path: root/gas/ChangeLog

Sat Jan  4 16:34:16 1992  John Gilmore  (gnu at cygnus.com)

	* as.h:  Move redefine of xfree.
	* read.c:  Use xmalloc, not malloc.
	* expr.c, input-scrub.c, write.c:  Avoid "know" macros
	that cover more than one line, since this causes char strings
	with embedded newlines.	

Wed Dec 11 18:13:07 1991  Steve Chamberlain  (sac at rtl.cygnus.com)
	* configure.in: handle rename of obj-coffbfd
	* config/obj-coffbfd.c: lint, don't fixup relocs if H8300, use
	tc_reloc_mangle to prepare relocation if one available
	* config/tc-h8300.h: added tc_reloc_mangle routine

Tue Dec 10 04:07:28 1991  K. Richard Pixley  (rich at rtl.cygnus.com)

	* input-file.c: don't close a NULL file pointer.

	* Makefile.in: infodir belongs in datadir.

Sat Dec  7 17:13:42 1991  K. Richard Pixley  (rich at rtl.cygnus.com)

	* Makefile.in: fix fake-as target to also create as.new.

Fri Dec  6 23:16:43 1991  K. Richard Pixley  (rich at rtl.cygnus.com)

	* Makefile.in: install using INSTALL_DATA and INSTALL_PROGRAM.
	  correct include directory locations. added standards.text
	  support.

	* configure.in: mark this directory as target dependent.

Thu Dec  5 22:46:23 1991  K. Richard Pixley  (rich at rtl.cygnus.com)

	* Makefile.in: idestdir and ddestdir go away.  Added
	  copyrights and shift gpl to v2.  Added ChangeLog if it
	  didn't exist. docdir and mandir now keyed off datadir by
	  default.

Wed Dec  4 10:28:48 1991  Steve Chamberlain  (sac at cygnus.com)

	* config/obj-coff-bfd.c (crawl_symbol_chain): only move a
	symbol to the end of a file if it really is external.


Sat Nov 30 23:03:59 1991  Steve Chamberlain  (sac at rtl.cygnus.com)

	* as.h: reloc.h->aout/reloc.h.
	* a.out.gnu.h: reloc.h->aout/reloc.h

Thu Nov 28 13:21:41 1991  Steve Chamberlain  (sac at rtl.cygnus.com)

	* config/obj-coff-bfd: made multiple sections with subsections
	work.
	* config/tc-h8300.c: made $ statement sep, and int alloc 2 bytes.


Mon Nov 25 17:26:22 1991  Steve Chamberlain  (sac at cygnus.com)

	* as.c: (perform_an_assembly_pass): moved declaration of i to
	avoid a syntax error. Now always create the three default
	sections when in MANY_SEGMENTS mode.
	* as.h: New improved way of specifying MANY_SEGMENTS
	* configure.in: hds object format is now coff-bfd
	* subsegs.h: add line numbers and section stuff to the
	segment_info structure. (But go back later and fix.. this is
	COFF specific)
	

Thu Oct 24 17:25:59 1991  Steve Chamberlain  (steve at rtl.cygnus.com)

	* config/obj-coff.c (obj_coff_dim): Allows newline to end a
	.dim statement.

Mon Oct 21 09:25:33 1991  Steve Chamberlain  (steve at rtl.cygnus.com)

	* configure.in: added ebmon entry.
	* output-file.c (output_file_create, output_file_close) added
	stuff to use bfd.
	* write.c: more bfd support
	* config/obj-coff.c  config/obj-coff.h use bfd for swapping
	header info.
	* tc-a29k: added the reloc types we support now
	

Mon Oct 14 11:44:03 1991  Roland H. Pesch  (pesch at cygnus.com)

	* doc/Makefile (as.ms): one more sed workaround for texi2roff
	(as.mm, as.me): new targets

Fri Oct 11 18:18:37 1991  Roland H. Pesch  (pesch at cygnus.com)

	* doc/Makefile: new targets as.info, as.ms
	doc/as.texinfo: minor changes for texi2roff-2
	Makefile.in: revise 'doc' target.	

Fri Jan  4 12:48:22 EST 1991	Jay Fenlason (hack@ai.mit.edu)

	* messages.c  Moved as_perror from input-scrub.c  Modified the
	error messages to look better.

	* output-file.c Don't call as_fatal, just call exit()

	expr.c Slightly improve checking for foo-foo case in
	clean_up_expression().  Detect foo: bar: ... foo-bar...

Tue Dec  4 16:29:20 EST 1990	Jay Fenlason (hack@ai.mit.edu)

	* m68k.c  Fixed an obscure bug involving AOFF mode with a
	large constant displacement (Was forgetting to output the
	extension word.)

	make-gas.com  Added a three line patch from Eric Youngdale that
	makes it possible to submit make-gas.com to a batch queue.

Wed Nov 21 15:07:51 EST 1990	Jay Fenlason (hack@ai.mit.edu)

	* vms.c  (VMS_TBT_Routine_END)  Add a four line patch from
	Eric Youngdale.

Tue Nov 13 14:02:15 EST 1990	Jay Fenlason (hack@ai.mti.edu)

	* vms-dbg.c (VMS_DBG_record())  Another one character patch from
	Eric Youngdale.

Mon Oct 29 15:49:21 EST 1990	Jay Fenlason (hack@ai.mit.edu)

	* read.c  Replace some as_warn calls with as_bad.

Fri Oct 26 15:21:15 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* i386.c, i860.c, ns32k.c	Add const changes.

Mon Oct 22 14:04:26 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* sparc.c	Add const changes.

	* make-gas.com	define const= for VMS, since many older versions of
	GCC don't work correctly with const under VMS.

Thu Oct 18 12:44:11 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* i860.c i860-opcode.h  Added patches from rgb@mcc.com

	* read.c, symbols.c, vms.c, + new_file vms-dbg-module.c
	Added Eric Youngdale's <YOUNGDALE@v6550c.nrl.navy.mil> VMS debugging
	patches, so debugging GCC output now works.

	* hash.c (hash_grow)  Remember to blank out the wall entry in the new
	hash table.  This is important on systems where malloc() returns
	non-zero storage. . .

Tue Oct 16 10:11:35 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* output-file.c (output_file_create)  if output filename is given as
	'-', write to stdout.

	* m68k.c Finally get the PCREL code to work right.  Add relaxation of
	PCREL stuff  This small fix from Ken Woodland
	(kenny%datacube.uucp@uunet.uu.net).

	* m68k.c  Added some const declarations to constants.  (md_relax_table,
	md_pseudo_table, etc. . .)

Thu Oct 11 11:15:10 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* Makefile, read.c, write.c Include the i860 port.
	(New files i860.c i860-opcode.h i860.h)

	* m68k.c Fix some addressing modes, (AOFF, AINDEX, etc) to work in
	PC relative mode.

	* (all over)  Raeburn's const hacking.  This reduces the data-space size by
	declaring many tables, etc, as 'const'.

Mon Oct 22 22:48:22 1990  John Gilmore  (gnu at cygint)

	Make gas work if you turn on the know() checks.

	* app.c:  Only pass a single space through:  the one after
	the opcode.  All other whitespace is removed, to match the
	expectations of the parser in read.c.

	* as.h: Remove obsolete comments.  Remove JF's NDEBUG so
	that know() can really work if you turn it on.  Make
	SEG_MAXIMUM_ORDINAL == SEG_REGISTER.

	* expr.c (operand):  Change BITS_PER_INT to 8*sizeof(int).

	* input-scrub.c:  strlen("\0") doesn't return 1...
	(as_where):  Add space after line number in errors, like gcc.

	* m68k.c (s_bss):  Fake .bss into data section 255.
	We can't cope with a real "BSS section" yet, but we can at
	least do the right thing less efficiently (with lots of
	zeroes).

	* read.c:  Turn lots of as_warn()'s into as_bad()'s.

	* read.h (SKIP_WHITESPACE):  Replace last instance of ASSERT()
	with know().

	* sparc.c (s_seg):  We can't put frags into the BSS segment
	yet, so just fake bss seg as 255th data segment.

	* vax.c:  Remove \'s from continued macro *parameters*.  These
	must have been added after the last time someone turned on
	know() checking...

	* write.c (relax_segment):  Refine what we know() about the
	symbols referenced during relaxation.	

	* Makefile (OTHER_ALIGN):  Remove, handled in tables now.
	Flip options a bit.  These options really ought to go
	elsewhere.

Sun Oct 21 03:57:21 1990  John Gilmore  (gnu at cygint)

	Sun-3 fixes.

	* expr.c, write.c:  Missing semicolon after know().

	* write.c (fixup_segment):  Allow pc-relative accesses to undefined
	external symbols.  Previously this would turn off pc-rel calc
	of displacement, while leaving pc-rel opcode alone, botching.

 	* m68k.c (m68k_ip):  Allow pc-relative effective addresses
	for source operands.  "pea" instructions are a good example
	where we can shorten from abs long to pc+16bit.
	(md_convert_frag):  Fix "JBSR" comments to refer to "JSR", the
	actual instruction.  Insert comments about bug in 68000 bcc
	and dbcc long code (that doesn't get exercised much).  Add
	comments about long pcrel displacements for -pic.  Remove
	"this code has not been tested" comment.
	(md_estimate_size_before_relax):  Now that fixup_segment
	doesn't shortcircuit pc-relative fixups for undefined symbols,
	only output them if -pic; else turn them absolute, which is
	slightly faster.  More JBSR->JSR comments.
	(md_parse_options):  Parse -pic.

Fri Oct 19 14:35:20 1990  John Gilmore  (gnu at cygint)

	* Make sparc assembler more compatible with Sun assembler.
	sparc.c:  reformat pseudo-op table to match main table.
	(md_assemble):  Add SPECIAL_CASE_FDIV to assemble FDIV*
	instructions as fdiv followed by fmovs to get around chip bug.
	(s_common, s_seg):  Accept "bss" section name.
	(md_assemble):  Handle "set" instructions with absolute
	operands, that only take one instruction rather than two.
	sparc-opcode.h (fdiv*):  Mark instructions "S"pecial.
	subsegs.c (subseg_change):  Move tail pointer too.
	symbols.c (colon):  Allow new definitions to override .comm symbols,
	as in VMS.  Sun CC depends on this.
	write.c (new_fix):  Always take r_type argument, not just on sparc.
	Chain fixP's in order, using tail pointer, so relocation
	records come out in forward order like Sun as.  Remove SPARC
	ifdefs.
	write.h: Add seg_fix_tailP, data_fix_tailP, text_fix_tailP.

	* am29k.c: Use s_align_bytes rather than a local copy.

	* read.c (s_align):  Rather than ifdef it, make two functions,
	s_align_bytes and s_align_ptwo.  Individual pseudo-op tables
	can call whichever one they like.

	* write.c (append):  Move from append.c to here.
	append.c:  Remove file.	

	* Makefile (MDSRC, mdsrc):  Easy way to edit all md.c's.
	Fix options.  Add option for -DDEBUG for know() and assert().
	Remove append.c, am29k.h.  Don't build special read-sparc.o.
	Remove sparc.h.  "make clean" removes am29k .o's.  Add
	dependencies on reloc.h.

Thu Oct 18 17:56:34 1990  John Gilmore  (gnu at cygint)

	* Generalize sparc extensions to relocation info.  Gas now
	keeps relocation information internally in a different format
	than how it is stored in the resulting .o.  md_ri_to_bytes()
	converts to external format.  md_reloc_size says how large
	each relocation record is in external format.
	sparc.h:  Remove this file.  Rename to reloc.h.  Rename struct
	to reloc_info_generic.
	reloc.h:  Add relocation types for AMD 29000.
	read.c, write.c:  Always call fix_new with reloc-type argument.
	write.c (emit_relocations): Make md_ri_to_bytes write directly
	to output area rather than overwriting its argument then
	bcopying it.
	md.h:  Declare md_reloc_size and md_ri_to_bytes.
	i386.c, am29k.c, vax.c, ms32k.c, m68k.c, sparc.c: include reloc.h.
	(md_reloc_size): Specify correct value.
	(md_ri_to_bytes):  Convert format from internal to external.

	write.c (write_object_file):  Call md_section_align() which
	rounds section sizes up if desired. 
	sparc.c (md_section_align):  Round to 8 byte boundary.
	i386.c, am29k.c, vax.c, ns32k.c, m68k.c (md_section_align):  Nop.

Mon Oct 15 22:06:11 1990  John Gilmore  (gnu at cygint)

	Changes in support of the AMD 29000 version of gas.

	* am29k-opcode.h:  Add dummy entry to end so we can examine
	item N+1 without exceeding table.

	* am29k.h:  New include file, derived from sparc.h.  Kludged
	together, still needs major work to get relocation working.

	* am29k.c:  New file, derived from sparc.c.
	Put 29k-specific ASM29K pseudo-ops into table.
	Change comment_chars to ASM29K.
	Change s_seg to s_use.
	Change s_sparc_align to s_29k_align, default operand to 4.
	(define_some_regs):  Define special register names.
	(md_begin):  Preprocess opcode table to mash together all
	the variants of single opcodes.  This simplifies later handling.
	Call define_some_regs to preset special reg names.
	(parse_operand):  Add, parses out an operand from a stmt.
	(machine_ip):  Simplify, since 29K is simpler asm language.
	Handle the various keyletters in the opcode table.


	Handle include files in the assembler, with a .include
	pseudo-op.
	* as.h (input_scrub_include_file):  declare.
	* as.c (perform_an_assembly_pass):  Avoid buffer hacking.
	Start us off in text segment.
	* read.c (read_a_source_file):  Take a name as argument,
	internalize all buffer handling.  Don't start a new text
	subsegment on each entry.  Actually use the start and end
	pointers returned by input_scrub_next_buffer.
	(s_include):  Call input_scrub_include_file for .include.
	* input-scrub.c:  Fix comments.
	(struct input_save):  Add, for saving state at .include.
	(input_scrub_push, input_scrub_pop):  Add, push & pop state.
	(input_scrub_begin):  Initialize next_saved_file.
	(input_scrub_end):  Free buffer.
	(input_scrub_include_file):  Add, to include a file.
	(input_scrub_close):  Add, to close a file.
	(input_scrub_next_buffer):  Set buffer-start address for
	caller.  If we hit EOF and were included, pop to previous file.
	* input-file.c:  Remove old includes.  Remove old file-descriptor
	hacking code, that was commented out.
	(struct saved_file):  Add, for saving state at .include.
	(input_file_push, input_file_pop):  Add, push & pop state.
	(input_file_open):  Don't buffer all files in one place.
	(input_file_close):  Add, close input file.
	* input-file.h: Declare new functions.
	* app.c: (struct app_save):  Add, for saving state at .include.
	(app_push, app_pop):  Add, push and pop state.
	(do_scrub_next_char):  Move its static state variables out so
	they can be saved and restored.


	* app.c:  Allow overriding of character meanings by machine
	dependent strings.  Avoid wiring character constants into app.c.
	(do_scrub_begin):  New meanings override old ones, not "OR" them.
	Only handle /* comments if / is not already in use.
	(do_scrub_next_char):  Reorganize mass of nested if's
	into a switch for speed.  Don't assume ';' is line terminator.
	Reorganize switch on characters, into a switch on their (machine-
	dependent) lexer table meanings.
	

	Encapsulate knowledge of segment types in fewer places.
	This allows us to add the "SEG_REGISTER" type, as well as
	providing flexibility for eventual COFF/ELF support.
	* struc-symbol.h (symbol_to_segment, symbol_to_segment_type,
	set_symbol_segment, set_symbol_segment_keep_ext,
	segment_name):  Define macros to encapsulate this info.
	* as.h:  Remove externs for seg_name, seg_N_TYPE, N_TYPE_seg.
	* symbols.c (symbol_new):  Change 'type' arg to 'seg'.
	* expr.c, i386.c, m68k.c, ns32k.c, read.c, symbols.c, vax.c,
	write.c:  Use macros.
	* i386.c, m68k.c, ns32k.c, vax.c, write.c: Change 2nd arg type of
	md_estimate_size_before_relax.
	* expr.c, read.c:  Change 'type' arg to symbol_new.
	* read.c, symbols.c, vax.c, write.c:  Move md.h to end of includes.


	Allow expressions to evaluate to registers.
	* as.h:  Add SEG_REGISTER.
	* struc-symbol.h:  Add fake N_REGISTER type.
	* subseg.c:  Add types to tables.
	* expr.c (operand):  Symbols of SEG_REGISTER type are
	immediately evaluated like those of SEG_ABSOLUTE.
	(clean_up_expression):  Clean up SEG_REGISTER exprs.


	Allow machine descriptions to cleanly extend the set of
	possible operands.
	* expr.c (operand):  Call md_operand before rejecting an
	operand as unacceptable.
	* md.h:  declare.
	* i386.c, ns32k.c, m68k.c, sparc.c, vax.c:  Define null function.
	* am29k.c (md_operand):  Use this for %% and & prefix operators.
	

	Allow machine descriptions to cleanly permit symbols to be
	predefined upon first usage.
	* symbols.c (symbol_find_or_make):  Call md_undefined_symbol
	before making an undefined symbol.
	* md.h:  declare.
	* i386.c, ns32k.c, m68k.c, sparc.c, vax.c:  Define null function.
	* am29k.c (md_undefined_symbol):  use this for the local and
	global register names; since there are hundreds of them, it
	only defines them upon their first use.
	* expr.c (operand):  Call symbol_find_or_make rather than
	roll our own undefined symbols.


	Miscellaneous changes:

	* as.h (know):  Make this useful if DEBUG defined.

	* write.h: Support SPARC-like relocation throughout all
	versions.

	* read.c (read_a_source_file): Report the name of invalid
	pseudo-ops.  Don't double-report junk characters.  Close the
	input file, which gas never used to do.
	(ignore_rest_of_line):  Report junk chars in ascii if
	printable.
	(s_ignore):  Ignore entire statement; used for 'listing
	control' statements from ASM29K, e.g. .eject.

	* read.c (s_lsym):  Handle register equates too.

	* read.c:  Add most ASM29K pseudo-ops to the master table.
	Not all are implemented yet.

	* cond.c:  New file, for functions implementing conditional
	assembly pseudo-ops: .ifdef, .ifndef, .else, .endif, .ifseq,
	.ifsne, .end.  So far, they are just stubbed out.

	* read.c (pobegin):  Allow the machine dependent pseudo-op
	table to override the generic one.  Remove #ifdef SPARC
	on .word, since it can now be overridden.

	* expr.c (operand):  Support radix-2 constants.  Kill off
	support for octals with digits '8' and '9'.  Initial steps
	toward more general support for local-labels.

	* symbols.h (symbol_table_lookup):  Remove macro, change all
	occurrences (in read.c, expr.c, symbols.c) to symbol_find.

	* read.h (is_end_of_line):  Define for external use.

	* i386.c (alloca):  Use builtin_alloca or include or extern.

	* Makefile:  Add ALL and all: entries.  Add asm29k entries.
	Add cond.c and cond.o.  Remove special handling for messages.o.
	Add lint entry.

Thu Sep 27 13:43:49 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* m68k.c (get_num)  Fix so that 1:w is treated properly.

	* Makefile  Replace references to a.out.h with a.out.gnu.h

Tue Sep 25 15:50:36 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* sparc.c  (md_number_to_imm)  Fix so that RELOC_32 locations contain
	zero, so that it will work with some sparc loaders which don't assume
	that the locations in question do.  A xix line patch from Michael Bloom
	(usc!srhqla!quad1!psivax!ttidca!mb@zaphod.mps.ohio-state.edu)

Mon Sep 24 11:43:15 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* as.c  #include <sys/types.h> if _POSIX_SOURCE defined.  This because
	<signal.h> uses pid_t that's defined in it.

	* m68k.c reverse the sense of -l option, and allow :w and :l to
	override the default size of AOFF indexes.

	Also, allow -l to shorten branches to unknown labels from LONG to WORD.

Thu Sep 13 17:05:09 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* vax.c (md_parse_option)  Don't print a warning msg if given -J

Wed Sep  5 14:26:14 EDT 1990	Jay Fenlason

	* expr.c  (operand)  Don't get garbaged high-order bits when given a
	lot of leading zeroes.

Tue Sep  4 11:40:21 EDT 1990	Jay Fenlason

	* read.c (pseudo_set) Compain if we're setting the symbol to the
	difference of two symbols which are in different frags.  (We can't
	find out how far apart they are.)

Wed Aug 15 12:18:58 EDT 1990	Jay Fenlason

	* m68k.c (m68k_ip_op)  Dyke out the code that deals with parsing
	:[wl] at the end of expressions since it is handled in get_num()
	and this was putting the result in the wrong place anyway.
	Corrected a couple of other references to ->isiz instead of con?->e_siz

Mon Aug 13 15:53:46 EDT 1990	Jay Fenlason

	* read.c  Handle .align specially on the sparc, or any other machine
	where OTHER_ALIGN is defined.  Added option and comments about it
	to Makefile.
v
Fri Aug 10 12:24:33 EDT 1990	Jay Fenlason

	* m68k.c (get_num)  Handle SEG_PASS1 expressions.

Mon Aug  6 16:32:29 EDT 1990	Jay Fenlason

	* write.c  (fixup_segment) Added two patches for the NS32k
	and SEQUENT  A six line patch from Ian Dall
	(asgard!aegir!hugin!augean!sibyl!ian@munnari.oz.au)

Wed Aug  1 13:30:48 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* m68k.c  Include REGISTER_PREFIX ifdefs.

	* write.c Include LOCAL_LABEL() and DOT_LABEL_PREFIX feature.

	* vax.c (md_parse_option)  Accept -H option.

	* vms.c New version of case hasher, etc.  These from Eric Youngdale
 	  <YOUNGDALE@v6550c.nrl.navy.mil>

Fri Jul 20 13:39:02 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* README	Added README.APOLLO and README.COFF stuff

Wed Jul 18 16:29:22 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* Makefile  Added option for SEQUENT_COMPATABILITY

	* ns32k.c  Add configurable syntax feature from
		ian@sibyl.eleceng.ua.oz@augean.ua.oz.au
		and SEQUENT_COMPATABILITY

	* ns32k-opcode.h  Add configurable syntax feature.

Mon Jul 16 11:44:14 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* write.c (relax_segment)  On ns32k, add fragP->fr_pcrel_adjust to
	aim.
		(fixup_segment)  On ns32k, don't subtract size from
	add_number on pcrel external symbols.

	* ns32k.c (md_relax_table)  Use correct max displacements.
	This is a six-line patch from ian@sibyl.eleceng.ua.oz.au

	* ns32k.c (md_atof, convert_iif)  Emit floating point numbers in
	the correct byte order.  A seven line patch from
 	ian@sibyl.eleceng.ua.oz.au

	* ns32k.c (all over)  Some lint fixes.

Mon Jul  9 13:17:00 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* app.c (do_scrub_next_char)  If a comment is #{whitespace}
	don't treat the next line as comment also.

	* m68k.c (...)  Accept apc@(num:[wl]), etc.

	* i386.c (md_assemble) Get bitfields correct when doing cross
	assembly to 386.  A two line patch from Michael Bloom.
	(usc!srhqla!quad1!ttidca!mb@zaphod.mps.ohio-state.edu).

	* README.APOLLO a new file with vasta@apollo's name, address
	and phone # in it.

	* make-gas.com Deleted references to gdb source files.

Fri Jul  6 14:34:27 EDT 1990	Jay Fenlason (hack@ai.mit.edu)

	* i386.c  Ignore the .optim directive

	* input-file.c  Change from _IOLBF to _IOFBF in setbuffer emulation
	for SYSV.

Mon Jun 18 15:36:49 EDT 1990 Jay Fenlason (hack@ai.mit.edu)

	* sparc.c #ifdef DONTDEF s_sparc_align, since it isn't called from
	  anywhere.

Fri Jun 15 15:53:30 EDT 1990 Jay Fenlason (hack@ai.mit.edu)

	* vax.c (md_parse_option)  make the code agree with the documentation
	  on the behaviour of the -d option.

Thu Jun  7 14:23:54 EDT 1990 Jay Fenlason (hack@ai.mit.edu)

	* atof-ieee.c  (gen_to_words)  Assemble 0r-0 correctly.

	* Makefile Remove last references to gdb*.c files.

	* version.c  New version 1.36

Tue May 22 13:22:26 EDT 1990 Jay Fenlason (hack@ai.mit.edu)

	* Makefile  Mention a work-around for a possible problem with HPUX7.0

Mon May 21 14:06:04 1990  Jim Kingdon  (kingdon at pogo.ai.mit.edu)

	* sparc.c (sparc_ip): Change error message from "not in hash table"
	to "unknown opcode".

Wed May 16 15:33:14 EDT 1990	hack@wookumz

	* i386.c (i386_operand)  Print error msg if given an operand like
	4(mumble) which we can't parse.

	* m68k.c (md_assemble)  Add '3' to the list of operand-places that
	can be found in 'symbol-dependent info'.  Also change
	'confusing width' diagnostic to something more meaningful.

Fri May 11 12:09:21 EDT 1990	hack@wookumz

	app.c (do_scrub_next_char)  Don't flush the line after a line
	consisting of a single '/'  A one-line patch from Mike Kupfer
	(kupfer@orc.olivetti.com)

	* i386.c (md_assemble)  Call frag_wane() before calling frag_new()
	A one line patch from Steve Bleazard (steve@robobar.co.uk

Tue May  8 12:56:25 EDT 1990	hack@wookumz

	* atof-generic.c (atof-generic)  Modified the Infinity detection code
	  to accept 0rinfinity and 0r-infinity as well as 0rinf and 0r-inf

Thu Apr 26 15:17:31 EDT 1990	hack@wookumz

	* atof-ieee.c  Change value of NaNs to 0x7fff ffff (float) and
	0x7fff ffff ffff ffff (double)  If you want some other value for
	NaN, use .long and spell it out yourself.

	atof-generic.c  (atof_generic)  Cleaned up code that detects NaN
 	and Inf.

	vax.c (md_assemble)  print a useful error message if expression()
	returns SEG_PASS1 or SEG_DIFFERENCE and we can't deal with those.

Thu Apr 19 10:30:47 EDT 1990	hack@wookumz

	* input-scrub.c (AFTER_STRING)  Make AFTER_STRING contain a null
	so that the strstr() call when looking for "#NO_APP" after a "#APP"
	will work.  A two character patch from Bruce Robertson
	(bruce@heather.pooh.com)

	* Makefile, i386.c  Use atof-ieee.c instead of atof-i386.c

Mon Apr 16 16:20:55 EDT 1990	hack@wookumz

	* m68k.c (md_relax_table)  Many of the offsets were off by two.
	Fixed some generic spacing problems thoughout the file.

Thu Apr 12 12:22:35 EDT 1990	hack@wookumz

	* sparc.c  (md_ri_to_chars)  Handle little-endian cross assembly.

	* write.c (relax_segment)  Compare addresses correctly to avoid
	  accidentally relaxing a branch that we don't have to.
	  These small changes from John Gilmore (gnu@toad.com)

Fri Apr  6 12:52:15 EDT 1990	hack@wookumz

	* Makefile, expr.c symbols.c  Correctly document the SUN_ASM_SYNTAX
	option, and make it work.

Tue Mar 20 12:46:59 EST 1990

	* as.c (main)  Only trap SIGHUP, SIGINT, SIGPIPE, and SIGTERM,
		and only if they aren't being ignored.  A three line patch
		from Paul Eggert (eggert@twinsun.com)

	* write.c (relax_segment)  Correct typo 'growth - ' should have been
		growth = 

	* atof-vax.c (next_bits, flonum_gen2vax)  Clean up by sharing some
		variables.  While we're at it, fix next_bits so that it
		doesn't use littlenums that don't exist. . .

Tue Mar 13 16:23:21 EST 1990	hack@wookumz

	* Rename atof-m68k.c atof-ieee.c

	* Delete atof-ns32k.c

	* m68k.c sparc.c ns32k.c atof-ieee.c  Call atof-ieee instead of
		atof-m68k or atof-ns32k

	* Makefile	Compile with atof-ieee.c instead of atof-ns32k.c or
		atof-m68k.c

Mon Mar 12 14:06:55 EST 1990	hack@wookumz

	* as.c  If the signal handler gets called twice, exit immediatly.

	* ns32k.c  Call gen_to_words with a pointer of the proper type, and
	  call md_number_to_chars to put the results in the proper byte-order.
	  Whoever wrote this code was *sloppy*!

	* Makefile ns32k.o depends on ns32k.c

	* vax.c  (md_parse_option)  If VMS, accept -+ and -h options.

	* vms.c (VMS_Case_Hack_Symbol)  Replace #if NO_CASE_HACKING
		with references to the -h option.  These small VMS patches
		from Angel Li (angel@flipper.miami.edu).

Thu Mar  8 19:18:59 EST 1990	hack@wookumz
	* vms.c		Some trivial patches from Eric Youngdale
			(YOUNGDALE@v6550c.nrl.navy.mil)

Wed Mar  7 17:12:09 EST 1990	hack@wookumz
	* make-gas.com	(Define error as as_fatal when compiling vax.c and vms.c
			A two line patch from Eric Youngdale
			(YOUNGDALE@v6550c.nrl.navy.mil)

Tue Mar  6 16:01:09 EST 1990	hack@wookumz

	* Makefile  Include ns32k options in makefile.  A small patch from
	David Taylor (taylor@think.com).

	* as.c read.c write.c Makefile  #ifdef DONTDEF out all the gdb
	symbol stuff, since it isn't used anymore and it doesn't work.

Mon Mar  5 14:51:04 EST 1990	hack@wookumz

	* i386.c (md_assemble) Replace memchr() with index().

	* as.c  Trap signals 1 through NSIG, print an error msg, and don't
	produce an object file.

	* m68k.c Added a hack so that fsincosx fpx,fpy:fpz works.

	* messages.c New function: as_bad  This is like as_warn, except
	-W doesn't disable it, and calling it inhibits production of an
	object file and causes a non-zero exit code.

Tue Feb 13 14:25:53 EST 1990	hack@wookumz
	* Makefile  Include G0 and LOADLIBES for Sequent Symmetry.
	Based on a small patch from Johan Widen (jw@sics.se)

Thu Feb  1 14:08:58 EST 1990	hack@wookumz
	* m68k.c  Replace 'abort' with 'abort()' which will work.

Wed Jan 24 17:15:08 EST 1990	hack@ai.mit.edu

	* read.c  (ignore_rest_of_line)  Have it print the first junk char
	in both decimal and %c form.

	(read_a_source_file)  On bad pseudo-op, print out the unknown
	pseudo-op's name.

Tue Jan 23 13:12:48 EST 1990	hack@ai.mit.edu

	* read.c (pseudo_set)	If the symbol is external, have it remain
	external.

	* i386-opcode.h  Allow jc as a synonym for jb and jnc as a syn for jnb.


Wed Jan  3 09:35:31 EST 1990	hack@ai.mit.edu

	* ns32k.c [cpureg_032]  Change register id of psr from 0x0b to 0x0d
	* ns32k-opcode.h	Change shift-counts for lsh and lshd
	to one byte instead of 2 and 4.
	A Trivial patch from John F. Peters (think!ames!practic.com!jfp@eddie)

Tue Dec  5 16:37:44 EST 1989	hack@ai.mit.edu

	* ns32k.c (md_create_{long,short}_jump)  Six line patch from
	John F Peters (think!ames!vine!practice.com!jfp) to use the
	correct addressing mode and byte-order for broken-word stuff.

	* write.c (write_object_file)  One line patch to call fix_new_ns32k
	with the correct # of args.

Fri Dec  1 16:44:21 EST 1989	hack@ai.mit.edu

	* atof-generic.c, flonum-mult.c  A real fix for the trailing-zeroes
	problem from Georg Feil (ghfeil@white.toronto.edu)  (two line change)

Mon Nov 27 15:30:46 EST 1989	hack@ai.mit.edu

	* i386-opcode.h  Fixed opcode-table entry for ljmp.  A one char
	patch from eliot@mgm.mit.edu

Mon Nov 20 12:41:28 EST 1989	hack@ai.mit.edu

	* expr.c  Replace the generic_buffer hack with a more portable one */

	* atof-generic.c (atof_generic)  Ignore trailing zeroes after a decimal
	point.  For some reason trailing zeroes (but not trailing nonzeroes) were
	causing loss of precision.  I don't know why. . .

	* vms.c Change copyright notice.  Install changes from Kenneth Adelman
	(adelman@tgv.com) for c++?  (A dozen lines or so)

Mon Nov 13 11:48:44 EST 1989	hack@ai.mit.edu

	* Makefile  Add BINDIR and use it to control where the executable is
	installed.

	* i386.c Use __builtin_alloca if possible (trivial patch from
	Marco S. Hyman pacbell!dumbcat!marc)

Mon Nov  6 18:24:47 EST 1989	hack@ai.mit.edu

	* version.c  New version: 1.35 will be distributed with the
 	1.36 gcc release.

Mon Oct 30 10:38:11 EST 1989	hack@ai.mit.edu

	* atof-m68k.c (atof_m68k)  Don't put the bits[] array on the stack,
	since it may be pointed to after atof-m68k exits.

Tue Oct 24 11:15:57 EDT 1989	hack@ai.mit.edu

	* atof-m68k.c  Added #define for bcopy on USG systems.
	#ifdef TEST the print_gen() function.

	* a.out.h  if USE_HP_INC_HDR then use ../binutils/hp-include/a.out.h

Fri Oct 13 14:36:48 EDT 1989	hack@ai.mit.edu

	* vax.c (all)  Ran vax through indent -gnu to make it readable.

	vax.c (vip_op)  Correctly assemble code like jmp $*0x11223344
	by setting vip_nbytes to 4 when using an immediate address.
	I hope this works!

	m68k.c (s_proc (new))  Added s_proc no-op pseudo-op.

	Makefile  Added instructions for compiling on Sequent Symmetry
	and HP 9000/300.

	a.out.h	Modified to compile on Sequent and HP above.  (HP port
	based on a msg from asjl@comp.vuw.ac.nz (real name unknown)).

Tue Oct 10 14:39:44 EDT 1989	hack@ai.mit.edu
	* vax.c (vip_op)	Fixed a typo in an error msg and cleaned
	up some spacing stuff.

Wed Sep 27 19:07:12 EDT 1989	hack@ai.mit.edu

	* app.c (do_scrub_next_char)	Fixed parsing of
		# <line> "file" garbage
	text so that it'll work again?  (8 line patch from Mike	Hibler
	(mike@cs.utah.edu))

Mon Sep 18 16:26:01 EDT 1989	hack@ai.mit.edu

	* app.c (do_scrub_next_char): Modify parsing of /* ... */ to work
	on the text /* ****/

	* sparc.c (sparc_ip):  Don't abort on insns that use the Alternate
	Spaces.  Try to assemble them correctly.

Thu Sep 14 11:42:44 EDT 1989	hack@ai.mit.edu

	* sparc.c (md_number_to_imm)  Dozen line patch from jkp@sauna.hut.fi
	(Jyrki Kuoppala) so that gas output will work with shared libraries.

	* ns32k.c Include <string.h> instead of <strings.h> if USG defined.

	(md_end)  free(freeptr_static) instead of free(freeptr) .

	* atof-ns32k.c  Include as.h so that sysV stuff (bzero) will be
	defined if needed.  These ns32k changes from
	nixbur!mollers.pad@seismo.css.gov (Josef Moellers)

Fri Sep  1 11:39:52 EDT 1989	hack@ai.mit.edu

	* atof-m68k.c (gen_to_words)  Get the sign right on negative
	floating-point numbers.

Wed Aug 30 13:59:57 EDT 1989	hack@ai.mit.edu

	* Makefile  Remove the rest of the $< entries that kill sun make

Fri Aug 25 15:00:30 EDT 1989	Nobody You Know (hack@ai.mit.edu)

	* atof-m68k.c (gen_to_words) deal with denormalized floating-point
	numbers.

Tue Aug 22 02:03:05 1989  Roland McGrath  (roland at hobbes.ai.mit.edu)

	* Makefile (gas-dist.tar): Put ChangeLog in the tar file.

	* version.c: Added comment telling Jay Fenl--I mean people--not to put
	changes in version.c, but to use ChangeLog instead.

	* version.c (version_string): Put "GNU" in all-caps.

	* version.c: Moved all comments about changes to ChangeLog (this file).
	Many anonymous entries have been attributed to Jay Fenlason (hack).

Thu Aug 17 15:53:57 1989  Jay Fenlason  (hack at apple-gunkies.ai.mit.edu)

	* Makefile: Removed $< references that seem
	to choke some versions of make.

	* frags.c (frag_grow): Fixed to deal with requests for very
	large frags (larger than frags.chunk_size).

	* app.c (do_scrub_next_char): Have it ignore any characters
	after the filename in a # line "filename".

	* sparc.c (s_common): On an error, don't print out
	input_line_pointer past the end of the line where the error is.

	* atof-generic.c (atof_generic): Accept any case for
	inf and nan.

	* m68k.c (m68_ip): Don't use PC-relative mode for alterable
	addressing modes.

Tue Aug 15 04:58:36 1989  Roland McGrath  (roland at apple-gunkies.ai.mit.edu)

	* sparc.c (md_begin): Rewrote this function to perform consistency
	checks with the new opcode table.

Fri Aug 11 16:01:16 1989  Roland McGrath  (roland at apple-gunkies.ai.mit.edu)

	* sparc-opcode.h (struct sparc_opcode): Replaced `mask' field with
	`lose'; removed `last' field.  Updated all opcodes accordingly.
	Fixed several opcodes that generated the wrong instructions.
	sparc.c (md_begin, sparc_ip): Changed to use new struct sparc_opcode.

Thu Aug  3 14:44:24 1989  Jay Fenlason  (hack at apple-gunkies.ai.mit.edu)

	* Makefile (a32k): Use read- and write-ns32k.o
	* ns32k.c (encode_operand): Make sure pcrel_adjust starts out zeroed.
	* read.c (cons): Call fix_new_ns32k() if NS32K is defined.
	* write.c (write_object_file): Ditto.
	These so that .word sym-sym (etc) will produce values with
	the proper byte-order.

Wed Aug 2 12:55:?? 1989  Jay Fenlason (hack at apple-gunkies.ai.mit.edu)

	* sparc.c (comment_chars[]): Removed '|' because it was causing
	problems.  Probably not the best fix, since I suspect other
	assemblers (68020) may get | in .stabs also, and the 68020 needs
	the '|' comment character.

Mon Jul 31 09:22:28 1989  Roland McGrath  (roland at apple-gunkies.ai.mit.edu)

	* sparc.c (sparc_ip): Allow the characters [0123] in opcodes.

Tue Jul 25 16:32:12 1989  Jay Fenlason  (hack)

	* atof-generic.c (atof_generic): Tried to keep
	size_of_digits_in_littlenum from going negative.

	* sparc-opcode.h: Added duplicate [i+1] entries to go with
	the [1+i] entries already there.  A kludgy fix, but it works.

Mon Jul 24 17:20:03 1989  Jay Fenlason  (hack)

	* write.c (relax_segment): Modified rs_org code so it won't
	occasionally dump core.

	* write.c (pseudo_set): Modified SEG_DIFFERENCE to (perhaps)
	allow one to set a symbol to the difference of two other symbols.

	* ns32k.c (convert_iif): Moved size_so_far+=size and size=0 inside
	 the check for a valid type.

	* sparc-opcode.h: Modified the entries for std "q,[1+i]", "D,[1+i]",
	and "Q,[1+i]".

(In version 1.34)  Jay Fenlason  (hack)

	* Makefile: Reorganized, added stuff to make asparc.

	* sparc.c, sparc-opcode.h, sparc.h: Sparc port.

	* write.c: Set the size of text and bss segments to a multiple of eight
	bytes.

	* m68k.c: Moved .single pseudo-op to machine independent part.

	* atof-generic.c: Fixed type in #ifdef __GNUC__.

	* sparc-opcode.h: Handle "mov REG, %y".

	* make-gas.com: Know that error.c no longer exists.

	* sparc.c: Handle [expr+reg].
	Don't call getExpression when looking for an immediate and getting
	something that starts with % and isn't %hi or %lo.

	* Teach the 68k about long conditional branches.

(In version 1.33)  Jay Fenlason  (hack)

	* Use __builtin_alloca if available.

	* README: Added more instructions for reporting bugs.

	* ns32k-opcode.h: Changed the acbb, acbw, and acbd insns.

	* vax.c: Replaced instances of LENGTH[STRING] with STRING[LENGTH].

	* ns32k.c (encode_operand): Increased max size of bit field for exts
	and inss instructions from 31 to 32 bits.

	* flonum-mult.c (flonum_multip): Fixed typo.

	* m68kc.: Allow #32 to be the same as #0 for bit-field ops.

	* make-gas.com, version.c, hex-value.c, flonum-const.c: VMS fixes.

	* ns32k.c, ns32k-opcode.h: More fixes from taylor@think.com.
	Mostly typos in comments, etc.

	* ns32k-opcode.h: Fixed size of immediate operands to andw and andd
	instructions.

(In version 1.32)  Jay Fenlason  (hack)

	* read.c (s_set): Fixed misnamed variable.

	* as.c: Don't hang if given an invalid option.

	* m68k.c: Fixed bug in creating absolute long addresses for branches.

	* ns3k*: Some small ns32k patches.

	* m68k.c: Recognize 0rnan, 0rinf, 0r-inf.

	* app.c: Don't dump core on unterminated strings.

	* symbols.c: Give reasonable error messages.

	* ns32k*: Allow -m32032 and -m32532 options.

	* atof-*.c: Added support for NaN, Inf, and -Inf in atof_generic and
	the various descriptions.

	* m68k.c (add_fix): Replace occurrences of "width==" with
	"(width)==".  This correct a precedence problem.

	* write.c, struc-symbol.h, m68k-opcode.h, m-hpux.h, Makefile: Changes
	for HP-UX from Chris Hanson (cph@kleph.ai.mit.edu).

	* m68k-opcode.h: Reorder movem insns so gdb will see the ones using the
	register list syntax first.

	* symbols.c (colon): Give more useful error messages when something was
	defined as a .comm and is now trying to be defined locally.
	Also, redefining a symbol is a fatal, not a warning.

	* m68k.c: Fixed a bug in using bignums as literal bit patterns for
	floating-point numbers.

(In version 1.31)  Jay Fenlason  (hack)

	* i386*: More patches.

	* Moved machine-dependent option parsing into the machine-dependent
	source files.

(In version 1.30)  Jay Fenlason  (hack)

	* i386*: New new version.

	* atof-m68k.c: Changed to be smaller, with somewhat better modularity.
	Also fixed an obscure bug wherein next_bits would return random bits.

	* m68k.c: Be more careful about creating PC-relative addressing modes
	on the 68000 and 68010.

	* frags.c (frag_new): Zero out the new frag.

	* Don't choke on "foo= bar" or on formfeeds.

	* read.c: Allow Sun-syntax local labels #ifdef SUN_ASM_SYNTAX.
	* m-sun3.h: Defined SUN_ASM_SYNTAX.

(In version 1.29)  Jay Fenlason  (hack)

	* i386.c: Newer version that fixes a bug wherein a jump instruction
	would be split between two frags.

	* i386*: New version.

	* m68k.c: #ifdef M_SUN and -m68010, produce Sun-2 executables.

(In version 1.28)  Jay Fenlason  (hack)

	* m68k.c: Added .single pseudo-op.

	* Made ". = X" and ".set .,X" equivalent to ".org X".
	The pseudo-symbol "." has the value of the location the assembler is
	currently assembling to.

(In version 1.27)  Jay Fenlason  (hack)

	* Merged ns32k and i386 support.

(In version 1.26)  Jay Fenlason  (hack)

	* Added partial ns32k support.

	* Added RMS's evil .word misfeature.  Invented the -k (kludge) option
	to warn that this misfeature was used.

	* Modified some files to get rid of warnings from GCC.

	* Added fix so that / can also be a comment character by itself.

(In version 1.25)  Jay Fenlason  (hack)

	* Installed patches for VMS.

	* as.h (SIZEOF_STRUCT_FRAG): Added space before backslash-newline.

	* messages.c: Fixed typo.

	* app.c: Handle : correctly.

	* error.c: Removed; no longer used.

	* m68k-opcode.h: Added fnop.
	Fixed to correctly handle fmovem with a register list and
	non-predecriment addressing mode.

	* m68k-opcode.h: Fixed to know about long form of FBcc insns.

	* write.c: Warn if a fixup ended up being wider than its field width.

(In version 1.24)  Jay Fenlason  (hack)

	* Accept and ignore -mc68010 and -m68010 switches.

	* Correctly assemble long subroutine calls on the 68000 without using a
	68020-specific instruction.

	* When calling with no filenames, read stdin.

(In version 1.23)  Jay Fenlason (hack)

	* app.c: Rewritten.

	* xmalloc.c, xrealloc.c: Replaced to work with GCC.

(In version 1.22)  Jay Fenlason  (hack)

	* write.c: Fixed a VMS bug.

	* m68k.c: Fixed a bug having to do with turning absolute into
	PC-relative.

	* atof-m68k.c (atof_m68k, gen_to_words): Try to avoid a problem with
	running off the end of the LITTLENUMS.

	* vax.c: Fixed so parenthesized expressions work.

	* atof-generic.c: Added a cast that fixes problems with some C
	compilers.

(In version 1.21)

	* Changes for VMS support and correct bitfield order for
	cross-assembly.

(In version 1.20)

	* m68k*: Fixed "fmovel #N, fpcr".  Added fpcr and fpsr to the list of
	registers.

(In version 1.19)

	* m68k.c? (md_convert_frag): Don't put the fixups for absolute long to
	PC-relative in the data segment.

	* atof-generic.c: #include <alloca.h> #ifdef sparc.

(In version 1.18)

	* Re-fixed _vfprintf stuff (?).

	* Made "movem REG, ADDR" work.

	* Improved preprocessing, without temporary files.

(In version 1.17)

	* Don't produce an undefined empty symbol for ".globl foo," (a line
	ending with a comma).

	* Fixed a bug wherein ".long X" became ".long 0" on the Sparc.

	* Fixed a bug which caused many "#APP" "#NO_APP" pairs to dump core.

	* Fixed calls to _doprnt to call _vfprintf #ifndef NO_VARARGS.

(In version 1.16)

	* Merged HP-UX changes from Chris Hanson (cph@zurich.ai.mit.edu).

	* flonum-multip.c: Renamed to flonum-mult.c.

	* m-hpux.h: Created.

	* m68k.c (bcopy): Fixed.

(In version 1.15)

	* struct-symbol.h: Renamed to struc-symbol.h.

(In version 1.14)

	* vax.c: Added a quick fix for the offset of fixed-width branches not
	fitting in the field given.

	* gdb-lines.c, read.c: Added support for .gdline and .gdbline
	pseudo-ops.

(In version 1.13)

	* read.c, atof-generic.c: Fixed bugs in reading in floating-point
	numbers.

	* m68k-opcode.h: Made "fmovep a0@, fp0" work.

(In version 1.12)

	* write.c: Fixed an obscure bug in relaction that would occasionally
	cause the assembler to stop relaxing when it really had at least one
	more pass to do.

(In version 1.11)

	* m68k*: Allow register lists in fmovem.

	* Added more floating-point exponents.

	* Print an error message on exponent overflow.

(In version 1.10)

	* Fixed floating point bugs that made it generate incorrect numbers for
	values over 10^16 or so.

(In version 1.09)

	* Fixed bug wherein you couldn't forward reference local label 0.

(In version 1.08)

	* m68k.c, m68k-opcode.h: Added support for fmovem with register lists.

	* Fixed an obscure bug having to do with generating PC-relative
	addressing mode for things in the middle of the instruction instead of
	at the end.

Wed Mar  1 15:29:24 1989  Randall Smith  (randy at apple-gunkies.ai.mit.edu)

	* *.*: Modified copyright notices to reflect new General Public
	License. 

	* Makefile: Added copyright notice.

Fri Feb 17 09:42:01 1989  Jay Fenlason  (hack at spiff)

	* Patched frags.c so that new frags start out bzero()ed.

Thu Jan 26 14:23:44 1989  Jay Fenlason  (hack at apple-gunkies.ai.mit.edu)

	* Added patches from pace to files as.h i386.c i386-opcode.h
	  imull foo,%eax no longer gets assembled into the 32-64 bit
	  multiply, which clobbers %edx behind gcc's back

	  jcxz/jecxz were backwards

	  There was a bug when using %ebp as a base register with no
	  displacement

	  Instructions like andb $0xffffff, %al used to put out too many
	  immediate bytes

	  The splitting jump instructions across frags could happen when
	  obstack_room()==6 too.

Local Variables:
mode: indented-text
left-margin: 8
version-control: never
End: