_default; + } + + return *type_addr; +} + +/* Read a native enumerated type and return it in GDB internal form. */ + +static struct type * +hpread_read_enum_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile) +{ + struct type *type; + struct pending **symlist, *osyms, *syms; + struct pending *local_list = NULL; + int o_nsyms, nsyms = 0; + dnttpointer mem; + union dnttentry *memp; + char *name; + long n; + struct symbol *sym; + + /* Allocate a GDB type. If we've already read in this enum type, + * it'll return the already built GDB type, so stop here. + * (Note: I added this check, to conform with what's done for + * struct, union, class. + * I assume this is OK. - RT) + */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_ENUM) + return type; + + /* HP C supports "sized enums", where a specifier such as "short" or + "char" can be used to get enums of different sizes. So don't assume + an enum is always 4 bytes long. pai/1997-08-21 */ + TYPE_LENGTH (type) = dn_bufp->denum.bitlength / 8; + + symlist = &file_symbols; + osyms = *symlist; + o_nsyms = osyms ? osyms->nsyms : 0; + + /* Get a name for each member and add it to our list of members. + * The list of "mem" SOM records we are walking should all be + * SOM type DNTT_TYPE_MEMENUM (not checked). + */ + mem = dn_bufp->denum.firstmem; + while (mem.word && mem.word != DNTTNIL) + { + memp = hpread_get_lntt (mem.dnttp.index, objfile); + + name = VT (objfile) + memp->dmember.name; + sym = (struct symbol *) obstack_alloc (&objfile->symbol_obstack, + sizeof (struct symbol)); + memset (sym, 0, sizeof (struct symbol)); + SYMBOL_NAME (sym) = obsavestring (name, strlen (name), + &objfile->symbol_obstack); + SYMBOL_CLASS (sym) = LOC_CONST; + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + SYMBOL_VALUE (sym) = memp->dmember.value; + add_symbol_to_list (sym, symlist); + nsyms++; + mem = memp->dmember.nextmem; + } + + /* Now that we know more about the enum, fill in more info. */ + TYPE_CODE (type) = TYPE_CODE_ENUM; + TYPE_FLAGS (type) &= ~TYPE_FLAG_STUB; + TYPE_NFIELDS (type) = nsyms; + TYPE_FIELDS (type) = (struct field *) + obstack_alloc (&objfile->type_obstack, sizeof (struct field) * nsyms); + + /* Find the symbols for the members and put them into the type. + The symbols can be found in the symlist that we put them on + to cause them to be defined. osyms contains the old value + of that symlist; everything up to there was defined by us. + + Note that we preserve the order of the enum constants, so + that in something like "enum {FOO, LAST_THING=FOO}" we print + FOO, not LAST_THING. */ + for (syms = *symlist, n = 0; syms; syms = syms->next) + { + int j = 0; + if (syms == osyms) + j = o_nsyms; + for (; j < syms->nsyms; j++, n++) + { + struct symbol *xsym = syms->symbol[j]; + SYMBOL_TYPE (xsym) = type; + TYPE_FIELD_NAME (type, n) = SYMBOL_NAME (xsym); + TYPE_FIELD_BITPOS (type, n) = SYMBOL_VALUE (xsym); + TYPE_FIELD_BITSIZE (type, n) = 0; + } + if (syms == osyms) + break; + } + + return type; +} + +/* Read and internalize a native function debug symbol. */ + +static struct type * +hpread_read_function_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile, int newblock) +{ + struct type *type, *type1; + struct pending *syms; + struct pending *local_list = NULL; + int nsyms = 0; + dnttpointer param; + union dnttentry *paramp; + char *name; + long n; + struct symbol *sym; + int record_args = 1; + + /* See if we've already read in this type. */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_FUNC) + { + record_args = 0; /* already read in, don't modify type */ + } + else + { + /* Nope, so read it in and store it away. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_FUNCTION || + dn_bufp->dblock.kind == DNTT_TYPE_MEMFUNC) + type1 = lookup_function_type (hpread_type_lookup (dn_bufp->dfunc.retval, + objfile)); + else if (dn_bufp->dblock.kind == DNTT_TYPE_FUNCTYPE) + type1 = lookup_function_type (hpread_type_lookup (dn_bufp->dfunctype.retval, + objfile)); + else /* expect DNTT_TYPE_FUNC_TEMPLATE */ + type1 = lookup_function_type (hpread_type_lookup (dn_bufp->dfunc_template.retval, + objfile)); + replace_type (type, type1); + + /* Mark it -- in the middle of processing */ + TYPE_FLAGS (type) |= TYPE_FLAG_INCOMPLETE; + } + + /* Now examine each parameter noting its type, location, and a + wealth of other information. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_FUNCTION || + dn_bufp->dblock.kind == DNTT_TYPE_MEMFUNC) + param = dn_bufp->dfunc.firstparam; + else if (dn_bufp->dblock.kind == DNTT_TYPE_FUNCTYPE) + param = dn_bufp->dfunctype.firstparam; + else /* expect DNTT_TYPE_FUNC_TEMPLATE */ + param = dn_bufp->dfunc_template.firstparam; + while (param.word && param.word != DNTTNIL) + { + paramp = hpread_get_lntt (param.dnttp.index, objfile); + nsyms++; + param = paramp->dfparam.nextparam; + + /* Get the name. */ + name = VT (objfile) + paramp->dfparam.name; + sym = (struct symbol *) obstack_alloc (&objfile->symbol_obstack, + sizeof (struct symbol)); + (void) memset (sym, 0, sizeof (struct symbol)); + SYMBOL_NAME (sym) = obsavestring (name, strlen (name), + &objfile->symbol_obstack); + + /* Figure out where it lives. */ + if (paramp->dfparam.regparam) + SYMBOL_CLASS (sym) = LOC_REGPARM; + else if (paramp->dfparam.indirect) + SYMBOL_CLASS (sym) = LOC_REF_ARG; + else + SYMBOL_CLASS (sym) = LOC_ARG; + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + if (paramp->dfparam.copyparam) + { + SYMBOL_VALUE (sym) = paramp->dfparam.location; +#ifdef HPREAD_ADJUST_STACK_ADDRESS + SYMBOL_VALUE (sym) + += HPREAD_ADJUST_STACK_ADDRESS (CURRENT_FUNCTION_VALUE (objfile)); +#endif + /* This is likely a pass-by-invisible reference parameter, + Hack on the symbol class to make GDB happy. */ + /* ??rehrauer: This appears to be broken w/r/t to passing + C values of type float and struct. Perhaps this ought + to be highighted as a special case, but for now, just + allowing these to be LOC_ARGs seems to work fine. + */ +#if 0 + SYMBOL_CLASS (sym) = LOC_REGPARM_ADDR; +#endif + } + else + SYMBOL_VALUE (sym) = paramp->dfparam.location; + + /* Get its type. */ + SYMBOL_TYPE (sym) = hpread_type_lookup (paramp->dfparam.type, objfile); + /* Add it to the symbol list. */ + /* Note 1 (RT) At the moment, add_symbol_to_list() is also being + * called on FPARAM symbols from the process_one_debug_symbol() + * level... so parameters are getting added twice! (this shows + * up in the symbol dump you get from "maint print symbols ..."). + * Note 2 (RT) I took out the processing of FPARAM from the + * process_one_debug_symbol() level, so at the moment parameters are only + * being processed here. This seems to have no ill effect. + */ + /* Note 3 (pai/1997-08-11) I removed the add_symbol_to_list() which put + each fparam on the local_symbols list from here. Now we use the + local_list to which fparams are added below, and set the param_symbols + global to point to that at the end of this routine. */ + /* elz: I added this new list of symbols which is local to the function. + this list is the one which is actually used to build the type for the + function rather than the gloabal list pointed to by symlist. + Using a global list to keep track of the parameters is wrong, because + this function is called recursively if one parameter happend to be + a function itself with more parameters in it. Adding parameters to the + same global symbol list would not work! + Actually it did work in case of cc compiled programs where you do + not check the parameter lists of the arguments. */ + add_symbol_to_list (sym, &local_list); + + } + + /* If type was read in earlier, don't bother with modifying + the type struct */ + if (!record_args) + goto finish; + + /* Note how many parameters we found. */ + TYPE_NFIELDS (type) = nsyms; + TYPE_FIELDS (type) = (struct field *) + obstack_alloc (&objfile->type_obstack, + sizeof (struct field) * nsyms); + + /* Find the symbols for the parameters and + use them to fill parameter-type information into the function-type. + The parameter symbols can be found in the local_list that we just put them on. */ + /* Note that we preserve the order of the parameters, so + that in something like "enum {FOO, LAST_THING=FOO}" we print + FOO, not LAST_THING. */ + + /* get the parameters types from the local list not the global list + so that the type can be correctly constructed for functions which + have function as parameters */ + for (syms = local_list, n = 0; syms; syms = syms->next) + { + int j = 0; + for (j = 0; j < syms->nsyms; j++, n++) + { + struct symbol *xsym = syms->symbol[j]; + TYPE_FIELD_NAME (type, n) = SYMBOL_NAME (xsym); + TYPE_FIELD_TYPE (type, n) = SYMBOL_TYPE (xsym); + TYPE_FIELD_ARTIFICIAL (type, n) = 0; + TYPE_FIELD_BITSIZE (type, n) = 0; + } + } + /* Mark it as having been processed */ + TYPE_FLAGS (type) &= ~(TYPE_FLAG_INCOMPLETE); + + /* Check whether we need to fix-up a class type with this function's type */ + if (fixup_class && (fixup_method == type)) + { + fixup_class_method_type (fixup_class, fixup_method, objfile); + fixup_class = NULL; + fixup_method = NULL; + } + + /* Set the param list of this level of the context stack + to our local list. Do this only if this function was + called for creating a new block, and not if it was called + simply to get the function type. This prevents recursive + invocations from trashing param_symbols. */ +finish: + if (newblock) + param_symbols = local_list; + + return type; +} + + +/* Read and internalize a native DOC function debug symbol. */ +/* This is almost identical to hpread_read_function_type(), except + * for references to dn_bufp->ddocfunc instead of db_bufp->dfunc. + * Since debug information for DOC functions is more likely to be + * volatile, please leave it this way. + */ +static struct type * +hpread_read_doc_function_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile, int newblock) +{ + struct type *type, *type1; + struct pending *syms; + struct pending *local_list = NULL; + int nsyms = 0; + dnttpointer param; + union dnttentry *paramp; + char *name; + long n; + struct symbol *sym; + int record_args = 1; + + /* See if we've already read in this type. */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_FUNC) + { + record_args = 0; /* already read in, don't modify type */ + } + else + { + /* Nope, so read it in and store it away. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_DOC_FUNCTION || + dn_bufp->dblock.kind == DNTT_TYPE_DOC_MEMFUNC) + type1 = lookup_function_type (hpread_type_lookup (dn_bufp->ddocfunc.retval, + objfile)); + replace_type (type, type1); + + /* Mark it -- in the middle of processing */ + TYPE_FLAGS (type) |= TYPE_FLAG_INCOMPLETE; + } + + /* Now examine each parameter noting its type, location, and a + wealth of other information. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_DOC_FUNCTION || + dn_bufp->dblock.kind == DNTT_TYPE_DOC_MEMFUNC) + param = dn_bufp->ddocfunc.firstparam; + while (param.word && param.word != DNTTNIL) + { + paramp = hpread_get_lntt (param.dnttp.index, objfile); + nsyms++; + param = paramp->dfparam.nextparam; + + /* Get the name. */ + name = VT (objfile) + paramp->dfparam.name; + sym = (struct symbol *) obstack_alloc (&objfile->symbol_obstack, + sizeof (struct symbol)); + (void) memset (sym, 0, sizeof (struct symbol)); + SYMBOL_NAME (sym) = name; + + /* Figure out where it lives. */ + if (paramp->dfparam.regparam) + SYMBOL_CLASS (sym) = LOC_REGPARM; + else if (paramp->dfparam.indirect) + SYMBOL_CLASS (sym) = LOC_REF_ARG; + else + SYMBOL_CLASS (sym) = LOC_ARG; + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + if (paramp->dfparam.copyparam) + { + SYMBOL_VALUE (sym) = paramp->dfparam.location; +#ifdef HPREAD_ADJUST_STACK_ADDRESS + SYMBOL_VALUE (sym) + += HPREAD_ADJUST_STACK_ADDRESS (CURRENT_FUNCTION_VALUE (objfile)); +#endif + /* This is likely a pass-by-invisible reference parameter, + Hack on the symbol class to make GDB happy. */ + /* ??rehrauer: This appears to be broken w/r/t to passing + C values of type float and struct. Perhaps this ought + to be highighted as a special case, but for now, just + allowing these to be LOC_ARGs seems to work fine. + */ +#if 0 + SYMBOL_CLASS (sym) = LOC_REGPARM_ADDR; +#endif + } + else + SYMBOL_VALUE (sym) = paramp->dfparam.location; + + /* Get its type. */ + SYMBOL_TYPE (sym) = hpread_type_lookup (paramp->dfparam.type, objfile); + /* Add it to the symbol list. */ + /* Note 1 (RT) At the moment, add_symbol_to_list() is also being + * called on FPARAM symbols from the process_one_debug_symbol() + * level... so parameters are getting added twice! (this shows + * up in the symbol dump you get from "maint print symbols ..."). + * Note 2 (RT) I took out the processing of FPARAM from the + * process_one_debug_symbol() level, so at the moment parameters are only + * being processed here. This seems to have no ill effect. + */ + /* Note 3 (pai/1997-08-11) I removed the add_symbol_to_list() which put + each fparam on the local_symbols list from here. Now we use the + local_list to which fparams are added below, and set the param_symbols + global to point to that at the end of this routine. */ + + /* elz: I added this new list of symbols which is local to the function. + this list is the one which is actually used to build the type for the + function rather than the gloabal list pointed to by symlist. + Using a global list to keep track of the parameters is wrong, because + this function is called recursively if one parameter happend to be + a function itself with more parameters in it. Adding parameters to the + same global symbol list would not work! + Actually it did work in case of cc compiled programs where you do not check the + parameter lists of the arguments. */ + add_symbol_to_list (sym, &local_list); + } + + /* If type was read in earlier, don't bother with modifying + the type struct */ + if (!record_args) + goto finish; + + /* Note how many parameters we found. */ + TYPE_NFIELDS (type) = nsyms; + TYPE_FIELDS (type) = (struct field *) + obstack_alloc (&objfile->type_obstack, + sizeof (struct field) * nsyms); + + /* Find the symbols for the parameters and + use them to fill parameter-type information into the function-type. + The parameter symbols can be found in the local_list that we just put them on. */ + /* Note that we preserve the order of the parameters, so + that in something like "enum {FOO, LAST_THING=FOO}" we print + FOO, not LAST_THING. */ + + /* get the parameters types from the local list not the global list + so that the type can be correctly constructed for functions which + have function as parameters + */ + for (syms = local_list, n = 0; syms; syms = syms->next) + { + int j = 0; + for (j = 0; j < syms->nsyms; j++, n++) + { + struct symbol *xsym = syms->symbol[j]; + TYPE_FIELD_NAME (type, n) = SYMBOL_NAME (xsym); + TYPE_FIELD_TYPE (type, n) = SYMBOL_TYPE (xsym); + TYPE_FIELD_ARTIFICIAL (type, n) = 0; + TYPE_FIELD_BITSIZE (type, n) = 0; + } + } + + /* Mark it as having been processed */ + TYPE_FLAGS (type) &= ~(TYPE_FLAG_INCOMPLETE); + + /* Check whether we need to fix-up a class type with this function's type */ + if (fixup_class && (fixup_method == type)) + { + fixup_class_method_type (fixup_class, fixup_method, objfile); + fixup_class = NULL; + fixup_method = NULL; + } + + /* Set the param list of this level of the context stack + to our local list. Do this only if this function was + called for creating a new block, and not if it was called + simply to get the function type. This prevents recursive + invocations from trashing param_symbols. */ +finish: + if (newblock) + param_symbols = local_list; + + return type; +} + + + +/* A file-level variable which keeps track of the current-template + * being processed. Set in hpread_read_struct_type() while processing + * a template type. Referred to in hpread_get_nth_templ_arg(). + * Yes, this is a kludge, but it arises from the kludge that already + * exists in symtab.h, namely the fact that they encode + * "template argument n" with fundamental type FT_TEMPLATE_ARG and + * bitlength n. This means that deep in processing fundamental types + * I need to ask the question "what template am I in the middle of?". + * The alternative to stuffing a global would be to pass an argument + * down the chain of calls just for this purpose. + * + * There may be problems handling nested templates... tough. + */ +static struct type *current_template = NULL; + +/* Read in and internalize a structure definition. + * This same routine is called for struct, union, and class types. + * Also called for templates, since they build a very similar + * type entry as for class types. + */ + +static struct type * +hpread_read_struct_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile) +{ + /* The data members get linked together into a list of struct nextfield's */ + struct nextfield + { + struct nextfield *next; + struct field field; + unsigned char attributes; /* store visibility and virtuality info */ +#define ATTR_VIRTUAL 1 +#define ATTR_PRIVATE 2 +#define ATTR_PROTECT 3 + }; + + + /* The methods get linked together into a list of struct next_fn_field's */ + struct next_fn_field + { + struct next_fn_field *next; + struct fn_fieldlist field; + struct fn_field fn_field; + int num_fn_fields; + }; + + /* The template args get linked together into a list of struct next_template's */ + struct next_template + { + struct next_template *next; + struct template_arg arg; + }; + + /* The template instantiations get linked together into a list of these... */ + struct next_instantiation + { + struct next_instantiation *next; + struct type *t; + }; + + struct type *type; + struct type *baseclass; + struct type *memtype; + struct nextfield *list = 0, *tmp_list = 0; + struct next_fn_field *fn_list = 0; + struct next_fn_field *fn_p; + struct next_template *t_new, *t_list = 0; + struct nextfield *new; + struct next_fn_field *fn_new; + struct next_instantiation *i_new, *i_list = 0; + int n, nfields = 0, n_fn_fields = 0, n_fn_fields_total = 0; + int n_base_classes = 0, n_templ_args = 0; + int ninstantiations = 0; + dnttpointer field, fn_field, parent; + union dnttentry *fieldp, *fn_fieldp, *parentp; + int i; + int static_member = 0; + int const_member = 0; + int volatile_member = 0; + unsigned long vtbl_offset; + int need_bitvectors = 0; + char *method_name = NULL; + char *method_alias = NULL; + + + /* Is it something we've already dealt with? */ + type = hpread_alloc_type (hp_type, objfile); + if ((TYPE_CODE (type) == TYPE_CODE_STRUCT) || + (TYPE_CODE (type) == TYPE_CODE_UNION) || + (TYPE_CODE (type) == TYPE_CODE_CLASS) || + (TYPE_CODE (type) == TYPE_CODE_TEMPLATE)) + return type; + + /* Get the basic type correct. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_STRUCT) + { + TYPE_CODE (type) = TYPE_CODE_STRUCT; + TYPE_LENGTH (type) = dn_bufp->dstruct.bitlength / 8; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_UNION) + { + TYPE_CODE (type) = TYPE_CODE_UNION; + TYPE_LENGTH (type) = dn_bufp->dunion.bitlength / 8; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS) + { + TYPE_CODE (type) = TYPE_CODE_CLASS; + TYPE_LENGTH (type) = dn_bufp->dclass.bitlength / 8; + + /* Overrides the TYPE_CPLUS_SPECIFIC(type) with allocated memory + * rather than &cplus_struct_default. + */ + allocate_cplus_struct_type (type); + + /* Fill in declared-type. + * (The C++ compiler will emit TYPE_CODE_CLASS + * for all 3 of "class", "struct" + * "union", and we have to look at the "class_decl" field if we + * want to know how it was really declared) + */ + /* (0==class, 1==union, 2==struct) */ + TYPE_DECLARED_TYPE (type) = dn_bufp->dclass.class_decl; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + /* Get the basic type correct. */ + TYPE_CODE (type) = TYPE_CODE_TEMPLATE; + allocate_cplus_struct_type (type); + TYPE_DECLARED_TYPE (type) = DECLARED_TYPE_TEMPLATE; + } + else + return type; + + + TYPE_FLAGS (type) &= ~TYPE_FLAG_STUB; + + /* For classes, read the parent list. + * Question (RT): Do we need to do this for templates also? + */ + if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS) + { + + /* First read the parent-list (classes from which we derive fields) */ + parent = dn_bufp->dclass.parentlist; + while (parent.word && parent.word != DNTTNIL) + { + parentp = hpread_get_lntt (parent.dnttp.index, objfile); + + /* "parentp" should point to a DNTT_TYPE_INHERITANCE record */ + + /* Get space to record the next field/data-member. */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + FIELD_BITSIZE (list->field) = 0; + + /* The "classname" field is actually a DNTT pointer to the base class */ + baseclass = hpread_type_lookup (parentp->dinheritance.classname, + objfile); + FIELD_TYPE (list->field) = baseclass; + + list->field.name = type_name_no_tag (FIELD_TYPE (list->field)); + + list->attributes = 0; + + /* Check for virtuality of base, and set the + * offset of the base subobject within the object. + * (Offset set to -1 for virtual bases (for now).) + */ + if (parentp->dinheritance.Virtual) + { + B_SET (&(list->attributes), ATTR_VIRTUAL); + parentp->dinheritance.offset = -1; + } + else + FIELD_BITPOS (list->field) = parentp->dinheritance.offset; + + /* Check visibility */ + switch (parentp->dinheritance.visibility) + { + case 1: + B_SET (&(list->attributes), ATTR_PROTECT); + break; + case 2: + B_SET (&(list->attributes), ATTR_PRIVATE); + break; + } + + n_base_classes++; + nfields++; + + parent = parentp->dinheritance.next; + } + } + + /* For templates, read the template argument list. + * This must be done before processing the member list, because + * the member list may refer back to this. E.g.: + * template <class T1, class T2> class q2 { + * public: + * T1 a; + * T2 b; + * }; + * We need to read the argument list "T1", "T2" first. + */ + if (dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + /* Kludge alert: This stuffs a global "current_template" which + * is referred to by hpread_get_nth_templ_arg(). The global + * is cleared at the end of this routine. + */ + current_template = type; + + /* Read in the argument list */ + field = dn_bufp->dtemplate.arglist; + while (field.word && field.word != DNTTNIL) + { + /* Get this template argument */ + fieldp = hpread_get_lntt (field.dnttp.index, objfile); + if (fieldp->dblock.kind != DNTT_TYPE_TEMPLATE_ARG) + { + warning ("Invalid debug info: Template argument entry is of wrong kind"); + break; + } + /* Bump the count */ + n_templ_args++; + /* Allocate and fill in a struct next_template */ + t_new = (struct next_template *) alloca (sizeof (struct next_template)); + t_new->next = t_list; + t_list = t_new; + t_list->arg.name = VT (objfile) + fieldp->dtempl_arg.name; + t_list->arg.type = hpread_read_templ_arg_type (field, fieldp, + objfile, t_list->arg.name); + /* Walk to the next template argument */ + field = fieldp->dtempl_arg.nextarg; + } + } + + TYPE_NTEMPLATE_ARGS (type) = n_templ_args; + + if (n_templ_args > 0) + TYPE_TEMPLATE_ARGS (type) = (struct template_arg *) + obstack_alloc (&objfile->type_obstack, sizeof (struct template_arg) * n_templ_args); + for (n = n_templ_args; t_list; t_list = t_list->next) + { + n -= 1; + TYPE_TEMPLATE_ARG (type, n) = t_list->arg; + } + + /* Next read in and internalize all the fields/members. */ + if (dn_bufp->dblock.kind == DNTT_TYPE_STRUCT) + field = dn_bufp->dstruct.firstfield; + else if (dn_bufp->dblock.kind == DNTT_TYPE_UNION) + field = dn_bufp->dunion.firstfield; + else if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS) + field = dn_bufp->dclass.memberlist; + else if (dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + field = dn_bufp->dtemplate.memberlist; + else + field.word = DNTTNIL; + + while (field.word && field.word != DNTTNIL) + { + fieldp = hpread_get_lntt (field.dnttp.index, objfile); + + /* At this point "fieldp" may point to either a DNTT_TYPE_FIELD + * or a DNTT_TYPE_GENFIELD record. + */ + vtbl_offset = 0; + static_member = 0; + const_member = 0; + volatile_member = 0; + + if (fieldp->dblock.kind == DNTT_TYPE_GENFIELD) + { + + /* The type will be GENFIELD if the field is a method or + * a static member (or some other cases -- see below) + */ + + /* Follow a link to get to the record for the field. */ + fn_field = fieldp->dgenfield.field; + fn_fieldp = hpread_get_lntt (fn_field.dnttp.index, objfile); + + /* Virtual funcs are indicated by a VFUNC which points to the + * real entry + */ + if (fn_fieldp->dblock.kind == DNTT_TYPE_VFUNC) + { + vtbl_offset = fn_fieldp->dvfunc.vtbl_offset; + fn_field = fn_fieldp->dvfunc.funcptr; + fn_fieldp = hpread_get_lntt (fn_field.dnttp.index, objfile); + } + + /* A function's entry may be preceded by a modifier which + * labels it static/constant/volatile. + */ + if (fn_fieldp->dblock.kind == DNTT_TYPE_MODIFIER) + { + static_member = fn_fieldp->dmodifier.m_static; + const_member = fn_fieldp->dmodifier.m_const; + volatile_member = fn_fieldp->dmodifier.m_volatile; + fn_field = fn_fieldp->dmodifier.type; + fn_fieldp = hpread_get_lntt (fn_field.dnttp.index, objfile); + } + + /* Check whether we have a method */ + if ((fn_fieldp->dblock.kind == DNTT_TYPE_MEMFUNC) || + (fn_fieldp->dblock.kind == DNTT_TYPE_FUNCTION) || + (fn_fieldp->dblock.kind == DNTT_TYPE_DOC_MEMFUNC) || + (fn_fieldp->dblock.kind == DNTT_TYPE_DOC_FUNCTION)) + { + /* Method found */ + + short ix = 0; + + /* Look up function type of method */ + memtype = hpread_type_lookup (fn_field, objfile); + + /* Methods can be seen before classes in the SOM records. + If we are processing this class because it's a parameter of a + method, at this point the method's type is actually incomplete; + we'll have to fix it up later; mark the class for this. */ + + if (TYPE_INCOMPLETE (memtype)) + { + TYPE_FLAGS (type) |= TYPE_FLAG_INCOMPLETE; + if (fixup_class) + warning ("Two classes to fix up for method?? Type information may be incorrect for some classes."); + if (fixup_method) + warning ("Two methods to be fixed up at once?? Type information may be incorrect for some classes."); + fixup_class = type; /* remember this class has to be fixed up */ + fixup_method = memtype; /* remember the method type to be used in fixup */ + } + + /* HP aCC generates operator names without the "operator" keyword, and + generates null strings as names for operators that are + user-defined type conversions to basic types (e.g. operator int ()). + So try to reconstruct name as best as possible. */ + + method_name = (char *) (VT (objfile) + fn_fieldp->dfunc.name); + method_alias = (char *) (VT (objfile) + fn_fieldp->dfunc.alias); + + if (!method_name || /* no name */ + !*method_name || /* or null name */ + cplus_mangle_opname (method_name, DMGL_ANSI)) /* or name is an operator like "<" */ + { + char *tmp_name = cplus_demangle (method_alias, DMGL_ANSI); + char *op_string = strstr (tmp_name, "operator"); + method_name = xmalloc (strlen (op_string) + 1); /* don't overwrite VT! */ + strcpy (method_name, op_string); + } + + /* First check if a method of the same name has already been seen. */ + fn_p = fn_list; + while (fn_p) + { + if (STREQ (fn_p->field.name, method_name)) + break; + fn_p = fn_p->next; + } + + /* If no such method was found, allocate a new entry in the list */ + if (!fn_p) + { + /* Get space to record this member function */ + /* Note: alloca used; this will disappear on routine exit */ + fn_new = (struct next_fn_field *) alloca (sizeof (struct next_fn_field)); + fn_new->next = fn_list; + fn_list = fn_new; + + /* Fill in the fields of the struct nextfield */ + + /* Record the (unmangled) method name */ + fn_list->field.name = method_name; + /* Initial space for overloaded methods */ + /* Note: xmalloc is used; this will persist after this routine exits */ + fn_list->field.fn_fields = (struct fn_field *) xmalloc (5 * (sizeof (struct fn_field))); + fn_list->field.length = 1; /* Init # of overloaded instances */ + fn_list->num_fn_fields = 5; /* # of entries for which space allocated */ + fn_p = fn_list; + ix = 0; /* array index for fn_field */ + /* Bump the total count of the distinctly named methods */ + n_fn_fields++; + } + else + /* Another overloaded instance of an already seen method name */ + { + if (++(fn_p->field.length) > fn_p->num_fn_fields) + { + /* Increase space allocated for overloaded instances */ + fn_p->field.fn_fields + = (struct fn_field *) xrealloc (fn_p->field.fn_fields, + (fn_p->num_fn_fields + 5) * sizeof (struct fn_field)); + fn_p->num_fn_fields += 5; + } + ix = fn_p->field.length - 1; /* array index for fn_field */ + } + + /* "physname" is intended to be the name of this overloaded instance. */ + if ((fn_fieldp->dfunc.language == HP_LANGUAGE_CPLUSPLUS) && + method_alias && + *method_alias) /* not a null string */ + fn_p->field.fn_fields[ix].physname = method_alias; + else + fn_p->field.fn_fields[ix].physname = method_name; + /* What's expected here is the function type */ + /* But mark it as NULL if the method was incompletely processed + We'll fix this up later when the method is fully processed */ + if (TYPE_INCOMPLETE (memtype)) + fn_p->field.fn_fields[ix].type = NULL; + else + fn_p->field.fn_fields[ix].type = memtype; + + /* For virtual functions, fill in the voffset field with the + * virtual table offset. (This is just copied over from the + * SOM record; not sure if it is what GDB expects here...). + * But if the function is a static method, set it to 1. + * + * Note that we have to add 1 because 1 indicates a static + * method, and 0 indicates a non-static, non-virtual method */ + + if (static_member) + fn_p->field.fn_fields[ix].voffset = VOFFSET_STATIC; + else + fn_p->field.fn_fields[ix].voffset = vtbl_offset ? vtbl_offset + 1 : 0; + + /* Also fill in the fcontext field with the current + * class. (The latter isn't quite right: should be the baseclass + * that defines the virtual function... Note we do have + * a variable "baseclass" that we could stuff into the fcontext + * field, but "baseclass" isn't necessarily right either, + * since the virtual function could have been defined more + * than one level up). + */ + + if (vtbl_offset != 0) + fn_p->field.fn_fields[ix].fcontext = type; + else + fn_p->field.fn_fields[ix].fcontext = NULL; + + /* Other random fields pertaining to this method */ + fn_p->field.fn_fields[ix].is_const = const_member; + fn_p->field.fn_fields[ix].is_volatile = volatile_member; /* ?? */ + switch (fieldp->dgenfield.visibility) + { + case 1: + fn_p->field.fn_fields[ix].is_protected = 1; + fn_p->field.fn_fields[ix].is_private = 0; + break; + case 2: + fn_p->field.fn_fields[ix].is_protected = 0; + fn_p->field.fn_fields[ix].is_private = 1; + break; + default: /* public */ + fn_p->field.fn_fields[ix].is_protected = 0; + fn_p->field.fn_fields[ix].is_private = 0; + } + fn_p->field.fn_fields[ix].is_stub = 0; + + /* HP aCC emits both MEMFUNC and FUNCTION entries for a method; + if the class points to the FUNCTION, there is usually separate + code for the method; but if we have a MEMFUNC, the method has + been inlined (and there is usually no FUNCTION entry) + FIXME Not sure if this test is accurate. pai/1997-08-22 */ + if ((fn_fieldp->dblock.kind == DNTT_TYPE_MEMFUNC) || + (fn_fieldp->dblock.kind == DNTT_TYPE_DOC_MEMFUNC)) + fn_p->field.fn_fields[ix].is_inlined = 1; + else + fn_p->field.fn_fields[ix].is_inlined = 0; + + fn_p->field.fn_fields[ix].dummy = 0; + + /* Bump the total count of the member functions */ + n_fn_fields_total++; + + } + else if (fn_fieldp->dblock.kind == DNTT_TYPE_SVAR) + { + /* This case is for static data members of classes */ + + /* pai:: FIXME -- check that "staticmem" bit is set */ + + /* Get space to record this static member */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + list->field.name = VT (objfile) + fn_fieldp->dsvar.name; + SET_FIELD_PHYSNAME (list->field, 0); /* initialize to empty */ + memtype = hpread_type_lookup (fn_fieldp->dsvar.type, objfile); + + FIELD_TYPE (list->field) = memtype; + list->attributes = 0; + switch (fieldp->dgenfield.visibility) + { + case 1: + B_SET (&(list->attributes), ATTR_PROTECT); + break; + case 2: + B_SET (&(list->attributes), ATTR_PRIVATE); + break; + } + nfields++; + } + + else if (fn_fieldp->dblock.kind == DNTT_TYPE_FIELD) + { + /* FIELDs follow GENFIELDs for fields of anonymous unions. + Code below is replicated from the case for FIELDs further + below, except that fieldp is replaced by fn_fieldp */ + if (!fn_fieldp->dfield.a_union) + warning ("Debug info inconsistent: FIELD of anonymous union doesn't have a_union bit set"); + /* Get space to record the next field/data-member. */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + list->field.name = VT (objfile) + fn_fieldp->dfield.name; + FIELD_BITPOS (list->field) = fn_fieldp->dfield.bitoffset; + if (fn_fieldp->dfield.bitlength % 8) + list->field.bitsize = fn_fieldp->dfield.bitlength; + else + list->field.bitsize = 0; + + memtype = hpread_type_lookup (fn_fieldp->dfield.type, objfile); + list->field.type = memtype; + list->attributes = 0; + switch (fn_fieldp->dfield.visibility) + { + case 1: + B_SET (&(list->attributes), ATTR_PROTECT); + break; + case 2: + B_SET (&(list->attributes), ATTR_PRIVATE); + break; + } + nfields++; + } + else if (fn_fieldp->dblock.kind == DNTT_TYPE_SVAR) + { + /* Field of anonymous union; union is not inside a class */ + if (!fn_fieldp->dsvar.a_union) + warning ("Debug info inconsistent: SVAR field in anonymous union doesn't have a_union bit set"); + /* Get space to record the next field/data-member. */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + list->field.name = VT (objfile) + fn_fieldp->dsvar.name; + FIELD_BITPOS (list->field) = 0; /* FIXME is this always true? */ + FIELD_BITSIZE (list->field) = 0; /* use length from type */ + memtype = hpread_type_lookup (fn_fieldp->dsvar.type, objfile); + list->field.type = memtype; + list->attributes = 0; + /* No info to set visibility -- always public */ + nfields++; + } + else if (fn_fieldp->dblock.kind == DNTT_TYPE_DVAR) + { + /* Field of anonymous union; union is not inside a class */ + if (!fn_fieldp->ddvar.a_union) + warning ("Debug info inconsistent: DVAR field in anonymous union doesn't have a_union bit set"); + /* Get space to record the next field/data-member. */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + list->field.name = VT (objfile) + fn_fieldp->ddvar.name; + FIELD_BITPOS (list->field) = 0; /* FIXME is this always true? */ + FIELD_BITSIZE (list->field) = 0; /* use length from type */ + memtype = hpread_type_lookup (fn_fieldp->ddvar.type, objfile); + list->field.type = memtype; + list->attributes = 0; + /* No info to set visibility -- always public */ + nfields++; + } + else + { /* Not a method, nor a static data member, nor an anon union field */ + + /* This case is for miscellaneous type entries (local enums, + local function templates, etc.) that can be present + inside a class. */ + + /* Enums -- will be handled by other code that takes care + of DNTT_TYPE_ENUM; here we see only DNTT_TYPE_MEMENUM so + it's not clear we could have handled them here at all. */ + /* FUNC_TEMPLATE: is handled by other code (?). */ + /* MEMACCESS: modified access for inherited member. Not + sure what to do with this, ignoriing it at present. */ + + /* What other entries can appear following a GENFIELD which + we do not handle above? (MODIFIER, VFUNC handled above.) */ + + if ((fn_fieldp->dblock.kind != DNTT_TYPE_MEMACCESS) && + (fn_fieldp->dblock.kind != DNTT_TYPE_MEMENUM) && + (fn_fieldp->dblock.kind != DNTT_TYPE_FUNC_TEMPLATE)) + warning ("Internal error: Unexpected debug record kind %d found following DNTT_GENFIELD", + fn_fieldp->dblock.kind); + } + /* walk to the next FIELD or GENFIELD */ + field = fieldp->dgenfield.nextfield; + + } + else if (fieldp->dblock.kind == DNTT_TYPE_FIELD) + { + + /* Ordinary structure/union/class field */ + struct type *anon_union_type; + + /* Get space to record the next field/data-member. */ + new = (struct nextfield *) alloca (sizeof (struct nextfield)); + new->next = list; + list = new; + + list->field.name = VT (objfile) + fieldp->dfield.name; + + + /* A FIELD by itself (without a GENFIELD) can also be a static member */ + if (fieldp->dfield.staticmem) + { + FIELD_BITPOS (list->field) = -1; + FIELD_BITSIZE (list->field) = 0; + } + else + /* Non-static data member */ + { + FIELD_BITPOS (list->field) = fieldp->dfield.bitoffset; + if (fieldp->dfield.bitlength % 8) + FIELD_BITSIZE (list->field) = fieldp->dfield.bitlength; + else + FIELD_BITSIZE (list->field) = 0; + } + + memtype = hpread_type_lookup (fieldp->dfield.type, objfile); + FIELD_TYPE (list->field) = memtype; + list->attributes = 0; + switch (fieldp->dfield.visibility) + { + case 1: + B_SET (&(list->attributes), ATTR_PROTECT); + break; + case 2: + B_SET (&(list->attributes), ATTR_PRIVATE); + break; + } + nfields++; + + + /* Note 1: First, we have to check if the current field is an anonymous + union. If it is, then *its* fields are threaded along in the + nextfield chain. :-( This was supposed to help debuggers, but is + really just a nuisance since we deal with anonymous unions anyway by + checking that the name is null. So anyway, we skip over the fields + of the anonymous union. pai/1997-08-22 */ + /* Note 2: In addition, the bitoffsets for the fields of the anon union + are relative to the enclosing struct, *NOT* relative to the anon + union! This is an even bigger nuisance -- we have to go in and munge + the anon union's type information appropriately. pai/1997-08-22 */ + + /* Both tasks noted above are done by a separate function. This takes us + to the next FIELD or GENFIELD, skipping anon unions, and recursively + processing intermediate types. */ + field = hpread_get_next_skip_over_anon_unions (1, field, &fieldp, objfile); + + } + else + { + /* neither field nor genfield ?? is this possible?? */ + /* pai:: FIXME walk to the next -- how? */ + warning ("Internal error: unexpected DNTT kind %d encountered as field of struct", + fieldp->dblock.kind); + warning ("Skipping remaining fields of struct"); + break; /* get out of loop of fields */ + } + } + + /* If it's a template, read in the instantiation list */ + if (dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + ninstantiations = 0; + field = dn_bufp->dtemplate.expansions; + while (field.word && field.word != DNTTNIL) + { + fieldp = hpread_get_lntt (field.dnttp.index, objfile); + + /* The expansions or nextexp should point to a tagdef */ + if (fieldp->dblock.kind != DNTT_TYPE_TAGDEF) + break; + + i_new = (struct next_instantiation *) alloca (sizeof (struct next_instantiation)); + i_new->next = i_list; + i_list = i_new; + i_list->t = hpread_type_lookup (field, objfile); + ninstantiations++; + + /* And the "type" field of that should point to a class */ + field = fieldp->dtag.type; + fieldp = hpread_get_lntt (field.dnttp.index, objfile); + if (fieldp->dblock.kind != DNTT_TYPE_CLASS) + break; + + /* Get the next expansion */ + field = fieldp->dclass.nextexp; + } + } + TYPE_NINSTANTIATIONS (type) = ninstantiations; + if (ninstantiations > 0) + TYPE_INSTANTIATIONS (type) = (struct type **) + obstack_alloc (&objfile->type_obstack, sizeof (struct type *) * ninstantiations); + for (n = ninstantiations; i_list; i_list = i_list->next) + { + n -= 1; + TYPE_INSTANTIATION (type, n) = i_list->t; + } + + + /* Copy the field-list to GDB's symbol table */ + TYPE_NFIELDS (type) = nfields; + TYPE_N_BASECLASSES (type) = n_base_classes; + TYPE_FIELDS (type) = (struct field *) + obstack_alloc (&objfile->type_obstack, sizeof (struct field) * nfields); + /* Copy the saved-up fields into the field vector. */ + for (n = nfields, tmp_list = list; tmp_list; tmp_list = tmp_list->next) + { + n -= 1; + TYPE_FIELD (type, n) = tmp_list->field; + } + + /* Copy the "function-field-list" (i.e., the list of member + * functions in the class) to GDB's symbol table + */ + TYPE_NFN_FIELDS (type) = n_fn_fields; + TYPE_NFN_FIELDS_TOTAL (type) = n_fn_fields_total; + TYPE_FN_FIELDLISTS (type) = (struct fn_fieldlist *) + obstack_alloc (&objfile->type_obstack, sizeof (struct fn_fieldlist) * n_fn_fields); + for (n = n_fn_fields; fn_list; fn_list = fn_list->next) + { + n -= 1; + TYPE_FN_FIELDLIST (type, n) = fn_list->field; + } + + /* pai:: FIXME -- perhaps each bitvector should be created individually */ + for (n = nfields, tmp_list = list; tmp_list; tmp_list = tmp_list->next) + { + n -= 1; + if (tmp_list->attributes) + { + need_bitvectors = 1; + break; + } + } + + if (need_bitvectors) + { + /* pai:: this step probably redundant */ + ALLOCATE_CPLUS_STRUCT_TYPE (type); + + TYPE_FIELD_VIRTUAL_BITS (type) = + (B_TYPE *) TYPE_ALLOC (type, B_BYTES (nfields)); + B_CLRALL (TYPE_FIELD_VIRTUAL_BITS (type), nfields); + + TYPE_FIELD_PRIVATE_BITS (type) = + (B_TYPE *) TYPE_ALLOC (type, B_BYTES (nfields)); + B_CLRALL (TYPE_FIELD_PRIVATE_BITS (type), nfields); + + TYPE_FIELD_PROTECTED_BITS (type) = + (B_TYPE *) TYPE_ALLOC (type, B_BYTES (nfields)); + B_CLRALL (TYPE_FIELD_PROTECTED_BITS (type), nfields); + + /* this field vector isn't actually used with HP aCC */ + TYPE_FIELD_IGNORE_BITS (type) = + (B_TYPE *) TYPE_ALLOC (type, B_BYTES (nfields)); + B_CLRALL (TYPE_FIELD_IGNORE_BITS (type), nfields); + + while (nfields-- > 0) + { + if (B_TST (&(list->attributes), ATTR_VIRTUAL)) + SET_TYPE_FIELD_VIRTUAL (type, nfields); + if (B_TST (&(list->attributes), ATTR_PRIVATE)) + SET_TYPE_FIELD_PRIVATE (type, nfields); + if (B_TST (&(list->attributes), ATTR_PROTECT)) + SET_TYPE_FIELD_PROTECTED (type, nfields); + + list = list->next; + } + } + else + { + TYPE_FIELD_VIRTUAL_BITS (type) = NULL; + TYPE_FIELD_PROTECTED_BITS (type) = NULL; + TYPE_FIELD_PRIVATE_BITS (type) = NULL; + } + + if (has_vtable (type)) + { + /* Allocate space for class runtime information */ + TYPE_RUNTIME_PTR (type) = (struct runtime_info *) xmalloc (sizeof (struct runtime_info)); + /* Set flag for vtable */ + TYPE_VTABLE (type) = 1; + /* The first non-virtual base class with a vtable. */ + TYPE_PRIMARY_BASE (type) = primary_base_class (type); + /* The virtual base list. */ + TYPE_VIRTUAL_BASE_LIST (type) = virtual_base_list (type); + } + else + TYPE_RUNTIME_PTR (type) = NULL; + + /* If this is a local type (C++ - declared inside a function), record file name & line # */ + if (hpread_get_scope_depth (dn_bufp, objfile, 1 /* no need for real depth */ )) + { + TYPE_LOCALTYPE_PTR (type) = (struct local_type_info *) xmalloc (sizeof (struct local_type_info)); + TYPE_LOCALTYPE_FILE (type) = (char *) xmalloc (strlen (current_subfile->name) + 1); + strcpy (TYPE_LOCALTYPE_FILE (type), current_subfile->name); + if (current_subfile->line_vector && (current_subfile->line_vector->nitems > 0)) + TYPE_LOCALTYPE_LINE (type) = current_subfile->line_vector->item[current_subfile->line_vector->nitems - 1].line; + else + TYPE_LOCALTYPE_LINE (type) = 0; + } + else + TYPE_LOCALTYPE_PTR (type) = NULL; + + /* Clear the global saying what template we are in the middle of processing */ + current_template = NULL; + + return type; +} + +/* Adjust the physnames for each static member of a struct + or class type to be something like "A::x"; then various + other pieces of code that do a lookup_symbol on the phyname + work correctly. + TYPE is a pointer to the struct/class type + NAME is a char * (string) which is the class/struct name + Void return */ + +static void +fix_static_member_physnames (struct type *type, char *class_name, + struct objfile *objfile) +{ + int i; + + /* We fix the member names only for classes or structs */ + if (TYPE_CODE (type) != TYPE_CODE_STRUCT) + return; + + for (i = 0; i < TYPE_NFIELDS (type); i++) + if (TYPE_FIELD_STATIC (type, i)) + { + if (TYPE_FIELD_STATIC_PHYSNAME (type, i)) + return; /* physnames are already set */ + + SET_FIELD_PHYSNAME (TYPE_FIELDS (type)[i], + obstack_alloc (&objfile->type_obstack, + strlen (class_name) + strlen (TYPE_FIELD_NAME (type, i)) + 3)); + strcpy (TYPE_FIELD_STATIC_PHYSNAME (type, i), class_name); + strcat (TYPE_FIELD_STATIC_PHYSNAME (type, i), "::"); + strcat (TYPE_FIELD_STATIC_PHYSNAME (type, i), TYPE_FIELD_NAME (type, i)); + } +} + +/* Fix-up the type structure for a CLASS so that the type entry + * for a method (previously marked with a null type in hpread_read_struct_type() + * is set correctly to METHOD. + * OBJFILE is as for other such functions. + * Void return. */ + +static void +fixup_class_method_type (struct type *class, struct type *method, + struct objfile *objfile) +{ + int i, j, k; + + if (!class || !method || !objfile) + return; + + /* Only for types that have methods */ + if ((TYPE_CODE (class) != TYPE_CODE_CLASS) && + (TYPE_CODE (class) != TYPE_CODE_UNION)) + return; + + /* Loop over all methods and find the one marked with a NULL type */ + for (i = 0; i < TYPE_NFN_FIELDS (class); i++) + for (j = 0; j < TYPE_FN_FIELDLIST_LENGTH (class, i); j++) + if (TYPE_FN_FIELD_TYPE (TYPE_FN_FIELDLIST1 (class, i), j) == NULL) + { + /* Set the method type */ + TYPE_FN_FIELD_TYPE (TYPE_FN_FIELDLIST1 (class, i), j) = method; + + /* Break out of both loops -- only one method to fix up in a class */ + goto finish; + } + +finish: + TYPE_FLAGS (class) &= ~TYPE_FLAG_INCOMPLETE; +} + + +/* If we're in the middle of processing a template, get a pointer + * to the Nth template argument. + * An example may make this clearer: + * template <class T1, class T2> class q2 { + * public: + * T1 a; + * T2 b; + * }; + * The type for "a" will be "first template arg" and + * the type for "b" will be "second template arg". + * We need to look these up in order to fill in "a" and "b"'s type. + * This is called from hpread_type_lookup(). + */ +static struct type * +hpread_get_nth_template_arg (struct objfile *objfile, int n) +{ + if (current_template != NULL) + return TYPE_TEMPLATE_ARG (current_template, n).type; + else + return lookup_fundamental_type (objfile, FT_TEMPLATE_ARG); +} + +/* Read in and internalize a TEMPL_ARG (template arg) symbol. */ + +static struct type * +hpread_read_templ_arg_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile, char *name) +{ + struct type *type; + + /* See if it's something we've already deal with. */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_TEMPLATE_ARG) + return type; + + /* Nope. Fill in the appropriate fields. */ + TYPE_CODE (type) = TYPE_CODE_TEMPLATE_ARG; + TYPE_LENGTH (type) = 0; + TYPE_NFIELDS (type) = 0; + TYPE_NAME (type) = name; + return type; +} + +/* Read in and internalize a set debug symbol. */ + +static struct type * +hpread_read_set_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile) +{ + struct type *type; + + /* See if it's something we've already deal with. */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_SET) + return type; + + /* Nope. Fill in the appropriate fields. */ + TYPE_CODE (type) = TYPE_CODE_SET; + TYPE_LENGTH (type) = dn_bufp->dset.bitlength / 8; + TYPE_NFIELDS (type) = 0; + TYPE_TARGET_TYPE (type) = hpread_type_lookup (dn_bufp->dset.subtype, + objfile); + return type; +} + +/* Read in and internalize an array debug symbol. */ + +static struct type * +hpread_read_array_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile) +{ + struct type *type; + + /* Allocate an array type symbol. + * Why no check for already-read here, like in the other + * hpread_read_xxx_type routines? Because it kept us + * from properly determining the size of the array! + */ + type = hpread_alloc_type (hp_type, objfile); + + TYPE_CODE (type) = TYPE_CODE_ARRAY; + + /* Although the hp-symtab.h does not *require* this to be the case, + * GDB is assuming that "arrayisbytes" and "elemisbytes" be consistent. + * I.e., express both array-length and element-length in bits, + * or express both array-length and element-length in bytes. + */ + if (!((dn_bufp->darray.arrayisbytes && dn_bufp->darray.elemisbytes) || + (!dn_bufp->darray.arrayisbytes && !dn_bufp->darray.elemisbytes))) + { + warning ("error in hpread_array_type.\n"); + return NULL; + } + else if (dn_bufp->darray.arraylength == 0x7fffffff) + { + /* The HP debug format represents char foo[]; as an array with + * length 0x7fffffff. Internally GDB wants to represent this + * as an array of length zero. + */ + TYPE_LENGTH (type) = 0; + } + else if (dn_bufp->darray.arrayisbytes) + TYPE_LENGTH (type) = dn_bufp->darray.arraylength; + else /* arraylength is in bits */ + TYPE_LENGTH (type) = dn_bufp->darray.arraylength / 8; + + TYPE_TARGET_TYPE (type) = hpread_type_lookup (dn_bufp->darray.elemtype, + objfile); + + /* The one "field" is used to store the subscript type */ + /* Since C and C++ multi-dimensional arrays are simply represented + * as: array of array of ..., we only need one subscript-type + * per array. This subscript type is typically a subrange of integer. + * If this gets extended to support languages like Pascal, then + * we need to fix this to represent multi-dimensional arrays properly. + */ + TYPE_NFIELDS (type) = 1; + TYPE_FIELDS (type) = (struct field *) + obstack_alloc (&objfile->type_obstack, sizeof (struct field)); + TYPE_FIELD_TYPE (type, 0) = hpread_type_lookup (dn_bufp->darray.indextype, + objfile); + return type; +} + +/* Read in and internalize a subrange debug symbol. */ +static struct type * +hpread_read_subrange_type (dnttpointer hp_type, union dnttentry *dn_bufp, + struct objfile *objfile) +{ + struct type *type; + + /* Is it something we've already dealt with. */ + type = hpread_alloc_type (hp_type, objfile); + if (TYPE_CODE (type) == TYPE_CODE_RANGE) + return type; + + /* Nope, internalize it. */ + TYPE_CODE (type) = TYPE_CODE_RANGE; + TYPE_LENGTH (type) = dn_bufp->dsubr.bitlength / 8; + TYPE_NFIELDS (type) = 2; + TYPE_FIELDS (type) + = (struct field *) obstack_alloc (&objfile->type_obstack, + 2 * sizeof (struct field)); + + if (dn_bufp->dsubr.dyn_low) + TYPE_FIELD_BITPOS (type, 0) = 0; + else + TYPE_FIELD_BITPOS (type, 0) = dn_bufp->dsubr.lowbound; + + if (dn_bufp->dsubr.dyn_high) + TYPE_FIELD_BITPOS (type, 1) = -1; + else + TYPE_FIELD_BITPOS (type, 1) = dn_bufp->dsubr.highbound; + TYPE_TARGET_TYPE (type) = hpread_type_lookup (dn_bufp->dsubr.subtype, + objfile); + return type; +} + +/* struct type * hpread_type_lookup(hp_type, objfile) + * Arguments: + * hp_type: A pointer into the DNTT specifying what type we + * are about to "look up"., or else [for fundamental types + * like int, float, ...] an "immediate" structure describing + * the type. + * objfile: ? + * Return value: A pointer to a "struct type" (representation of a + * type in GDB's internal symbol table - see gdbtypes.h) + * Routine description: + * There are a variety of places when scanning the DNTT when we + * need to interpret a "type" field. The simplest and most basic + * example is when we're processing the symbol table record + * for a data symbol (a SVAR or DVAR record). That has + * a "type" field specifying the type of the data symbol. That + * "type" field is either an "immediate" type specification (for the + * fundamental types) or a DNTT pointer (for more complicated types). + * For the more complicated types, we may or may not have already + * processed the pointed-to type. (Multiple data symbols can of course + * share the same type). + * The job of hpread_type_lookup() is to process this "type" field. + * Most of the real work is done in subroutines. Here we interpret + * the immediate flag. If not immediate, chase the DNTT pointer to + * find our way to the SOM record describing the type, switch on + * the SOM kind, and then call an appropriate subroutine depending + * on what kind of type we are constructing. (e.g., an array type, + * a struct/class type, etc). + */ +static struct type * +hpread_type_lookup (dnttpointer hp_type, struct objfile *objfile) +{ + union dnttentry *dn_bufp; + struct type *tmp_type; + + /* First see if it's a simple builtin type. */ + if (hp_type.dntti.immediate) + { + /* If this is a template argument, the argument number is + * encoded in the bitlength. All other cases, just return + * GDB's representation of this fundamental type. + */ + if (hp_type.dntti.type == HP_TYPE_TEMPLATE_ARG) + return hpread_get_nth_template_arg (objfile, hp_type.dntti.bitlength); + else + return lookup_fundamental_type (objfile, + hpread_type_translate (hp_type)); + } + + /* Not a builtin type. We'll have to read it in. */ + if (hp_type.dnttp.index < LNTT_SYMCOUNT (objfile)) + dn_bufp = hpread_get_lntt (hp_type.dnttp.index, objfile); + else + /* This is a fancy way of returning NULL */ + return lookup_fundamental_type (objfile, FT_VOID); + + switch (dn_bufp->dblock.kind) + { + case DNTT_TYPE_SRCFILE: + case DNTT_TYPE_MODULE: + case DNTT_TYPE_ENTRY: + case DNTT_TYPE_BEGIN: + case DNTT_TYPE_END: + case DNTT_TYPE_IMPORT: + case DNTT_TYPE_LABEL: + case DNTT_TYPE_FPARAM: + case DNTT_TYPE_SVAR: + case DNTT_TYPE_DVAR: + case DNTT_TYPE_CONST: + case DNTT_TYPE_MEMENUM: + case DNTT_TYPE_VARIANT: + case DNTT_TYPE_FILE: + case DNTT_TYPE_WITH: + case DNTT_TYPE_COMMON: + case DNTT_TYPE_COBSTRUCT: + case DNTT_TYPE_XREF: + case DNTT_TYPE_SA: + case DNTT_TYPE_MACRO: + case DNTT_TYPE_BLOCKDATA: + case DNTT_TYPE_CLASS_SCOPE: + case DNTT_TYPE_MEMACCESS: + case DNTT_TYPE_INHERITANCE: + case DNTT_TYPE_OBJECT_ID: + case DNTT_TYPE_FRIEND_CLASS: + case DNTT_TYPE_FRIEND_FUNC: + /* These are not types - something went wrong. */ + /* This is a fancy way of returning NULL */ + return lookup_fundamental_type (objfile, FT_VOID); + + case DNTT_TYPE_FUNCTION: + /* We wind up here when dealing with class member functions + * (called from hpread_read_struct_type(), i.e. when processing + * the class definition itself). + */ + return hpread_read_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_DOC_FUNCTION: + return hpread_read_doc_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_TYPEDEF: + { + /* A typedef - chase it down by making a recursive call */ + struct type *structtype = hpread_type_lookup (dn_bufp->dtype.type, + objfile); + + /* The following came from the base hpread.c that we inherited. + * It is WRONG so I have commented it out. - RT + *... + + char *suffix; + suffix = VT (objfile) + dn_bufp->dtype.name; + TYPE_NAME (structtype) = suffix; + + * ... further explanation .... + * + * What we have here is a typedef pointing to a typedef. + * E.g., + * typedef int foo; + * typedef foo fum; + * + * What we desire to build is (these are pictures + * of "struct type"'s): + * + * +---------+ +----------+ +------------+ + * | typedef | | typedef | | fund. type | + * | type| -> | type| -> | | + * | "fum" | | "foo" | | "int" | + * +---------+ +----------+ +------------+ + * + * What this commented-out code is doing is smashing the + * name of pointed-to-type to be the same as the pointed-from + * type. So we wind up with something like: + * + * +---------+ +----------+ +------------+ + * | typedef | | typedef | | fund. type | + * | type| -> | type| -> | | + * | "fum" | | "fum" | | "fum" | + * +---------+ +----------+ +------------+ + * + */ + + return structtype; + } + + case DNTT_TYPE_TAGDEF: + { + /* Just a little different from above. We have to tack on + * an identifier of some kind (struct, union, enum, class, etc). + */ + struct type *structtype = hpread_type_lookup (dn_bufp->dtype.type, + objfile); + char *prefix, *suffix; + suffix = VT (objfile) + dn_bufp->dtype.name; + + /* Lookup the next type in the list. It should be a structure, + * union, class, enum, or template type. + * We will need to attach that to our name. + */ + if (dn_bufp->dtype.type.dnttp.index < LNTT_SYMCOUNT (objfile)) + dn_bufp = hpread_get_lntt (dn_bufp->dtype.type.dnttp.index, objfile); + else + { + complain (&hpread_type_lookup_complaint); + return NULL; + } + + if (dn_bufp->dblock.kind == DNTT_TYPE_STRUCT) + { + prefix = "struct "; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_UNION) + { + prefix = "union "; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS) + { + /* Further field for CLASS saying how it was really declared */ + /* 0==class, 1==union, 2==struct */ + if (dn_bufp->dclass.class_decl == 0) + prefix = "class "; + else if (dn_bufp->dclass.class_decl == 1) + prefix = "union "; + else if (dn_bufp->dclass.class_decl == 2) + prefix = "struct "; + else + prefix = ""; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_ENUM) + { + prefix = "enum "; + } + else if (dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + prefix = "template "; + } + else + { + prefix = ""; + } + + /* Build the correct name. */ + TYPE_NAME (structtype) + = (char *) obstack_alloc (&objfile->type_obstack, + strlen (prefix) + strlen (suffix) + 1); + TYPE_NAME (structtype) = strcpy (TYPE_NAME (structtype), prefix); + TYPE_NAME (structtype) = strcat (TYPE_NAME (structtype), suffix); + TYPE_TAG_NAME (structtype) = suffix; + + /* For classes/structs, we have to set the static member "physnames" + to point to strings like "Class::Member" */ + if (TYPE_CODE (structtype) == TYPE_CODE_STRUCT) + fix_static_member_physnames (structtype, suffix, objfile); + + return structtype; + } + + case DNTT_TYPE_POINTER: + /* Pointer type - call a routine in gdbtypes.c that constructs + * the appropriate GDB type. + */ + return make_pointer_type ( + hpread_type_lookup (dn_bufp->dptr.pointsto, + objfile), + NULL); + + case DNTT_TYPE_REFERENCE: + /* C++ reference type - call a routine in gdbtypes.c that constructs + * the appropriate GDB type. + */ + return make_reference_type ( + hpread_type_lookup (dn_bufp->dreference.pointsto, + objfile), + NULL); + + case DNTT_TYPE_ENUM: + return hpread_read_enum_type (hp_type, dn_bufp, objfile); + case DNTT_TYPE_SET: + return hpread_read_set_type (hp_type, dn_bufp, objfile); + case DNTT_TYPE_SUBRANGE: + return hpread_read_subrange_type (hp_type, dn_bufp, objfile); + case DNTT_TYPE_ARRAY: + return hpread_read_array_type (hp_type, dn_bufp, objfile); + case DNTT_TYPE_STRUCT: + case DNTT_TYPE_UNION: + return hpread_read_struct_type (hp_type, dn_bufp, objfile); + case DNTT_TYPE_FIELD: + return hpread_type_lookup (dn_bufp->dfield.type, objfile); + + case DNTT_TYPE_FUNCTYPE: + /* Here we want to read the function SOMs and return a + * type for it. We get here, for instance, when processing + * pointer-to-function type. + */ + return hpread_read_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_PTRMEM: + /* Declares a C++ pointer-to-data-member type. + * The "pointsto" field defines the class, + * while the "memtype" field defines the pointed-to-type. + */ + { + struct type *ptrmemtype; + struct type *class_type; + struct type *memtype; + memtype = hpread_type_lookup (dn_bufp->dptrmem.memtype, + objfile), + class_type = hpread_type_lookup (dn_bufp->dptrmem.pointsto, + objfile), + ptrmemtype = alloc_type (objfile); + smash_to_member_type (ptrmemtype, class_type, memtype); + return make_pointer_type (ptrmemtype, NULL); + } + break; + + case DNTT_TYPE_PTRMEMFUNC: + /* Defines a C++ pointer-to-function-member type. + * The "pointsto" field defines the class, + * while the "memtype" field defines the pointed-to-type. + */ + { + struct type *ptrmemtype; + struct type *class_type; + struct type *functype; + struct type *retvaltype; + int nargs; + int i; + class_type = hpread_type_lookup (dn_bufp->dptrmem.pointsto, + objfile); + functype = hpread_type_lookup (dn_bufp->dptrmem.memtype, + objfile); + retvaltype = TYPE_TARGET_TYPE (functype); + nargs = TYPE_NFIELDS (functype); + ptrmemtype = alloc_type (objfile); + + smash_to_method_type (ptrmemtype, class_type, retvaltype, + TYPE_FIELDS (functype), + TYPE_NFIELDS (functype), + 0); + return make_pointer_type (ptrmemtype, NULL); + } + break; + + case DNTT_TYPE_CLASS: + return hpread_read_struct_type (hp_type, dn_bufp, objfile); + + case DNTT_TYPE_GENFIELD: + /* Chase pointer from GENFIELD to FIELD, and make recursive + * call on that. + */ + return hpread_type_lookup (dn_bufp->dgenfield.field, objfile); + + case DNTT_TYPE_VFUNC: + /* C++ virtual function. + * We get here in the course of processing a class type which + * contains virtual functions. Just go through another level + * of indirection to get to the pointed-to function SOM. + */ + return hpread_type_lookup (dn_bufp->dvfunc.funcptr, objfile); + + case DNTT_TYPE_MODIFIER: + /* Check the modifiers and then just make a recursive call on + * the "type" pointed to by the modifier DNTT. + * + * pai:: FIXME -- do we ever want to handle "m_duplicate" and + * "m_void" modifiers? Is static_flag really needed here? + * (m_static used for methods of classes, elsewhere). + */ + tmp_type = make_cv_type (dn_bufp->dmodifier.m_const, + dn_bufp->dmodifier.m_volatile, + hpread_type_lookup (dn_bufp->dmodifier.type, objfile), + 0); + return tmp_type; + + + case DNTT_TYPE_MEMFUNC: + /* Member function. Treat like a function. + * I think we get here in the course of processing a + * pointer-to-member-function type... + */ + return hpread_read_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_DOC_MEMFUNC: + return hpread_read_doc_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_TEMPLATE: + /* Template - sort of the header for a template definition, + * which like a class, points to a member list and also points + * to a TEMPLATE_ARG list of type-arguments. + */ + return hpread_read_struct_type (hp_type, dn_bufp, objfile); + + case DNTT_TYPE_TEMPLATE_ARG: + { + char *name; + /* The TEMPLATE record points to an argument list of + * TEMPLATE_ARG records, each of which describes one + * of the type-arguments. + */ + name = VT (objfile) + dn_bufp->dtempl_arg.name; + return hpread_read_templ_arg_type (hp_type, dn_bufp, objfile, name); + } + + case DNTT_TYPE_FUNC_TEMPLATE: + /* We wind up here when processing a TEMPLATE type, + * if the template has member function(s). + * Treat it like a FUNCTION. + */ + return hpread_read_function_type (hp_type, dn_bufp, objfile, 0); + + case DNTT_TYPE_LINK: + /* The LINK record is used to link up templates with instantiations. + * There is no type associated with the LINK record per se. + */ + return lookup_fundamental_type (objfile, FT_VOID); + + /* Also not yet handled... */ + /* case DNTT_TYPE_DYN_ARRAY_DESC: */ + /* case DNTT_TYPE_DESC_SUBRANGE: */ + /* case DNTT_TYPE_BEGIN_EXT: */ + /* case DNTT_TYPE_INLN: */ + /* case DNTT_TYPE_INLN_LIST: */ + /* case DNTT_TYPE_ALIAS: */ + default: + /* A fancy way of returning NULL */ + return lookup_fundamental_type (objfile, FT_VOID); + } +} + +static sltpointer +hpread_record_lines (struct subfile *subfile, sltpointer s_idx, + sltpointer e_idx, struct objfile *objfile, + CORE_ADDR offset) +{ + union sltentry *sl_bufp; + + while (s_idx <= e_idx) + { + sl_bufp = hpread_get_slt (s_idx, objfile); + /* Only record "normal" entries in the SLT. */ + if (sl_bufp->snorm.sltdesc == SLT_NORMAL + || sl_bufp->snorm.sltdesc == SLT_EXIT) + record_line (subfile, sl_bufp->snorm.line, + sl_bufp->snorm.address + offset); + else if (sl_bufp->snorm.sltdesc == SLT_NORMAL_OFFSET) + record_line (subfile, sl_bufp->snormoff.line, + sl_bufp->snormoff.address + offset); + s_idx++; + } + return e_idx; +} + +/* Given a function "f" which is a member of a class, find + * the classname that it is a member of. Used to construct + * the name (e.g., "c::f") which GDB will put in the + * "demangled name" field of the function's symbol. + * Called from hpread_process_one_debug_symbol() + * If "f" is not a member function, return NULL. + */ +char * +class_of (struct type *functype) +{ + struct type *first_param_type; + char *first_param_name; + struct type *pointed_to_type; + char *class_name; + + /* Check that the function has a first argument "this", + * and that "this" is a pointer to a class. If not, + * functype is not a member function, so return NULL. + */ + if (TYPE_NFIELDS (functype) == 0) + return NULL; + first_param_name = TYPE_FIELD_NAME (functype, 0); + if (first_param_name == NULL) + return NULL; /* paranoia */ + if (strcmp (first_param_name, "this")) + return NULL; + first_param_type = TYPE_FIELD_TYPE (functype, 0); + if (first_param_type == NULL) + return NULL; /* paranoia */ + if (TYPE_CODE (first_param_type) != TYPE_CODE_PTR) + return NULL; + + /* Get the thing that "this" points to, check that + * it's a class, and get its class name. + */ + pointed_to_type = TYPE_TARGET_TYPE (first_param_type); + if (pointed_to_type == NULL) + return NULL; /* paranoia */ + if (TYPE_CODE (pointed_to_type) != TYPE_CODE_CLASS) + return NULL; + class_name = TYPE_NAME (pointed_to_type); + if (class_name == NULL) + return NULL; /* paranoia */ + + /* The class name may be of the form "class c", in which case + * we want to strip off the leading "class ". + */ + if (strncmp (class_name, "class ", 6) == 0) + class_name += 6; + + return class_name; +} + +/* Internalize one native debug symbol. + * Called in a loop from hpread_expand_symtab(). + * Arguments: + * dn_bufp: + * name: + * section_offsets: + * objfile: + * text_offset: + * text_size: + * filename: + * index: Index of this symbol + * at_module_boundary_p Pointer to boolean flag to control caller's loop. + */ + +static void +hpread_process_one_debug_symbol (union dnttentry *dn_bufp, char *name, + struct section_offsets *section_offsets, + struct objfile *objfile, CORE_ADDR text_offset, + int text_size, char *filename, int index, + int *at_module_boundary_p) +{ + unsigned long desc; + int type; + CORE_ADDR valu; + int offset = ANOFFSET (section_offsets, SECT_OFF_TEXT (objfile)); + int data_offset = ANOFFSET (section_offsets, SECT_OFF_DATA (objfile)); + union dnttentry *dn_temp; + dnttpointer hp_type; + struct symbol *sym; + struct context_stack *new; + char *class_scope_name; + + /* Allocate one GDB debug symbol and fill in some default values. */ + sym = (struct symbol *) obstack_alloc (&objfile->symbol_obstack, + sizeof (struct symbol)); + memset (sym, 0, sizeof (struct symbol)); + SYMBOL_NAME (sym) = obsavestring (name, strlen (name), &objfile->symbol_obstack); + SYMBOL_LANGUAGE (sym) = language_auto; + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + SYMBOL_LINE (sym) = 0; + SYMBOL_VALUE (sym) = 0; + SYMBOL_CLASS (sym) = LOC_TYPEDEF; + + /* Just a trick in case the SOM debug symbol is a type definition. + * There are routines that are set up to build a GDB type symbol, given + * a SOM dnttpointer. So we set up a dummy SOM dnttpointer "hp_type". + * This allows us to call those same routines. + */ + hp_type.dnttp.extension = 1; + hp_type.dnttp.immediate = 0; + hp_type.dnttp.global = 0; + hp_type.dnttp.index = index; + + /* This "type" is the type of SOM record. + * Switch on SOM type. + */ + type = dn_bufp->dblock.kind; + switch (type) + { + case DNTT_TYPE_SRCFILE: + /* This type of symbol indicates from which source file or + * include file any following data comes. It may indicate: + * + * o The start of an entirely new source file (and thus + * a new module) + * + * o The start of a different source file due to #include + * + * o The end of an include file and the return to the original + * file. Thus if "foo.c" includes "bar.h", we see first + * a SRCFILE for foo.c, then one for bar.h, and then one for + * foo.c again. + * + * If it indicates the start of a new module then we must + * finish the symbol table of the previous module + * (if any) and start accumulating a new symbol table. + */ + + valu = text_offset; + if (!last_source_file) + { + /* + * A note on "last_source_file": this is a char* pointing + * to the actual file name. "start_symtab" sets it, + * "end_symtab" clears it. + * + * So if "last_source_file" is NULL, then either this is + * the first record we are looking at, or a previous call + * to "end_symtab()" was made to close out the previous + * module. Since we're now quitting the scan loop when we + * see a MODULE END record, we should never get here, except + * in the case that we're not using the quick look-up tables + * and have to use the old system as a fall-back. + */ + start_symtab (name, NULL, valu); + record_debugformat ("HP"); + SL_INDEX (objfile) = dn_bufp->dsfile.address; + } + + else + { + /* Either a new include file, or a SRCFILE record + * saying we are back in the main source (or out of + * a nested include file) again. + */ + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dsfile.address, + objfile, offset); + } + + /* A note on "start_subfile". This routine will check + * the name we pass it and look for an existing subfile + * of that name. There's thus only one sub-file for the + * actual source (e.g. for "foo.c" in foo.c), despite the + * fact that we'll see lots of SRCFILE entries for foo.c + * inside foo.c. + */ + start_subfile (name, NULL); + break; + + case DNTT_TYPE_MODULE: + /* + * We no longer ignore DNTT_TYPE_MODULE symbols. The module + * represents the meaningful semantic structure of a compilation + * unit. We expect to start the psymtab-to-symtab expansion + * looking at a MODULE entry, and to end it at the corresponding + * END MODULE entry. + * + *--Begin outdated comments + * + * This record signifies the start of a new source module + * In C/C++ there is no explicit "module" construct in the language, + * but each compilation unit is implicitly a module and they + * do emit the DNTT_TYPE_MODULE records. + * The end of the module is marked by a matching DNTT_TYPE_END record. + * + * The reason GDB gets away with ignoring the DNTT_TYPE_MODULE record + * is it notices the DNTT_TYPE_END record for the previous + * module (see comments under DNTT_TYPE_END case), and then treats + * the next DNTT_TYPE_SRCFILE record as if it were the module-start record. + * (i.e., it makes a start_symtab() call). + * This scheme seems a little convoluted, but I'll leave it + * alone on the principle "if it ain't broke don't fix + * it". (RT). + * + *-- End outdated comments + */ + + valu = text_offset; + if (!last_source_file) + { + /* Start of a new module. We know this because "last_source_file" + * is NULL, which can only happen the first time or if we just + * made a call to end_symtab() to close out the previous module. + */ + start_symtab (name, NULL, valu); + SL_INDEX (objfile) = dn_bufp->dmodule.address; + } + else + { + /* This really shouldn't happen if we're using the quick + * look-up tables, as it would mean we'd scanned past an + * END MODULE entry. But if we're not using the tables, + * we started the module on the SRCFILE entry, so it's ok. + * For now, accept this. + */ + /* warning( "Error expanding psymtab, missed module end, found entry for %s", + * name ); + */ + *at_module_boundary_p = -1; + } + + start_subfile (name, NULL); + break; + + case DNTT_TYPE_FUNCTION: + case DNTT_TYPE_ENTRY: + /* A function or secondary entry point. */ + valu = dn_bufp->dfunc.lowaddr + offset; + + /* Record lines up to this point. */ + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dfunc.address, + objfile, offset); + + WITHIN_FUNCTION (objfile) = 1; + CURRENT_FUNCTION_VALUE (objfile) = valu; + + /* Stack must be empty now. */ + if (context_stack_depth != 0) + complain (&lbrac_unmatched_complaint, (char *) symnum); + new = push_context (0, valu); + + /* Built a type for the function. This includes processing + * the symbol records for the function parameters. + */ + SYMBOL_CLASS (sym) = LOC_BLOCK; + SYMBOL_TYPE (sym) = hpread_read_function_type (hp_type, dn_bufp, objfile, 1); + + /* The "SYMBOL_NAME" field is expected to be the mangled name + * (if any), which we get from the "alias" field of the SOM record + * if that exists. + */ + if ((dn_bufp->dfunc.language == HP_LANGUAGE_CPLUSPLUS) && + dn_bufp->dfunc.alias && /* has an alias */ + *(char *) (VT (objfile) + dn_bufp->dfunc.alias)) /* not a null string */ + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->dfunc.alias; + else + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->dfunc.name; + + /* Special hack to get around HP compilers' insistence on + * reporting "main" as "_MAIN_" for C/C++ */ + if ((strcmp (SYMBOL_NAME (sym), "_MAIN_") == 0) && + (strcmp (VT (objfile) + dn_bufp->dfunc.name, "main") == 0)) + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->dfunc.name; + + /* The SYMBOL_CPLUS_DEMANGLED_NAME field is expected to + * be the demangled name. + */ + if (dn_bufp->dfunc.language == HP_LANGUAGE_CPLUSPLUS) + { + /* SYMBOL_INIT_DEMANGLED_NAME is a macro which winds up + * calling the demangler in libiberty (cplus_demangle()) to + * do the job. This generally does the job, even though + * it's intended for the GNU compiler and not the aCC compiler + * Note that SYMBOL_INIT_DEMANGLED_NAME calls the + * demangler with arguments DMGL_PARAMS | DMGL_ANSI. + * Generally, we don't want params when we display + * a demangled name, but when I took out the DMGL_PARAMS, + * some things broke, so I'm leaving it in here, and + * working around the issue in stack.c. - RT + */ + SYMBOL_INIT_DEMANGLED_NAME (sym, &objfile->symbol_obstack); + if ((SYMBOL_NAME (sym) == VT (objfile) + dn_bufp->dfunc.alias) && + (!SYMBOL_CPLUS_DEMANGLED_NAME (sym))) + { + + /* Well, the symbol name is mangled, but the + * demangler in libiberty failed so the demangled + * field is still NULL. Try to + * do the job ourselves based on the "name" field + * in the SOM record. A complication here is that + * the name field contains only the function name + * (like "f"), whereas we want the class qualification + * (as in "c::f"). Try to reconstruct that. + */ + char *basename; + char *classname; + char *dem_name; + basename = VT (objfile) + dn_bufp->dfunc.name; + classname = class_of (SYMBOL_TYPE (sym)); + if (classname) + { + dem_name = xmalloc (strlen (basename) + strlen (classname) + 3); + strcpy (dem_name, classname); + strcat (dem_name, "::"); + strcat (dem_name, basename); + SYMBOL_CPLUS_DEMANGLED_NAME (sym) = dem_name; + SYMBOL_LANGUAGE (sym) = language_cplus; + } + } + } + + /* Add the function symbol to the list of symbols in this blockvector */ + if (dn_bufp->dfunc.global) + add_symbol_to_list (sym, &global_symbols); + else + add_symbol_to_list (sym, &file_symbols); + new->name = sym; + + /* Search forward to the next BEGIN and also read + * in the line info up to that point. + * Not sure why this is needed. + * In HP FORTRAN this code is harmful since there + * may not be a BEGIN after the FUNCTION. + * So I made it C/C++ specific. - RT + */ + if (dn_bufp->dfunc.language == HP_LANGUAGE_C || + dn_bufp->dfunc.language == HP_LANGUAGE_CPLUSPLUS) + { + while (dn_bufp->dblock.kind != DNTT_TYPE_BEGIN) + { + dn_bufp = hpread_get_lntt (++index, objfile); + if (dn_bufp->dblock.extension) + continue; + } + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dbegin.address, + objfile, offset); + SYMBOL_LINE (sym) = hpread_get_line (dn_bufp->dbegin.address, objfile); + } + record_line (current_subfile, SYMBOL_LINE (sym), valu); + break; + + case DNTT_TYPE_DOC_FUNCTION: + valu = dn_bufp->ddocfunc.lowaddr + offset; + + /* Record lines up to this point. */ + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->ddocfunc.address, + objfile, offset); + + WITHIN_FUNCTION (objfile) = 1; + CURRENT_FUNCTION_VALUE (objfile) = valu; + /* Stack must be empty now. */ + if (context_stack_depth != 0) + complain (&lbrac_unmatched_complaint, (char *) symnum); + new = push_context (0, valu); + + /* Built a type for the function. This includes processing + * the symbol records for the function parameters. + */ + SYMBOL_CLASS (sym) = LOC_BLOCK; + SYMBOL_TYPE (sym) = hpread_read_doc_function_type (hp_type, dn_bufp, objfile, 1); + + /* The "SYMBOL_NAME" field is expected to be the mangled name + * (if any), which we get from the "alias" field of the SOM record + * if that exists. + */ + if ((dn_bufp->ddocfunc.language == HP_LANGUAGE_CPLUSPLUS) && + dn_bufp->ddocfunc.alias && /* has an alias */ + *(char *) (VT (objfile) + dn_bufp->ddocfunc.alias)) /* not a null string */ + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->ddocfunc.alias; + else + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->ddocfunc.name; + + /* Special hack to get around HP compilers' insistence on + * reporting "main" as "_MAIN_" for C/C++ */ + if ((strcmp (SYMBOL_NAME (sym), "_MAIN_") == 0) && + (strcmp (VT (objfile) + dn_bufp->ddocfunc.name, "main") == 0)) + SYMBOL_NAME (sym) = VT (objfile) + dn_bufp->ddocfunc.name; + + if (dn_bufp->ddocfunc.language == HP_LANGUAGE_CPLUSPLUS) + { + + /* SYMBOL_INIT_DEMANGLED_NAME is a macro which winds up + * calling the demangler in libiberty (cplus_demangle()) to + * do the job. This generally does the job, even though + * it's intended for the GNU compiler and not the aCC compiler + * Note that SYMBOL_INIT_DEMANGLED_NAME calls the + * demangler with arguments DMGL_PARAMS | DMGL_ANSI. + * Generally, we don't want params when we display + * a demangled name, but when I took out the DMGL_PARAMS, + * some things broke, so I'm leaving it in here, and + * working around the issue in stack.c. - RT + */ + SYMBOL_INIT_DEMANGLED_NAME (sym, &objfile->symbol_obstack); + + if ((SYMBOL_NAME (sym) == VT (objfile) + dn_bufp->ddocfunc.alias) && + (!SYMBOL_CPLUS_DEMANGLED_NAME (sym))) + { + + /* Well, the symbol name is mangled, but the + * demangler in libiberty failed so the demangled + * field is still NULL. Try to + * do the job ourselves based on the "name" field + * in the SOM record. A complication here is that + * the name field contains only the function name + * (like "f"), whereas we want the class qualification + * (as in "c::f"). Try to reconstruct that. + */ + char *basename; + char *classname; + char *dem_name; + basename = VT (objfile) + dn_bufp->ddocfunc.name; + classname = class_of (SYMBOL_TYPE (sym)); + if (classname) + { + dem_name = xmalloc (strlen (basename) + strlen (classname) + 3); + strcpy (dem_name, classname); + strcat (dem_name, "::"); + strcat (dem_name, basename); + SYMBOL_CPLUS_DEMANGLED_NAME (sym) = dem_name; + SYMBOL_LANGUAGE (sym) = language_cplus; + } + } + } + + /* Add the function symbol to the list of symbols in this blockvector */ + if (dn_bufp->ddocfunc.global) + add_symbol_to_list (sym, &global_symbols); + else + add_symbol_to_list (sym, &file_symbols); + new->name = sym; + + /* Search forward to the next BEGIN and also read + * in the line info up to that point. + * Not sure why this is needed. + * In HP FORTRAN this code is harmful since there + * may not be a BEGIN after the FUNCTION. + * So I made it C/C++ specific. - RT + */ + if (dn_bufp->ddocfunc.language == HP_LANGUAGE_C || + dn_bufp->ddocfunc.language == HP_LANGUAGE_CPLUSPLUS) + { + while (dn_bufp->dblock.kind != DNTT_TYPE_BEGIN) + { + dn_bufp = hpread_get_lntt (++index, objfile); + if (dn_bufp->dblock.extension) + continue; + } + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dbegin.address, + objfile, offset); + SYMBOL_LINE (sym) = hpread_get_line (dn_bufp->dbegin.address, objfile); + } + record_line (current_subfile, SYMBOL_LINE (sym), valu); + break; + + case DNTT_TYPE_BEGIN: + /* Begin a new scope. */ + if (context_stack_depth == 1 /* this means we're at function level */ && + context_stack[0].name != NULL /* this means it's a function */ && + context_stack[0].depth == 0 /* this means it's the first BEGIN + we've seen after the FUNCTION */ + ) + { + /* This is the first BEGIN after a FUNCTION. + * We ignore this one, since HP compilers always insert + * at least one BEGIN, i.e. it's: + * + * FUNCTION + * argument symbols + * BEGIN + * local symbols + * (possibly nested BEGIN ... END's if there are inner { } blocks) + * END + * END + * + * By ignoring this first BEGIN, the local symbols get treated + * as belonging to the function scope, and "print func::local_sym" + * works (which is what we want). + */ + + /* All we do here is increase the depth count associated with + * the FUNCTION entry in the context stack. This ensures that + * the next BEGIN we see (if any), representing a real nested { } + * block, will get processed. + */ + + context_stack[0].depth++; + + } + else + { + + /* Record lines up to this SLT pointer. */ + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dbegin.address, + objfile, offset); + /* Calculate start address of new scope */ + valu = hpread_get_location (dn_bufp->dbegin.address, objfile); + valu += offset; /* Relocate for dynamic loading */ + /* We use the scope start DNTT index as nesting depth identifier! */ + desc = hpread_get_scope_start (dn_bufp->dbegin.address, objfile); + new = push_context (desc, valu); + } + break; + + case DNTT_TYPE_END: + /* End a scope. */ + + /* Valid end kinds are: + * MODULE + * FUNCTION + * WITH + * COMMON + * BEGIN + * CLASS_SCOPE + */ + + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dend.address, + objfile, offset); + switch (dn_bufp->dend.endkind) + { + case DNTT_TYPE_MODULE: + /* Ending a module ends the symbol table for that module. + * Calling end_symtab() has the side effect of clearing the + * last_source_file pointer, which in turn signals + * process_one_debug_symbol() to treat the next DNTT_TYPE_SRCFILE + * record as a module-begin. + */ + valu = text_offset + text_size + offset; + + /* Tell our caller that we're done with expanding the + * debug information for a module. + */ + *at_module_boundary_p = 1; + + /* Don't do this, as our caller will do it! + + * (void) end_symtab (valu, objfile, 0); + */ + break; + + case DNTT_TYPE_FUNCTION: + /* Ending a function, well, ends the function's scope. */ + dn_temp = hpread_get_lntt (dn_bufp->dend.beginscope.dnttp.index, + objfile); + valu = dn_temp->dfunc.hiaddr + offset; + /* Insert func params into local list */ + merge_symbol_lists (¶m_symbols, &local_symbols); + new = pop_context (); + /* Make a block for the local symbols within. */ + finish_block (new->name, &local_symbols, new->old_blocks, + new->start_addr, valu, objfile); + WITHIN_FUNCTION (objfile) = 0; /* This may have to change for Pascal */ + local_symbols = new->locals; + param_symbols = new->params; + break; + + case DNTT_TYPE_BEGIN: + if (context_stack_depth == 1 && + context_stack[0].name != NULL && + context_stack[0].depth == 1) + { + /* This is the END corresponding to the + * BEGIN which we ignored - see DNTT_TYPE_BEGIN case above. + */ + context_stack[0].depth--; + } + else + { + /* Ending a local scope. */ + valu = hpread_get_location (dn_bufp->dend.address, objfile); + /* Why in the hell is this needed? */ + valu += offset + 9; /* Relocate for dynamic loading */ + new = pop_context (); + desc = dn_bufp->dend.beginscope.dnttp.index; + if (desc != new->depth) + complain (&lbrac_mismatch_complaint, (char *) symnum); + + /* Make a block for the local symbols within. */ + finish_block (new->name, &local_symbols, new->old_blocks, + new->start_addr, valu, objfile); + local_symbols = new->locals; + param_symbols = new->params; + } + break; + + case DNTT_TYPE_WITH: + /* Since we ignore the DNTT_TYPE_WITH that starts the scope, + * we can ignore the DNTT_TYPE_END that ends it. + */ + break; + + case DNTT_TYPE_COMMON: + /* End a FORTRAN common block. We don't currently handle these */ + complain (&hpread_unhandled_end_common_complaint); + break; + + case DNTT_TYPE_CLASS_SCOPE: + + /* pai: FIXME Not handling nested classes for now -- must + * maintain a stack */ + class_scope_name = NULL; + +#if 0 + /* End a class scope */ + valu = hpread_get_location (dn_bufp->dend.address, objfile); + /* Why in the hell is this needed? */ + valu += offset + 9; /* Relocate for dynamic loading */ + new = pop_context (); + desc = dn_bufp->dend.beginscope.dnttp.index; + if (desc != new->depth) + complain (&lbrac_mismatch_complaint, (char *) symnum); + /* Make a block for the local symbols within. */ + finish_block (new->name, &local_symbols, new->old_blocks, + new->start_addr, valu, objfile); + local_symbols = new->locals; + param_symbols = new->params; +#endif + break; + + default: + complain (&hpread_unexpected_end_complaint); + break; + } + break; + + /* DNTT_TYPE_IMPORT is not handled */ + + case DNTT_TYPE_LABEL: + SYMBOL_NAMESPACE (sym) = LABEL_NAMESPACE; + break; + + case DNTT_TYPE_FPARAM: + /* Function parameters. */ + /* Note 1: This code was present in the 4.16 sources, and then + removed, because fparams are handled in + hpread_read_function_type(). However, while fparam symbols + are indeed handled twice, this code here cannot be removed + because then they don't get added to the local symbol list of + the function's code block, which leads to a failure to look + up locals, "this"-relative member names, etc. So I've put + this code back in. pai/1997-07-21 */ + /* Note 2: To fix a defect, we stopped adding FPARAMS to local_symbols + in hpread_read_function_type(), so FPARAMS had to be handled + here. I changed the location to be the appropriate argument + kinds rather than LOC_LOCAL. pai/1997-08-08 */ + /* Note 3: Well, the fix in Note 2 above broke argument printing + in traceback frames, and further it makes assumptions about the + order of the FPARAM entries from HP compilers (cc and aCC in particular + generate them in reverse orders -- fixing one breaks for the other). + So I've added code in hpread_read_function_type() to add fparams + to a param_symbols list for the current context level. These are + then merged into local_symbols when a function end is reached. + pai/1997-08-11 */ + + break; /* do nothing; handled in hpread_read_function_type() */ + +#if 0 /* Old code */ + if (dn_bufp->dfparam.regparam) + SYMBOL_CLASS (sym) = LOC_REGISTER; + else if (dn_bufp->dfparam.indirect) + SYMBOL_CLASS (sym) = LOC_REF_ARG; + else + SYMBOL_CLASS (sym) = LOC_ARG; + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + if (dn_bufp->dfparam.copyparam) + { + SYMBOL_VALUE (sym) = dn_bufp->dfparam.location; +#ifdef HPREAD_ADJUST_STACK_ADDRESS + SYMBOL_VALUE (sym) + += HPREAD_ADJUST_STACK_ADDRESS (CURRENT_FUNCTION_VALUE (objfile)); +#endif + } + else + SYMBOL_VALUE (sym) = dn_bufp->dfparam.location; + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->dfparam.type, objfile); + add_symbol_to_list (sym, &fparam_symbols); + break; +#endif + + case DNTT_TYPE_SVAR: + /* Static variables. */ + SYMBOL_CLASS (sym) = LOC_STATIC; + + /* Note: There is a case that arises with globals in shared + * libraries where we need to set the address to LOC_INDIRECT. + * This case is if you have a global "g" in one library, and + * it is referenced "extern <type> g;" in another library. + * If we're processing the symbols for the referencing library, + * we'll see a global "g", but in this case the address given + * in the symbol table contains a pointer to the real "g". + * We use the storage class LOC_INDIRECT to indicate this. RT + */ + if (is_in_import_list (SYMBOL_NAME (sym), objfile)) + SYMBOL_CLASS (sym) = LOC_INDIRECT; + + SYMBOL_VALUE_ADDRESS (sym) = dn_bufp->dsvar.location + data_offset; + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->dsvar.type, objfile); + + if (dn_bufp->dsvar.global) + add_symbol_to_list (sym, &global_symbols); + + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (sym, &local_symbols); + + else + add_symbol_to_list (sym, &file_symbols); + + if (dn_bufp->dsvar.thread_specific) + { + /* Thread-local variable. + */ + SYMBOL_CLASS (sym) = LOC_THREAD_LOCAL_STATIC; + SYMBOL_BASEREG (sym) = CR27_REGNUM; + + if (objfile->flags & OBJF_SHARED) + { + /* + * This variable is not only thread local but + * in a shared library. + * + * Alas, the shared lib structures are private + * to "somsolib.c". But C lets us point to one. + */ + struct so_list *so; + + if (objfile->obj_private == NULL) + error ("Internal error in reading shared library information."); + + so = ((obj_private_data_t *) (objfile->obj_private))->so_info; + if (so == NULL) + error ("Internal error in reading shared library information."); + + /* Thread-locals in shared libraries do NOT have the + * standard offset ("data_offset"), so we re-calculate + * where to look for this variable, using a call-back + * to interpret the private shared-library data. + */ + SYMBOL_VALUE_ADDRESS (sym) = dn_bufp->dsvar.location + + so_lib_thread_start_addr (so); + } + } + break; + + case DNTT_TYPE_DVAR: + /* Dynamic variables. */ + if (dn_bufp->ddvar.regvar) + SYMBOL_CLASS (sym) = LOC_REGISTER; + else + SYMBOL_CLASS (sym) = LOC_LOCAL; + + SYMBOL_VALUE (sym) = dn_bufp->ddvar.location; +#ifdef HPREAD_ADJUST_STACK_ADDRESS + SYMBOL_VALUE (sym) + += HPREAD_ADJUST_STACK_ADDRESS (CURRENT_FUNCTION_VALUE (objfile)); +#endif + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->ddvar.type, objfile); + if (dn_bufp->ddvar.global) + add_symbol_to_list (sym, &global_symbols); + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (sym, &local_symbols); + else + add_symbol_to_list (sym, &file_symbols); + break; + + case DNTT_TYPE_CONST: + /* A constant (pascal?). */ + SYMBOL_CLASS (sym) = LOC_CONST; + SYMBOL_VALUE (sym) = dn_bufp->dconst.location; + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->dconst.type, objfile); + if (dn_bufp->dconst.global) + add_symbol_to_list (sym, &global_symbols); + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (sym, &local_symbols); + else + add_symbol_to_list (sym, &file_symbols); + break; + + case DNTT_TYPE_TYPEDEF: + /* A typedef. We do want to process these, since a name is + * added to the namespace for the typedef'ed name. + */ + SYMBOL_NAMESPACE (sym) = VAR_NAMESPACE; + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->dtype.type, objfile); + if (dn_bufp->dtype.global) + add_symbol_to_list (sym, &global_symbols); + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (sym, &local_symbols); + else + add_symbol_to_list (sym, &file_symbols); + break; + + case DNTT_TYPE_TAGDEF: + { + int global = dn_bufp->dtag.global; + /* Structure, union, enum, template, or class tag definition */ + /* We do want to process these, since a name is + * added to the namespace for the tag name (and if C++ class, + * for the typename also). + */ + SYMBOL_NAMESPACE (sym) = STRUCT_NAMESPACE; + + /* The tag contains in its "type" field a pointer to the + * DNTT_TYPE_STRUCT, DNTT_TYPE_UNION, DNTT_TYPE_ENUM, + * DNTT_TYPE_CLASS or DNTT_TYPE_TEMPLATE + * record that actually defines the type. + */ + SYMBOL_TYPE (sym) = hpread_type_lookup (dn_bufp->dtype.type, objfile); + TYPE_NAME (sym->type) = SYMBOL_NAME (sym); + TYPE_TAG_NAME (sym->type) = SYMBOL_NAME (sym); + if (dn_bufp->dtag.global) + add_symbol_to_list (sym, &global_symbols); + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (sym, &local_symbols); + else + add_symbol_to_list (sym, &file_symbols); + + /* If this is a C++ class, then we additionally + * need to define a typedef for the + * class type. E.g., so that the name "c" becomes visible as + * a type name when the user says "class c { ... }". + * In order to figure this out, we need to chase down the "type" + * field to get to the DNTT_TYPE_CLASS record. + * + * We also add the typename for ENUM. Though this isn't + * strictly correct, it is necessary because of the debug info + * generated by the aCC compiler, in which we cannot + * distinguish between: + * enum e { ... }; + * and + * typedef enum { ... } e; + * I.e., the compiler emits the same debug info for the above + * two cases, in both cases "e" appearing as a tagdef. + * Therefore go ahead and generate the typename so that + * "ptype e" will work in the above cases. + * + * We also add the typename for TEMPLATE, so as to allow "ptype t" + * when "t" is a template name. + */ + if (dn_bufp->dtype.type.dnttp.index < LNTT_SYMCOUNT (objfile)) + dn_bufp = hpread_get_lntt (dn_bufp->dtag.type.dnttp.index, objfile); + else + { + complain (&hpread_tagdef_complaint); + return; + } + if (dn_bufp->dblock.kind == DNTT_TYPE_CLASS || + dn_bufp->dblock.kind == DNTT_TYPE_ENUM || + dn_bufp->dblock.kind == DNTT_TYPE_TEMPLATE) + { + struct symbol *newsym; + + newsym = (struct symbol *) obstack_alloc (&objfile->symbol_obstack, + sizeof (struct symbol)); + memset (newsym, 0, sizeof (struct symbol)); + SYMBOL_NAME (newsym) = name; + SYMBOL_LANGUAGE (newsym) = language_auto; + SYMBOL_NAMESPACE (newsym) = VAR_NAMESPACE; + SYMBOL_LINE (newsym) = 0; + SYMBOL_VALUE (newsym) = 0; + SYMBOL_CLASS (newsym) = LOC_TYPEDEF; + SYMBOL_TYPE (newsym) = sym->type; + if (global) + add_symbol_to_list (newsym, &global_symbols); + else if (WITHIN_FUNCTION (objfile)) + add_symbol_to_list (newsym, &local_symbols); + else + add_symbol_to_list (newsym, &file_symbols); + } + } + break; + + case DNTT_TYPE_POINTER: + /* Declares a pointer type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_ENUM: + /* Declares an enum type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_MEMENUM: + /* Member of enum */ + /* Ignored at this level, but hpread_read_enum_type() will take + * care of walking the list of enumeration members. + */ + break; + + case DNTT_TYPE_SET: + /* Declares a set type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_SUBRANGE: + /* Declares a subrange type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_ARRAY: + /* Declares an array type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_STRUCT: + case DNTT_TYPE_UNION: + /* Declares an struct/union type. + * Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_FIELD: + /* Structure/union/class field */ + /* Ignored at this level, but hpread_read_struct_type() will take + * care of walking the list of structure/union/class members. + */ + break; + + /* DNTT_TYPE_VARIANT is not handled by GDB */ + + /* DNTT_TYPE_FILE is not handled by GDB */ + + case DNTT_TYPE_FUNCTYPE: + /* Function type */ + /* Ignored at this level, handled within hpread_type_lookup() */ + break; + + case DNTT_TYPE_WITH: + /* This is emitted within methods to indicate "with <class>" + * scoping rules (i.e., indicate that the class data members + * are directly visible). + * However, since GDB already infers this by looking at the + * "this" argument, interpreting the DNTT_TYPE_WITH + * symbol record is unnecessary. + */ + break; + + case DNTT_TYPE_COMMON: + /* FORTRAN common. Not yet handled. */ + complain (&hpread_unhandled_common_complaint); + break; + + /* DNTT_TYPE_COBSTRUCT is not handled by GDB. */ + /* DNTT_TYPE_XREF is not handled by GDB. */ + /* DNTT_TYPE_SA is not handled by GDB. */ + /* DNTT_TYPE_MACRO is not handled by GDB */ + + case DNTT_TYPE_BLOCKDATA: + /* Not sure what this is - part of FORTRAN support maybe? + * Anyway, not yet handled. + */ + complain (&hpread_unhandled_blockdata_complaint); + break; + + case DNTT_TYPE_CLASS_SCOPE: + + + + /* The compiler brackets member functions with a CLASS_SCOPE/END + * pair of records, presumably to put them in a different scope + * from the module scope where they are normally defined. + * E.g., in the situation: + * void f() { ... } + * void c::f() { ...} + * The member function "c::f" will be bracketed by a CLASS_SCOPE/END. + * This causes "break f" at the module level to pick the + * the file-level function f(), not the member function + * (which needs to be referenced via "break c::f"). + * + * Here we record the class name to generate the demangled names of + * member functions later. + * + * FIXME Not being used now for anything -- cplus_demangle seems + * enough for getting the class-qualified names of functions. We + * may need this for handling nested classes and types. */ + + /* pai: FIXME Not handling nested classes for now -- need to + * maintain a stack */ + + dn_temp = hpread_get_lntt (dn_bufp->dclass_scope.type.dnttp.index, objfile); + if (dn_temp->dblock.kind == DNTT_TYPE_TAGDEF) + class_scope_name = VT (objfile) + dn_temp->dtag.name; + else + class_scope_name = NULL; + +#if 0 + + /* Begin a new scope. */ + SL_INDEX (objfile) = hpread_record_lines (current_subfile, + SL_INDEX (objfile), + dn_bufp->dclass_scope.address, + objfile, offset); + valu = hpread_get_location (dn_bufp->dclass_scope.address, objfile); + valu += offset; /* Relocate for dynamic loading */ + desc = hpread_get_scope_start (dn_bufp->dclass_scope.address, objfile); + /* We use the scope start DNTT index as the nesting depth identifier! */ + new = push_context (desc, valu); +#endif + break; + + case DNTT_TYPE_REFERENCE: + /* Declares a C++ reference type. Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_PTRMEM: + /* Declares a C++ pointer-to-data-member type. This does not + * need to be handled at this level; being a type description it + * is instead handled at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_PTRMEMFUNC: + /* Declares a C++ pointer-to-function-member type. This does not + * need to be handled at this level; being a type description it + * is instead handled at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_CLASS: + /* Declares a class type. + * Should not be necessary to do anything + * with the type at this level; these are processed + * at the hpread_type_lookup() level. + */ + break; + + case DNTT_TYPE_GENFIELD: + /* I believe this is used for class member functions */ + /* Ignored at this level, but hpread_read_struct_type() will take + * care of walking the list of class members. + */ + break; + + case DNTT_TYPE_VFUNC: + /* Virtual function */ + /* This does not have to be handled at this level; handled in + * the course of processing class symbols. + */ + break; + + case DNTT_TYPE_MEMACCESS: + /* DDE ignores this symbol table record. + * It has something to do with "modified access" to class members. + * I'll assume we can safely ignore it too. + */ + break; + + case DNTT_TYPE_INHERITANCE: + /* These don't have to be handled here, since they are handled + * within hpread_read_struct_type() in the process of constructing + * a class type. + */ + break; + + case DNTT_TYPE_FRIEND_CLASS: + case DNTT_TYPE_FRIEND_FUNC: + /* These can safely be ignored, as GDB doesn't need this + * info. DDE only uses it in "describe". We may later want + * to extend GDB's "ptype" to give this info, but for now + * it seems safe enough to ignore it. + */ + break; + + case DNTT_TYPE_MODIFIER: + /* Intended to supply "modified access" to a type */ + /* From the way DDE handles this, it looks like it always + * modifies a type. Therefore it is safe to ignore it at this + * level, and handle it in hpread_type_lookup(). + */ + break; + + case DNTT_TYPE_OBJECT_ID: + /* Just ignore this - that's all DDE does */ + break; + + case DNTT_TYPE_MEMFUNC: + /* Member function */ + /* This does not have to be handled at this level; handled in + * the course of processing class symbols. + */ + break; + + case DNTT_TYPE_DOC_MEMFUNC: + /* Member function */ + /* This does not have to be handled at this level; handled in + * the course of processing class symbols. + */ + break; + + case DNTT_TYPE_TEMPLATE: + /* Template - sort of the header for a template definition, + * which like a class, points to a member list and also points + * to a TEMPLATE_ARG list of type-arguments. + * We do not need to process TEMPLATE records at this level though. + */ + break; + + case DNTT_TYPE_TEMPLATE_ARG: + /* The TEMPLATE record points to an argument list of + * TEMPLATE_ARG records, each of which describes one + * of the type-arguments. + * We do not need to process TEMPLATE_ARG records at this level though. + */ + break; + + case DNTT_TYPE_FUNC_TEMPLATE: + /* This will get emitted for member functions of templates. + * But we don't need to process this record at this level though, + * we will process it in the course of processing a TEMPLATE + * record. + */ + break; + + case DNTT_TYPE_LINK: + /* The LINK record is used to link up templates with instantiations. */ + /* It is not clear why this is needed, and furthermore aCC does + * not appear to generate this, so I think we can safely ignore it. - RT + */ + break; + + /* DNTT_TYPE_DYN_ARRAY_DESC is not handled by GDB */ + /* DNTT_TYPE_DESC_SUBRANGE is not handled by GDB */ + /* DNTT_TYPE_BEGIN_EXT is not handled by GDB */ + /* DNTT_TYPE_INLN is not handled by GDB */ + /* DNTT_TYPE_INLN_LIST is not handled by GDB */ + /* DNTT_TYPE_ALIAS is not handled by GDB */ + + default: + break; + } +} + +/* Get nesting depth for a DNTT entry. + * DN_BUFP points to a DNTT entry. + * OBJFILE is the object file. + * REPORT_NESTED is a flag; if 0, real nesting depth is + * reported, if it is 1, the function simply returns a + * non-zero value if the nesting depth is anything > 0. + * + * Return value is an integer. 0 => not a local type / name + * positive return => type or name is local to some + * block or function. + */ + + +/* elz: ATTENTION: FIXME: NOTE: WARNING!!!! + this function now returns 0 right away. It was taking too much time + at start up. Now, though, the local types are not handled correctly. + */ + + +static int +hpread_get_scope_depth (union dnttentry *dn_bufp, struct objfile *objfile, + int report_nested) +{ + register int index; + register union dnttentry *dn_tmp; + register short depth = 0; +/****************************/ + return 0; +/****************************/ + + index = (((char *) dn_bufp) - LNTT (objfile)) / (sizeof (struct dntt_type_block)); + + while (--index >= 0) + { + dn_tmp = hpread_get_lntt (index, objfile); + switch (dn_tmp->dblock.kind) + { + case DNTT_TYPE_MODULE: + return depth; + case DNTT_TYPE_END: + /* index is signed int; dnttp.index is 29-bit unsigned int! */ + index = (int) dn_tmp->dend.beginscope.dnttp.index; + break; + case DNTT_TYPE_BEGIN: + case DNTT_TYPE_FUNCTION: + case DNTT_TYPE_DOC_FUNCTION: + case DNTT_TYPE_WITH: + case DNTT_TYPE_COMMON: + case DNTT_TYPE_CLASS_SCOPE: + depth++; + if (report_nested) + return 1; + break; + default: + break; + } + } + return depth; +} + +/* Adjust the bitoffsets for all fields of an anonymous union of + type TYPE by negative BITS. This handles HP aCC's hideous habit + of giving members of anonymous unions bit offsets relative to the + enclosing structure instead of relative to the union itself. */ + +static void +hpread_adjust_bitoffsets (struct type *type, int bits) +{ + register int i; + + /* This is done only for unions; caller had better check that + it is an anonymous one. */ + if (TYPE_CODE (type) != TYPE_CODE_UNION) + return; + + /* Adjust each field; since this is a union, there are no base + classes. Also no static membes. Also, no need for recursion as + the members of this union if themeselves structs or unions, have + the correct bitoffsets; if an anonymous union is a member of this + anonymous union, the code in hpread_read_struct_type() will + adjust for that. */ + + for (i = 0; i < TYPE_NFIELDS (type); i++) + TYPE_FIELD_BITPOS (type, i) -= bits; +} + +/* Because of quirks in HP compilers' treatment of anonymous unions inside + classes, we have to chase through a chain of threaded FIELD entries. + If we encounter an anonymous union in the chain, we must recursively skip over + that too. + + This function does a "next" in the chain of FIELD entries, but transparently + skips over anonymous unions' fields (recursively). + + Inputs are the number of times to do "next" at the top level, the dnttpointer + (FIELD) and entry pointer (FIELDP) for the dntt record corresponding to it, + and the ubiquitous objfile parameter. (Note: FIELDP is a **.) Return value + is a dnttpointer for the new field after all the skipped ones */ + +static dnttpointer +hpread_get_next_skip_over_anon_unions (int skip_fields, dnttpointer field, + union dnttentry **fieldp, + struct objfile *objfile) +{ + struct type *anon_type; + register int i; + int bitoffset; + char *name; + + for (i = 0; i < skip_fields; i++) + { + /* Get type of item we're looking at now; recursively processes the types + of these intermediate items we skip over, so they aren't lost. */ + anon_type = hpread_type_lookup ((*fieldp)->dfield.type, objfile); + anon_type = CHECK_TYPEDEF (anon_type); + bitoffset = (*fieldp)->dfield.bitoffset; + name = VT (objfile) + (*fieldp)->dfield.name; + /* First skip over one item to avoid stack death on recursion */ + field = (*fieldp)->dfield.nextfield; + *fieldp = hpread_get_lntt (field.dnttp.index, objfile); + /* Do we have another anonymous union? If so, adjust the bitoffsets + of its members and skip over its members. */ + if ((TYPE_CODE (anon_type) == TYPE_CODE_UNION) && + (!name || STREQ (name, ""))) + { + hpread_adjust_bitoffsets (anon_type, bitoffset); + field = hpread_get_next_skip_over_anon_unions (TYPE_NFIELDS (anon_type), field, fieldp, objfile); + } + } + return field; +} |