What has changed since GDB-3.5? (Organized release by release) *** Changes in GDB-4.9: (This is a prototype to remind us of things that should be announced in the next release...) 'Cfront' style demangling has had its name changed to 'ARM' style, to emphasize that it was written from the specifications in the Annotated Reference Manual, not to be compatible with AT&T cfront. Despite disclaimers, it still generated too much confusion with users attempting to use gdb with AT&T cfront. * Simulators GDB now uses a standard remote interface to a simulator library. So far, the library contains simulators for the Zilog Z8001/2, the Hitachi H8/300, H8/500 and Super-H. * New targets supported H8/300 simulator h8300-hitachi-hms or h8300hms H8/500 simulator h8500-hitachi-hms or h8500hms SH simulator sh-hitachi-hms or sh Z8000 simulator z8k-zilog-none or z8ksim Cross-debugging to GO32 targets is supported. It requires a custom version of the i386-stub.c module which is integrated with the GO32 memory extender. Msg follows: Date: Tue, 16 Feb 93 02:34:20 EST From: "Mark W. Eichin" Message-Id: <9302160734.AA09302@tweedledumb.cygnus.com> To: gnu@cygnus.com Cc: ian@cygnus.com, gnu@cygnus.com, gumby@cygnus.com, gdb@cygnus.com In-Reply-To: gnu@cygnus.com's message of Mon, 15 Feb 93 22:30:09 -0800 <9302160630.AA00786@cygnus.com> Subject: GO32 debugging in devo/gdb SUB: GO32 debugging in devo/gdb SUM: , gnu->eichin, ian, gnu, gumby, gdb My impression is that devo/gdb supports remote debugging of GO32 programs. Is this true? Yes. I think that even the 4.7 release had everything needed. What does a user have to have in the GO32 environment in order to do this? (My guess: our custom-modified GO32. Did we send the changes back to DJ and did they ever get integrated into the standard GO32?) I asked DJ if he wanted the changes; at the time, he was very busy having a daughter. He's back on the net now, I'll give him another try. My changes are to GO32 1.07 and the entire source (and an executable) are checked in to cvs; the current GO32 is 1.08, I haven't tried updating the changes. What does a user have to actually do in GO32 in order for this to work? E.g. there seems to be no user-level documentation for this feature. GO32 includes "go32.exe" and "debug32.exe"; my version is "dser32.exe". With a serial link on com1 to the host, use the mode command on the target to set the baud rate, then "dser32 a.out" and start up gdb (configured -target go32), target remote /dev/ttya. Shoudl just work from there. I'm wondering if we can announce this as part of what's supported in gdb-4.8. The hard part is the extender itself -- it needs to be built with a native 16-bit compiler (such as Turbo C with Turbo Assembler -- about $300 in software, which I do own -- and the assembly code uses enough high level features (like structs) that it isn't portable to other assemblers.) We have no way to build it with any free tools. I think we can ship (or at least make available) the executable for the DOS side, I don't think Turbo C has any runtime restrictions. _Mark_ *** Changes in GDB-4.8: * HP Precision Architecture supported GDB now supports HP PA-RISC machines running HPUX. A preliminary version of this support was available as a set of patches from the University of Utah. GDB does not support debugging of programs compiled with the HP compiler, because HP will not document their file format. Instead, you must use GCC (version 2.3.2 or later) and PA-GAS (as available from jaguar.cs.utah.edu:/dist/pa-gas.u4.tar.Z). Many problems in the preliminary version have been fixed. * Faster and better demangling We have improved template demangling and fixed numerous bugs in the GNU style demangler. It can now handle type modifiers such as `static' or `const'. Wide character types (wchar_t) are now supported. Demangling of each symbol is now only done once, and is cached when the symbol table for a file is read in. This results in a small increase in memory usage for C programs, a moderate increase in memory usage for C++ programs, and a fantastic speedup in symbol lookups. `Cfront' style demangling still doesn't work with AT&T cfront. It was written from the specifications in the Annotated Reference Manual, which AT&T's compiler does not actually implement. * G++ multiple inheritance compiler problem In the 2.3.2 release of gcc/g++, how the compiler resolves multiple inheritance lattices was reworked to properly discover ambiguities. We recently found an example which causes this new algorithm to fail in a very subtle way, producing bad debug information for those classes. The file 'gcc.patch' (in this directory) can be applied to gcc to circumvent the problem. A future GCC release will contain a complete fix. The previous G++ debug info problem (mentioned below for the gdb-4.7 release) is fixed in gcc version 2.3.2. * Improved configure script The `configure' script will now attempt to guess your system type if you don't supply a host system type. The old scheme of supplying a host system triplet is preferable over using this. All the magic is done in the new `config.guess' script. Examine it for details. We have also brought our configure script much more in line with the FSF's version. It now supports the --with-xxx options. In particular, `--with-minimal-bfd' can be used to make the GDB binary image smaller. The resulting GDB will not be able to read arbitrary object file formats -- only the format ``expected'' to be used on the configured target system. We hope to make this the default in a future release. * Documentation improvements There's new internal documentation on how to modify GDB, and how to produce clean changes to the code. We implore people to read it before submitting changes. The GDB manual uses new, sexy Texinfo conditionals, rather than arcane M4 macros. The new texinfo.tex is provided in this release. Pre-built `info' files are also provided. To build `info' files from scratch, you will need the latest `makeinfo' release, which will be available in a future texinfo-X.Y release. *NOTE* The new texinfo.tex can cause old versions of TeX to hang. We're not sure exactly which versions have this problem, but it has been seen in 3.0. We highly recommend upgrading to TeX version 3.141 or better. If that isn't possible, there is a patch in `texinfo/tex3patch' that will modify `texinfo/texinfo.tex' to work around this problem. * New features GDB now supports array constants that can be used in expressions typed in by the user. The syntax is `{element, element, ...}'. Ie: you can now type `print {1, 2, 3}', and it will build up an array in memory malloc'd in the target program. The new directory `gdb/sparclite' contains a program that demonstrates how the sparc-stub.c remote stub runs on a Fujitsu SPARClite processor. * New native hosts supported HP/PA-RISC under HPUX using GNU tools hppa1.1-hp-hpux 386 CPUs running SCO Unix 3.2v4 i386-unknown-sco3.2v4 * New targets supported AMD 29k family via UDI a29k-amd-udi or udi29k * New file formats supported BFD now supports reading HP/PA-RISC executables (SOM file format?), HPUX core files, and SCO 3.2v2 core files. * Major bug fixes Attaching to processes now works again; thanks for the many bug reports. We have also stomped on a bunch of core dumps caused by printf_filtered("%s") problems. We eliminated a copyright problem on the rpc and ptrace header files for VxWorks, which was discovered at the last minute during the 4.7 release. You should now be able to build a VxWorks GDB. You can now interrupt gdb while an attached process is running. This will cause the attached process to stop, and give control back to GDB. We fixed problems caused by using too many file descriptors for reading symbols from object files and libraries. This was especially a problem for programs that used many (~100) shared libraries. The `step' command now only enters a subroutine if there is line number information for the subroutine. Otherwise it acts like the `next' command. Previously, `step' would enter subroutines if there was any debugging information about the routine. This avoids problems when using `cc -g1' on MIPS machines. * Internal improvements GDB's internal interfaces have been improved to make it easier to support debugging of multiple languages in the future. GDB now uses a common structure for symbol information internally. Minimal symbols (derived from linkage symbols in object files), partial symbols (from a quick scan of debug information), and full symbols contain a common subset of information, making it easier to write shared code that handles any of them. * New command line options We now accept --silent as an alias for --quiet. * Mmalloc licensing The memory-mapped-malloc library is now licensed under the GNU Library General Public License. *** Changes in GDB-4.7: * Host/native/target split GDB has had some major internal surgery to untangle the support for hosts and remote targets. Now, when you configure GDB for a remote target, it will no longer load in all of the support for debugging local programs on the host. When fully completed and tested, this will ensure that arbitrary host/target combinations are possible. The primary conceptual shift is to separate the non-portable code in GDB into three categories. Host specific code is required any time GDB is compiled on that host, regardless of the target. Target specific code relates to the peculiarities of the target, but can be compiled on any host. Native specific code is everything else: it can only be built when the host and target are the same system. Child process handling and core file support are two common `native' examples. GDB's use of /proc for controlling Unix child processes is now cleaner. It has been split out into a single module under the `target_ops' vector, plus two native-dependent functions for each system that uses /proc. * New hosts supported HP/Apollo 68k (under the BSD domain) m68k-apollo-bsd or apollo68bsd 386 CPUs running various BSD ports i386-unknown-bsd or 386bsd 386 CPUs running SCO Unix i386-unknown-scosysv322 or i386sco * New targets supported Fujitsu SPARClite sparclite-fujitsu-none or sparclite 68030 and CPU32 m68030-*-*, m68332-*-* * New native hosts supported 386 CPUs running various BSD ports i386-unknown-bsd or 386bsd (386bsd is not well tested yet) 386 CPUs running SCO Unix i386-unknown-scosysv322 or sco * New file formats supported BFD now supports COFF files for the Zilog Z8000 microprocessor. It supports reading of `a.out.adobe' object files, which are an a.out format extended with minimal information about multiple sections. * New commands `show copying' is the same as the old `info copying'. `show warranty' is the same as `info warrantee'. These were renamed for consistency. The old commands continue to work. `info handle' is a new alias for `info signals'. You can now define pre-command hooks, which attach arbitrary command scripts to any command. The commands in the hook will be executed prior to the user's command. You can also create a hook which will be executed whenever the program stops. See gdb.texinfo. * C++ improvements We now deal with Cfront style name mangling, and can even extract type info from mangled symbols. GDB can automatically figure out which symbol mangling style your C++ compiler uses. Calling of methods and virtual functions has been improved as well. * Major bug fixes The crash that occured when debugging Sun Ansi-C compiled binaries is fixed. This was due to mishandling of the extra N_SO stabs output by the compiler. We also finally got Ultrix 4.2 running in house, and fixed core file support, with help from a dozen people on the net. John M. Farrell discovered that the reason that single-stepping was so slow on all of the Mips based platforms (primarily SGI and DEC) was that we were trying to demangle and lookup a symbol used for internal purposes on every instruction that was being stepped through. Changing the name of that symbol so that it couldn't be mistaken for a C++ mangled symbol sped things up a great deal. Rich Pixley sped up symbol lookups in general by getting much smarter about when C++ symbol mangling is necessary. This should make symbol completion (TAB on the command line) much faster. It's not as fast as we'd like, but it's significantly faster than gdb-4.6. * AMD 29k support A new user controllable variable 'call_scratch_address' can specify the location of a scratch area to be used when GDB calls a function in the target. This is necessary because the usual method of putting the scratch area on the stack does not work in systems that have separate instruction and data spaces. We integrated changes to support the 29k UDI (Universal Debugger Interface), but discovered at the last minute that we didn't have all of the appropriate copyright paperwork. We are working with AMD to resolve this, and hope to have it available soon. * Remote interfaces We have sped up the remote serial line protocol, especially for targets with lots of registers. It now supports a new `expedited status' ('T') message which can be used in place of the existing 'S' status message. This allows the remote stub to send only the registers that GDB needs to make a quick decision about single-stepping or conditional breakpoints, eliminating the need to fetch the entire register set for each instruction being stepped through. The GDB remote serial protocol now implements a write-through cache for registers, only re-reading the registers if the target has run. There is also a new remote serial stub for SPARC processors. You can find it in gdb-4.7/gdb/sparc-stub.c. This was written to support the Fujitsu SPARClite processor, but will run on any stand-alone SPARC processor with a serial port. * Configuration Configure.in files have become much easier to read and modify. A new `table driven' format makes it more obvious what configurations are supported, and what files each one uses. * Library changes There is a new opcodes library which will eventually contain all of the disassembly routines and opcode tables. At present, it only contains Sparc and Z8000 routines. This will allow the assembler, debugger, and disassembler (binutils/objdump) to share these routines. The libiberty library is now copylefted under the GNU Library General Public License. This allows more liberal use, and was done so libg++ can use it. This makes no difference to GDB, since the Library License grants all the rights from the General Public License. * Documentation The file gdb-4.7/gdb/doc/stabs.texinfo is a (relatively) complete reference to the stabs symbol info used by the debugger. It is (as far as we know) the only published document on this fascinating topic. We encourage you to read it, compare it to the stabs information on your system, and send improvements on the document in general (to bug-gdb@prep.ai.mit.edu). And, of course, many bugs have been fixed. *** Changes in GDB-4.6: * Better support for C++ function names GDB now accepts as input the "demangled form" of C++ overloaded function names and member function names, and can do command completion on such names (using TAB, TAB-TAB, and ESC-?). The names have to be quoted with a pair of single quotes. Examples are 'func (int, long)' and 'obj::operator==(obj&)'. Make use of command completion, it is your friend. GDB also now accepts a variety of C++ mangled symbol formats. They are the GNU g++ style, the Cfront (ARM) style, and the Lucid (lcc) style. You can tell GDB which format to use by doing a 'set demangle-style {gnu, lucid, cfront, auto}'. 'gnu' is the default. Do a 'set demangle-style foo' for the list of formats. * G++ symbol mangling problem Recent versions of gcc have a bug in how they emit debugging information for C++ methods (when using dbx-style stabs). The file 'gcc.patch' (in this directory) can be applied to gcc to fix the problem. Alternatively, if you can't fix gcc, you can #define GCC_MANGLE_BUG when compling gdb/symtab.c. The usual symptom is difficulty with setting breakpoints on methods. GDB complains about the method being non-existent. (We believe that version 2.2.2 of GCC has this problem.) * New 'maintenance' command All of the commands related to hacking GDB internals have been moved out of the main command set, and now live behind the 'maintenance' command. This can also be abbreviated as 'mt'. The following changes were made: dump-me -> maintenance dump-me info all-breakpoints -> maintenance info breakpoints printmsyms -> maintenance print msyms printobjfiles -> maintenance print objfiles printpsyms -> maintenance print psymbols printsyms -> maintenance print symbols The following commands are new: maintenance demangle Call internal GDB demangler routine to demangle a C++ link name and prints the result. maintenance print type Print a type chain for a given symbol * Change to .gdbinit file processing We now read the $HOME/.gdbinit file before processing the argv arguments (e.g. reading symbol files or core files). This allows global parameters to be set, which will apply during the symbol reading. The ./.gdbinit is still read after argv processing. * New hosts supported Solaris-2.0 !!! sparc-sun-solaris2 or sun4sol2 Linux support i386-unknown-linux or linux We are also including code to support the HP/PA running BSD and HPUX. This is almost guaranteed not to work, as we didn't have time to test or build it for this release. We are including it so that the more adventurous (or masochistic) of you can play with it. We also had major problems with the fact that the compiler that we got from HP doesn't support the -g option. It costs extra. * New targets supported Hitachi H8/300 h8300-hitachi-hms or h8300hms * More smarts about finding #include files GDB now remembers the compilation directory for all include files, and for all files from which C is generated (like yacc and lex sources). This greatly improves GDB's ability to find yacc/lex sources, and include files, especially if you are debugging your program from a directory different from the one that contains your sources. We also fixed a bug which caused difficulty with listing and setting breakpoints in include files which contain C code. (In the past, you had to try twice in order to list an include file that you hadn't looked at before.) * Interesting infernals change GDB now deals with arbitrary numbers of sections, where the symbols for each section must be relocated relative to that section's landing place in the target's address space. This work was needed to support ELF with embedded stabs used by Solaris-2.0. * Bug fixes (of course!) There have been loads of fixes for the following things: mips, rs6000, 29k/udi, m68k, g++, type handling, elf/dwarf, m88k, i960, stabs, DOS(GO32), procfs, etc... See the ChangeLog for details. *** Changes in GDB-4.5: * New machines supported (host and target) IBM RS6000 running AIX rs6000-ibm-aix or rs6000 SGI Irix-4.x mips-sgi-irix4 or iris4 * New malloc package GDB now uses a new memory manager called mmalloc, based on gmalloc. Mmalloc is capable of handling mutiple heaps of memory. It is also capable of saving a heap to a file, and then mapping it back in later. This can be used to greatly speedup the startup of GDB by using a pre-parsed symbol table which lives in a mmalloc managed heap. For more details, please read mmalloc/mmalloc.texi. * info proc The 'info proc' command (SVR4 only) has been enhanced quite a bit. See 'help info proc' for details. * MIPS ecoff symbol table format The code that reads MIPS symbol table format is now supported on all hosts. Thanks to MIPS for releasing the sym.h and symconst.h files to make this possible. * File name changes for MS-DOS Many files in the config directories have been renamed to make it easier to support GDB on MS-DOSe systems (which have very restrictive file name conventions :-( ). MS-DOSe host support (under DJ Delorie's GO32 environment) is close to working but has some remaining problems. Note that debugging of DOS programs is not supported, due to limitations in the ``operating system'', but it can be used to host cross-debugging. * Cross byte order fixes Many fixes have been made to support cross debugging of Sparc and MIPS targets from hosts whose byte order differs. * New -mapped and -readnow options If memory-mapped files are available on your system through the 'mmap' system call, you can use the -mapped option on the `file' or `symbol-file' commands to cause GDB to write the symbols from your program into a reusable file. If the program you are debugging is called `/path/fred', the mapped symbol file will be `./fred.syms'. Future GDB debugging sessions will notice the presence of this file, and will quickly map in symbol information from it, rather than reading the symbol table from the executable program. Using the '-mapped' option in a GDB `file' or `symbol-file' command has the same effect as starting GDB with the '-mapped' command-line option. You can cause GDB to read the entire symbol table immediately by using the '-readnow' option with any of the commands that load symbol table information (or on the GDB command line). This makes the command slower, but makes future operations faster. The -mapped and -readnow options are typically combined in order to build a `fred.syms' file that contains complete symbol information. A simple GDB invocation to do nothing but build a `.syms' file for future use is: gdb -batch -nx -mapped -readnow programname The `.syms' file is specific to the host machine on which GDB is run. It holds an exact image of GDB's internal symbol table. It cannot be shared across multiple host platforms. * longjmp() handling GDB is now capable of stepping and nexting over longjmp(), _longjmp(), and siglongjmp() without losing control. This feature has not yet been ported to all systems. It currently works on many 386 platforms, all MIPS-based platforms (SGI, DECstation, etc), and Sun3/4. * Solaris 2.0 Preliminary work has been put in to support the new Solaris OS from Sun. At this time, it can control and debug processes, but it is not capable of reading symbols. * Bug fixes As always, many many bug fixes. The major areas were with g++, and mipsread. People using the MIPS-based platforms should experience fewer mysterious crashes and trashed symbol tables. *** Changes in GDB-4.4: * New machines supported (host and target) SCO Unix on i386 IBM PC clones i386-sco-sysv or i386sco (except core files) BSD Reno on Vax vax-dec-bsd Ultrix on Vax vax-dec-ultrix * New machines supported (target) AMD 29000 embedded, using EBMON a29k-none-none * C++ support GDB continues to improve its handling of C++. `References' work better. The demangler has also been improved, and now deals with symbols mangled as per the Annotated C++ Reference Guide. GDB also now handles `stabs' symbol information embedded in MIPS `ecoff' symbol tables. Since the ecoff format was not easily extensible to handle new languages such as C++, this appeared to be a good way to put C++ debugging info into MIPS binaries. This option will be supported in the GNU C compiler, version 2, when it is released. * New features for SVR4 GDB now handles SVR4 shared libraries, in the same fashion as SunOS shared libraries. Debugging dynamically linked programs should present only minor differences from debugging statically linked programs. The `info proc' command will print out information about any process on an SVR4 system (including the one you are debugging). At the moment, it prints the address mappings of the process. If you bring up GDB on another SVR4 system, please send mail to bug-gdb@prep.ai.mit.edu to let us know what changes were reqired (if any). * Better dynamic linking support in SunOS Reading symbols from shared libraries which contain debugging symbols now works properly. However, there remain issues such as automatic skipping of `transfer vector' code during function calls, which make it harder to debug code in a shared library, than to debug the same code linked statically. * New Getopt GDB is now using the latest `getopt' routines from the FSF. This version accepts the -- prefix for options with long names. GDB will continue to accept the old forms (-option and +option) as well. Various single letter abbreviations for options have been explicity added to the option table so that they won't get overshadowed in the future by other options that begin with the same letter. * Bugs fixed The `cleanup_undefined_types' bug that many of you noticed has been squashed. Many assorted bugs have been handled. Many more remain to be handled. See the various ChangeLog files (primarily in gdb and bfd) for details. *** Changes in GDB-4.3: * New machines supported (host and target) Amiga 3000 running Amix m68k-cbm-svr4 or amix NCR 3000 386 running SVR4 i386-ncr-svr4 or ncr3000 Motorola Delta 88000 running Sys V m88k-motorola-sysv or delta88 * Almost SCO Unix support We had hoped to support: SCO Unix on i386 IBM PC clones i386-sco-sysv or i386sco (except for core file support), but we discovered very late in the release that it has problems with process groups that render gdb unusable. Sorry about that. I encourage people to fix it and post the fixes. * Preliminary ELF and DWARF support GDB can read ELF object files on System V Release 4, and can handle debugging records for C, in DWARF format, in ELF files. This support is preliminary. If you bring up GDB on another SVR4 system, please send mail to bug-gdb@prep.ai.mit.edu to let us know what changes were reqired (if any). * New Readline GDB now uses the latest `readline' library. One user-visible change is that two tabs will list possible command completions, which previously required typing M-? (meta-question mark, or ESC ?). * Bugs fixed The `stepi' bug that many of you noticed has been squashed. Many bugs in C++ have been handled. Many more remain to be handled. See the various ChangeLog files (primarily in gdb and bfd) for details. * State of the MIPS world (in case you wondered): GDB can understand the symbol tables emitted by the compilers supplied by most vendors of MIPS-based machines, including DEC. These symbol tables are in a format that essentially nobody else uses. Some versions of gcc come with an assembler post-processor called mips-tfile. This program is required if you want to do source-level debugging of gcc-compiled programs. I believe FSF does not ship mips-tfile with gcc version 1, but it will eventually come with gcc version 2. Debugging of g++ output remains a problem. g++ version 1.xx does not really support it at all. (If you're lucky, you should be able to get line numbers and stack traces to work, but no parameters or local variables.) With some work it should be possible to improve the situation somewhat. When gcc version 2 is released, you will have somewhat better luck. However, even then you will get confusing results for inheritance and methods. We will eventually provide full debugging of g++ output on DECstations. This will probably involve some kind of stabs-in-ecoff encapulation, but the details have not been worked out yet. *** Changes in GDB-4.2: * Improved configuration Only one copy of `configure' exists now, and it is not self-modifying. Porting BFD is simpler. * Stepping improved The `step' and `next' commands now only stop at the first instruction of a source line. This prevents the multiple stops that used to occur in switch statements, for-loops, etc. `Step' continues to stop if a function that has debugging information is called within the line. * Bug fixing Lots of small bugs fixed. More remain. * New host supported (not target) Intel 386 PC clone running Mach i386-none-mach *** Changes in GDB-4.1: * Multiple source language support GDB now has internal scaffolding to handle several source languages. It determines the type of each source file from its filename extension, and will switch expression parsing and number formatting to match the language of the function in the currently selected stack frame. You can also specifically set the language to be used, with `set language c' or `set language modula-2'. * GDB and Modula-2 GDB now has preliminary support for the GNU Modula-2 compiler, currently under development at the State University of New York at Buffalo. Development of both GDB and the GNU Modula-2 compiler will continue through the fall of 1991 and into 1992. Other Modula-2 compilers are currently not supported, and attempting to debug programs compiled with them will likely result in an error as the symbol table is read. Feel free to work on it, though! There are hooks in GDB for strict type checking and range checking, in the `Modula-2 philosophy', but they do not currently work. * set write on/off GDB can now write to executable and core files (e.g. patch a variable's value). You must turn this switch on, specify the file ("exec foo" or "core foo"), *then* modify it, e.g. by assigning a new value to a variable. Modifications take effect immediately. * Automatic SunOS shared library reading When you run your program, GDB automatically determines where its shared libraries (if any) have been loaded, and reads their symbols. The `share' command is no longer needed. This also works when examining core files. * set listsize You can specify the number of lines that the `list' command shows. The default is 10. * New machines supported (host and target) SGI Iris (MIPS) running Irix V3: mips-sgi-irix or iris Sony NEWS (68K) running NEWSOS 3.x: m68k-sony-sysv or news Ultracomputer (29K) running Sym1: a29k-nyu-sym1 or ultra3 * New hosts supported (not targets) IBM RT/PC: romp-ibm-aix or rtpc * New targets supported (not hosts) AMD 29000 embedded with COFF a29k-none-coff AMD 29000 embedded with a.out a29k-none-aout Ultracomputer remote kernel debug a29k-nyu-kern * New remote interfaces AMD 29000 Adapt AMD 29000 Minimon *** Changes in GDB-4.0: * New Facilities Wide output is wrapped at good places to make the output more readable. Gdb now supports cross-debugging from a host machine of one type to a target machine of another type. Communication with the target system is over serial lines. The ``target'' command handles connecting to the remote system; the ``load'' command will download a program into the remote system. Serial stubs for the m68k and i386 are provided. Gdb also supports debugging of realtime processes running under VxWorks, using SunRPC Remote Procedure Calls over TCP/IP to talk to a debugger stub on the target system. New CPUs supported include the AMD 29000 and Intel 960. GDB now reads object files and symbol tables via a ``binary file'' library, which allows a single copy of GDB to debug programs of multiple object file types such as a.out and coff. There is now a GDB reference card in "doc/refcard.tex". (Make targets refcard.dvi and refcard.ps are available to format it). * Control-Variable user interface simplified All variables that control the operation of the debugger can be set by the ``set'' command, and displayed by the ``show'' command. For example, ``set prompt new-gdb=>'' will change your prompt to new-gdb=>. ``Show prompt'' produces the response: Gdb's prompt is new-gdb=>. What follows are the NEW set commands. The command ``help set'' will print a complete list of old and new set commands. ``help set FOO'' will give a longer description of the variable FOO. ``show'' will show all of the variable descriptions and their current settings. confirm on/off: Enables warning questions for operations that are hard to recover from, e.g. rerunning the program while it is already running. Default is ON. editing on/off: Enables EMACS style command line editing of input. Previous lines can be recalled with control-P, the current line can be edited with control-B, you can search for commands with control-R, etc. Default is ON. history filename NAME: NAME is where the gdb command history will be stored. The default is .gdb_history, or the value of the environment variable GDBHISTFILE. history size N: The size, in commands, of the command history. The default is 256, or the value of the environment variable HISTSIZE. history save on/off: If this value is set to ON, the history file will be saved after exiting gdb. If set to OFF, the file will not be saved. The default is OFF. history expansion on/off: If this value is set to ON, then csh-like history expansion will be performed on command line input. The default is OFF. radix N: Sets the default radix for input and output. It can be set to 8, 10, or 16. Note that the argument to "radix" is interpreted in the current radix, so "set radix 10" is always a no-op. height N: This integer value is the number of lines on a page. Default is 24, the current `stty rows'' setting, or the ``li#'' setting from the termcap entry matching the environment variable TERM. width N: This integer value is the number of characters on a line. Default is 80, the current `stty cols'' setting, or the ``co#'' setting from the termcap entry matching the environment variable TERM. Note: ``set screensize'' is obsolete. Use ``set height'' and ``set width'' instead. print address on/off: Print memory addresses in various command displays, such as stack traces and structure values. Gdb looks more ``symbolic'' if you turn this off; it looks more ``machine level'' with it on. Default is ON. print array on/off: Prettyprint arrays. New convenient format! Default is OFF. print demangle on/off: Print C++ symbols in "source" form if on, "raw" form if off. print asm-demangle on/off: Same, for assembler level printouts like instructions. print vtbl on/off: Prettyprint C++ virtual function tables. Default is OFF. * Support for Epoch Environment. The epoch environment is a version of Emacs v18 with windowing. One new command, ``inspect'', is identical to ``print'', except that if you are running in the epoch environment, the value is printed in its own window. * Support for Shared Libraries GDB can now debug programs and core files that use SunOS shared libraries. Symbols from a shared library cannot be referenced before the shared library has been linked with the program (this happens after you type ``run'' and before the function main() is entered). At any time after this linking (including when examining core files from dynamically linked programs), gdb reads the symbols from each shared library when you type the ``sharedlibrary'' command. It can be abbreviated ``share''. sharedlibrary REGEXP: Load shared object library symbols for files matching a unix regular expression. No argument indicates to load symbols for all shared libraries. info sharedlibrary: Status of loaded shared libraries. * Watchpoints A watchpoint stops execution of a program whenever the value of an expression changes. Checking for this slows down execution tremendously whenever you are in the scope of the expression, but is quite useful for catching tough ``bit-spreader'' or pointer misuse problems. Some machines such as the 386 have hardware for doing this more quickly, and future versions of gdb will use this hardware. watch EXP: Set a watchpoint (breakpoint) for an expression. info watchpoints: Information about your watchpoints. delete N: Deletes watchpoint number N (same as breakpoints). disable N: Temporarily turns off watchpoint number N (same as breakpoints). enable N: Re-enables watchpoint number N (same as breakpoints). * C++ multiple inheritance When used with a GCC version 2 compiler, GDB supports multiple inheritance for C++ programs. * C++ exception handling Gdb now supports limited C++ exception handling. Besides the existing ability to breakpoint on an exception handler, gdb can breakpoint on the raising of an exception (before the stack is peeled back to the handler's context). catch FOO: If there is a FOO exception handler in the dynamic scope, set a breakpoint to catch exceptions which may be raised there. Multiple exceptions (``catch foo bar baz'') may be caught. info catch: Lists all exceptions which may be caught in the current stack frame. * Minor command changes The command ``call func (arg, arg, ...)'' now acts like the print command, except it does not print or save a value if the function's result is void. This is similar to dbx usage. The ``up'' and ``down'' commands now always print the frame they end up at; ``up-silently'' and `down-silently'' can be used in scripts to change frames without printing. * New directory command 'dir' now adds directories to the FRONT of the source search path. The path starts off empty. Source files that contain debug information about the directory in which they were compiled can be found even with an empty path; Sun CC and GCC include this information. If GDB can't find your source file in the current directory, type "dir .". * Configuring GDB for compilation For normal use, type ``./configure host''. See README or gdb.texinfo for more details. GDB now handles cross debugging. If you are remotely debugging between two different machines, type ``./configure host -target=targ''. Host is the machine where GDB will run; targ is the machine where the program that you are debugging will run.