If you find inaccuracies in this list, please send mail to gdb-patches@sourceware.cygnus.com. If you would like to work on any of these, you should consider sending mail to the same address, to find out whether anyone else is working on it. Known problems in GDB 5.0 ========================= Below is a list of problems identified during the GDB 5.0 release cycle. People hope to have these problems fixed in a follow-on release. -- The BFD directory requires bug-fixed AUTOMAKE et.al. AUTOMAKE 1.4 incorrectly set the TEXINPUTS environment variable. It contained the full path to texinfo.tex when it should have only contained the directory. The bug has been fixed in the current AUTOMAKE sources. Automake snapshots can be found in: ftp://sourceware.cygnus.com/pub/gdb/snapshots and ftp://sourceware.cygnus.com/pub/binutils -- RFD: infrun.c: No bpstat_stop_status call after proceed over break? http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00665.html GDB misses watchpoint triggers after proceeding over a breakpoint on x86 targets. -- x86 linux GDB and SIGALRM (???) http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00803.html This problem has been fixed, but a regression test still needs to be added to the testsuite: http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00309.html Mark -- Revised UDP support (was: Re: [Fwd: [patch] UDP transport support]) http://sourceware.cygnus.com/ml/gdb-patches/2000-04/msg00000.html (Broken) support for GDB's remote protocol across UDP is to be included in the follow-on release. -- Can't build IRIX -> arm GDB. http://sourceware.cygnus.com/ml/gdb-patches/2000-04/msg00356.html David Whedon writes: > Now I'm building for an embedded arm target. If there is a way of turning > remote-rdi off, I couldn't find it. It looks like it gets built by default > in gdb/configure.tgt(line 58) Anyway, the build dies in > gdb/rdi-share/unixcomm.c. SERPORT1 et. al. never get defined because we > aren't one of the architectures supported. -- Problem with weak functions http://sourceware.cygnus.com/ml/gdb/2000-05/msg00060.html Dan Nicolaescu writes: > It seems that gdb-4.95.1 does not display correctly the function when > stoping in weak functions. > > It stops in a function that is defined as weak, not in the function > that is actualy run... -- GDB 5.0 doesn't work on Linux/SPARC -- Code Cleanups: Next Release =========================== The following are small cleanups that will hopefully be completed by the follow on to 5.0. -- Delete macro TARGET_BYTE_ORDER_SELECTABLE. Patches in the database. -- Purge PARAMS. Eliminate all uses of PARAMS in GDB's source code. -- Fix copyright notices. Turns out that ``1998-2000'' isn't considered valid :-( http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00467.html -- Find something better than DEFAULT_BFD_ARCH, DEFAULT_BFD_VEC to determine the default isa/byte-order. -- Rely on BFD_BIG_ENDIAN and BFD_LITTLE_ENDIAN instead of host dependant BIG_ENDIAN and LITTLE_ENDIAN. -- Code Cleanups: General ====================== The following are more general cleanups and fixes. They are not tied to any specific release. -- Eliminate more compiler warnings. Of course there also needs to be the usual debate over which warnings are valid and how to best go about this. One method: choose a single option; get agreement that it is reasonable; try it out to see if there isn't anything silly about it (-Wunused-parameters is an example of that) then incrementally hack away. The other method is to enable all warnings and eliminate them from one file at a time. -- Elimination of ``(catch_errors_ftype *) func''. Like make_cleanup_func it isn't portable. http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00791.html http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00814.html -- Nuke USG define. -- [PATCH/5] src/intl/Makefile.in:distclean additions http://sourceware.cygnus.com/ml/gdb-patches/2000-04/msg00363.html Do not forget to merge the patch back into the trunk. -- Rationalize the host-endian code (grep for HOST_BYTE_ORDER). At present defs.h includes (which is linux specific) yet almost nothing depends on it. Suggest "gdb_endian.h" which can also handle and only include that where it is really needed. -- Replace asprintf() calls with xasprintf() calls. As with things like strdup() most calls to asprintf() don't check the return value. -- Replace strsave() + mstrsave() with libiberty:xstrdup(). -- Replace savestring() with something from libiberty. An xstrldup()? but that would have different semantics. -- Rationalize use of floatformat_unknown in GDB sources. Instead of defaulting to floatformat_unknown, should hosts/targets specify the value explicitly? http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00447.html -- Add a ``name'' member to include/floatformat.h:struct floatformat. Print that name in gdbarch.c. -- Sort out the harris mess in include/floatformat.h (it hardwires two different floating point formats). -- See of the GDB local floatformat_do_doublest() and libiberty's floatformat_to_double (which was once GDB's ...) can be merged some how. -- Eliminate mmalloc() from GDB. Also eliminate it from defs.h. -- Eliminate PTR. ISO-C allows ``void *''. -- Eliminate abort (). GDB should never abort. GDB should either throw ``error ()'' or ``internal_error ()''. Better still GDB should naturally unwind with an error status. -- GDB probably doesn't build on FreeBSD pre 2.2.x http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00378.html Fixes to get FreeBSD working on 2.2.x, 3.x and 4.x caused the code to suffer bit rot. -- Deprecate "fg". Apparently ``fg'' is actually continue. http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00417.html -- Deprecate current use of ``floatformat_unknown''. Require all targets to explicitly provide their float format instead of defaulting to floatformat unknown. Doing the latter leads to nasty bugs. http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00447.html -- Rationalize floatformat_to_double() vs floatformat_to_doublest(). Looks like GDB migrated floatformat_to_double() to libiberty but then turned around and created a ..._to_doublest() the latter containing several bug fixes. http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00472.html -- Move floatformat_ia64_ext to libiberty/include floatformat.[ch]. http://sourceware.cygnus.com/ml/gdb-patches/2000-05/msg00466.html -- Follow through `make check' with --enable-shared. When the srcware tree is configured with --enable-shared, the `expect' program won't run properly. Jim Wilson found out gdb has a local hack to set LD_LIBRARY_PATH, but, AFAIK, no other project has been hacked similarly. http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00845.html -- The ``maintenance deprecate set endian big'' command doesn't notice that it is deprecating ``set endian'' and not ``set endian big'' (big is implemented using an enum). Is anyone going to notice this? -- When tab expanding something like ``set arch'' ignore the deprecated ``set archdebug'' and expand to ``set architecture''. -- Eliminate ``arm_register_names[j] = (char *) regnames[j]'' and the like from arm-tdep.c. -- Fix uses of ->function.cfunc = set_function(). The command.c code calls sfunc() when a set command. Rather than change it suggest fixing the callback function so that it is more useful. See: http://sourceware.cygnus.com/ml/gdb-patches/2000-06/msg00062.html See also ``Fix implementation of ``target xxx''.'' below. -- IRIX 3.x support is probably broken. -- Delete sim/SIM_HAVE_BREAKPOINTS and gdb/SIM_HAS_BREAKPOINTS. http://sourceware.cygnus.com/ml/gdb-patches/2000-07/msg00042.html Apart from the d30v, are there any sim/common simulators that make use of this? A brief summary of what happended is that sim/common/sim-break.c was created as a good idea. It turned out a better idea was to use SIM_SIGBREAK and have GDB pass back sim_resume (..., SIGBREAK). -- parse.c:build_parse() has a buffer overrun. -- New Features and Fixes ====================== These are harder than cleanups but easier than work involving fundamental architectural change. -- Add built-by, build-date, tm, xm, nm and anything else into gdb binary so that you can see how the GDB was created. Some of these (*m.h) would be added to the generated config.h. That in turn would fix a long standing bug where by the build process many not notice a changed tm.h file. Since everything depends on config.h, a change to *m.h forces a change to config.h and, consequently forces a rebuild. -- Add an "info bfd" command that displays supported object formats, similarly to objdump -i. Is there a command already? -- Fix ``I'm sorry, Dave, I can't do that.'' from symfile.c. This requires internationalization. -- Convert GDB build process to AUTOMAKE. See also sub-directory configure below. The current convention is (kind of) to use $(
_h) in all dependency lists. It isn't done in a consistent way. -- Cleanup configury support for optional sub-directories. Check how GCC handles multiple front ends for an example of how things could work. A tentative first step is to rationalize things so that all sub directories are handled in a fashion similar to gdb/mi. See also automake above. -- Restructure gdb directory tree so that it avoids any 8.3 and 14 filename problems. -- Add a transcript mechanism to GDB. Such a mechanism might log all gdb input and output to a file in a form that would allow it to be replayed. It could involve ``gdb --transcript=FILE'' or it could involve ``(gdb) transcript file''. -- Can the xdep files be replaced by autoconf? -- Document trace machinery -- Document ui-out and ui-file. http://sourceware.cygnus.com/ml/gdb/2000-04/msg00121.html -- Update texinfo.tex to latest? -- Incorporate agentexpr.texi into gdb.texinfo agentexpr.texi mostly describes the details of the byte code used for tracepoints, not the internals of the support for this in GDB. So it looks like gdb.texinfo is a better place for this information. http://sourceware.cygnus.com/ml/gdb-patches/2000-04/msg00566.html -- Document overlay machinery. -- ``(gdb) catch signal SIGNAL'' Overlaps with ``handle SIGNAL'' but the implied behavour is different. You can attach commands to a catch but not a handle. A handle has a limited number of hardwired actions. -- Get the TUI working on all platforms. -- Add support for ``gdb --- PROGRAM ARGS ...''. Add support for ``gdb -cmd=...'' Along with many variations. Check: ????? for a full discussion. for a discussion. -- Implement ``(gdb) !ls''. Which is very different from ``(gdb) ! ls''. Implementing the latter is trivial. http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00034.html -- Change the (char *list[]) to (const char (*)[]) so that dynamic lists can be passed. -- When tab expanding something like ``set arch'' ignore the deprecated ``set archdebug'' and expand to ``set architecture''. -- Replace the code that uses the host FPU with an emulator of the target FPU. -- Thread Support ============== -- Generic: lin-thread cannot handle thread exit (Mark Kettenis, Michael Snyder) http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00525.html The thread_db assisted debugging code doesn't handle exiting threads properly, at least in combination with glibc 2.1.3 (the framework is there, just not the actual code). There are at least two problems that prevent this from working. As an additional reference point, the pre thread_db code did not work either. -- GNU/Linux/x86 and random thread signals (and Solaris/SPARC but not Solaris/x86). http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00336.html Christopher Blizzard writes: So, I've done some more digging into this and it looks like Jim Kingdon has reported this problem in the past: http://sourceware.cygnus.com/ml/bug-gdb/1999-10/msg00058.html I can reproduce this problem both with and without Tom's patch. Has anyone seen this before? Maybe have a solution for it hanging around? :) There's a test case for this documented at: when debugging threaded applications you get extra SIGTRAPs http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=9565 [There should be a GDB testcase - cagney] -- GDB5 TOT on unixware 7 http://sourceware.cygnus.com/ml/gdb/2000-04/msg00119.html Robert Lipe writes: > I just spun the top of tree of the GDB5 branch on UnixWare 7. As a > practical matter, the current thread support is somewhat more annoying > than when GDB was thread-unaware. -- Migrate qfThreadInfo packet -> qThreadInfo. (Andrew Cagney) Add support for packet enable/disable commands with these thread packets. General cleanup. [PATCH] Document the ThreadInfo remote protocol queries http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00832.html [PATCH] "info threads" queries for remote.c http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00831.html -- Language Support ================ New languages come onto the scene all the time. -- Pascal (Pierre Muller, David Taylor) Pierre Muller has contributed patches for adding Pascal Language support to GDB. 2 pascal language patches inserted in database http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00521.html Indent -gnu ? http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00496.html -- Java (Anthony Green, David Taylor) Anthony Green has a number of Java patches that did not make it into the 5.0 release. Patch: java tests http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00512.html Patch: java booleans http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00515.html Patch: handle N_MAIN stab http://sourceware.cygnus.com/ml/gdb-patches/2000-q1/msg00527.html -- [Comming...] Modify gdb to work correctly with Pascal. -- Re: Various C++ things value_headof/value_from_vtable_info are worthless, and should be removed. The one place in printcmd.c that uses it should use the RTTI functions. RTTI for g++ should be using the typeinfo functions rather than the vtables. The typeinfo functions are always at offset 4 from the beginning of the vtable, and are always right. The vtables will have weird names like E::VB sometimes. The typeinfo function will always be "E type_info function", or somesuch. value_virtual_fn_field needs to be fixed so there are no failures for virtual functions for C++ using g++. Testsuite cases are the major priority right now for C++ support, since i have to make a lot of changes that could potentially break each other. -- Add support for Modula3 Get DEC/Compaq to contribute their Modula-3 support. -- Remote Protocol Support ======================= -- set/show remote X-packet ... ``(gdb) help set remote X-packet'' doesn't list the applicable responses. The help message needs to be expanded. -- Remote protocol doco feedback. Too much feedback to mention needs to be merged in (901660). Search for the word ``remote''. http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00023.html http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00056.html http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00382.html -- GDB doesn't recover gracefully from remote protocol errors. GDB wasn't checking for NAKs from the remote target. Instead a NAK is ignored and a timeout is required before GDB retries. A pre-cursor to fixing this this is making GDB's remote protocol packet more robust. While downloading to a remote protocol target, gdb ignores packet errors in so far as it will continue to edownload with chunk N+1 even if chunk N was not correctly sent. This causes gdb.base/remote.exp to take a painfully long time to run. As a PS that test needs to be fixed so that it builds on 16 bit machines. -- Add the cycle step command. http://sourceware.cygnus.com/ml/gdb/2000-q1/msg00237.html -- Resolve how to scale things to support very large packets. -- Resolve how to handle a target that changes things like its endianess on the fly - should it be returned in the ``T'' packet? Underlying problem is that the register file is target endian. If the target endianess changes gdb doesn't know. -- Symbol Support ============== If / when GDB starts to support the debugging of multi-processor (rather than multi-thread) applications the symtab code will need to be updated a little so that several independant symbol tables are active at a given time. The other interesting change is a clarification of the exact meaning of CORE_ADDR and that has had consequences for a few targets (that were abusing that data type). -- Investiagate ways of reducing memory. -- Investigate ways of improving load time. -- Get the d10v to use POINTER_TO_ADDRESS and ADDRESS_TO_POINTER. Consequence of recent symtab clarification. No marks for figuring out who maintains the d10v. -- Get the MIPS to correctly sign extend all address <-> pointer conversions. Consequence of recent symtab clarification. No marks for figuring out who maintains the MIPS. -- GDB truncates 64 bit enums. http://sourceware.cygnus.com/ml/gdb-patches/2000-06/msg00290.html -- Testsuite Support ================= There are never to many testcases. -- Better thread testsuite. -- Better C++ testsuite. -- Look at adding a GDB specific testsuite directory so that white box tests of key internals can be added (eg ui_file). -- Separate out tests that involve the floating point (FP). (Something for people brining up new targets). FP and non-fp tests are combined. I think there should be set of basic tests that exercise pure integer support and then a more expanded set that exercise FP and FP/integer interactions. As an example, the MIPS, for n32 as problems with passing FP's and structs. Since most inferior call tests include FP it is difficult to determine of the integer tests are ok. -- Architectural Changes: General ============================== These are harder than simple cleanups / fixes and, consequently involve more work. Typically an Architectural Change will be broken down into a more digestible set of cleanups and fixes. -- Cleanup software single step. At present many targets implement software single step by directly blatting memory (see rs6000-tdep.c). Those targets should register the applicable breakpoints using the breakpoint framework. Perhaphs a new internal breakpoint class ``step'' is needed. -- Replace READ_FP() with FRAME_HANDLE(). READ_FP() is a hangover from the days of the vax when the ABI really did have a frame pointer register. Modern architectures typically construct a virtual frame-handle from the stack pointer and various other bits of string. Unfortunatly GDB still treats this synthetic FP register as though it is real. That in turn really confuses users (arm and ``print $fp'' VS ``info registers fp''). The synthetic FP should be separated out of the true register set presented to the user. -- Register Cache Cleanup (below from Andrew Cagney) I would depict the current register architecture as something like: High GDB --> Low GDB | | \|/ \|/ --- REG NR ----- | register + REGISTER_BYTE(reg_nr) | \|/ ------------------------- | extern register[] | ------------------------- where neither the high (valops.c et.al.) or low gdb (*-tdep.c) are really clear on what mechanisms they should be using to manipulate that buffer. Further, much code assumes, dangerously, that registers are contigious. Having got mips-tdep.c to support multiple ABIs, believe me, that is a bad assumption. Finally, that register cache layout is determined by the current remote/local target and _not_ the less specific target ISA. In fact, in many cases it is determined by the somewhat arbitrary layout of the [gG] packets! How I would like the register file to work is more like: High GDB | \|/ pseudo reg-nr | map pseudo <-> random cache bytes | \|/ ------------ | register | | cache | ------------ /|\ | map random cache bytes to target dependant i-face /|\ | target dependant such as [gG] packet or ptrace buffer The main objectives being: o a clear separation between the low level target and the high level GDB o a mechanism that solves the general problem of register aliases, overlaps etc instead of treating them as optional extras that can be wedged in as an after thought (that is a reasonable description of the current code). Identify then solve the hard case and the rest just falls out. GDB solved the easy case and then tried to ignore the real world :-) o a removal of the assumption that the mapping between the register cache and virtual registers is largely static. If you flip the USR/SSR stack register select bit in the status-register then the corresponding stack registers should reflect the change. o a mechanism that clearly separates the gdb internal register cache from any target (not architecture) dependant specifics such as [gG] packets. Of course, like anything, it sounds good in theory. In reality, it would have to contend with many<->many relationships at both the virt<->cache and cache<->target level. For instance: virt<->cache Modifying an mmx register may involve scattering values across both FP and mmpx specific parts of a buffer cache<->target When writing back a SP it may need to both be written to both SP and USP. Hmm, Rather than let this like the last time it was discussed, just slip, I'm first going to add this e-mail (+ references) to TODO. I'd then like to sketch out a broad strategy I think could get us there. First thing I'd suggest is separating out the ``extern registers[]'' code so that we can at least identify what is using it. At present things are scattered across many files. That way we can at least pretend that there is a cache instead of a global array :-) I'd then suggest someone putting up a proposal for the pseudo-reg / high-level side interface so that code can be adopted to it. For old code, initially a blanket rename of write_register_bytes() to deprecated_write_register_bytes() would help. Following that would, finaly be the corresponding changes to the target. -- Check that GDB can handle all BFD architectures (Andrew Cagney) There should be a test that checks that BFD/GDB are in sync with regard to architecture changes. Something like a test that first queries GDB for all supported architectures and then feeds each back to GDB.. Anyone interested in learning how to write tests? :-) -- Architectural Change: Multi-arch et al. ======================================= The long term objective is to remove all assumptions that there is a single target with a single address space with a single instruction set architecture and single application binary interface. This is an ongoing effort. The first milestone is to enable ``multi-arch'' where by all architectural decisions are made at runtime. It should be noted that ``gdbarch'' is really ``gdbabi'' and ``gdbisa''. Once things are multi-arched breaking that down correctly will become much easier. -- GDBARCH cleanup (Andrew Cagney) The non-generated parts of gdbarch.{sh,h,c} should be separated out into arch-utils.[hc]. Document that gdbarch_init_ftype could easily fail because it didn't identify an architecture. -- Fix BELIEVE_PPC_PROMOTION. Change it to BELIEVE_PPC_PROMOTION_P? At present there is still #ifdef BELIEVE_PPC_PROMOTION code in the symtab file. -- Fix target_signal_from_host() etc. The name is wrong for starters. ``target_signal'' should probably be ``gdb_signal''. ``from_host'' should be ``from_target_signal''. After that it needs to be multi-arched and made independant of any host signal numbering. -- Update ALPHA so that it uses ``struct frame_extra_info'' instead of EXTRA_FRAME_INFO. This is a barrier to replacing mips_extra_func_info with something that works with multi-arch. -- Multi-arch mips_extra_func_info. This first needs the alpha to be updated so that it uses ``struct frame_extra_info''. -- Rationalize TARGET_SINGLE_FORMAT and TARGET_SINGLE_BIT et al. Surely one of them is redundant. -- Convert ALL architectures to MULTI-ARCH. -- Select the initial multi-arch ISA / ABI based on --target or similar. At present the default is based on what ever is first in the BFD archures table. It should be determined based on the ``--target=...'' name. -- Make MIPS pure multi-arch. It is only at the multi-arch enabled stage. -- Truly multi-arch. Enable the code to recognize --enable-targets=.... like BINUTILS does. Can the tm.h and nm.h files be eliminated by multi-arch. -- Architectural Change: MI, LIBGDB and scripting languages ======================================================== See also architectural changes related to the event loop. LIBGDB can't be finished until there is a generic event loop being used by all targets. The long term objective is it to be possible to integrate GDB into scripting languages. -- Implement generic ``(gdb) commmand > file'' Once everything is going through ui_file it should be come fairly easy. http://sourceware.cygnus.com/ml/gdb/2000-04/msg00104.html -- Replace gdb_stdtarg with gdb_targout (and possibly gdb_targerr). gdb_stdtarg is easily confused with gdb_stdarg. -- Extra ui_file methods - dump. Very useful for whitebox testing. -- Eliminate error_begin(). With ui_file, there is no need for the statefull error_begin () function. -- Send normal output to gdb_stdout. Send error messages to gdb_stderror. Send debug and log output log gdb_stdlog. GDB still contains many cases where (f)printf or printf_filtered () is used when it should be sending the messages to gdb_stderror or gdb_stdlog. The thought of #defining printf to something has crossed peoples minds ;-) -- Re-do GDB's output pager. GDB's output pager still relies on people correctly using *_filtered for gdb_stdout and *_unfiltered for gdb_stdlog / gdb_stderr. Hopefully, with all normal output going to gdb_stdout, the pager can just look at the ui_file that the output is on and then use that to decide what to do about paging. Sounds good in theory. -- Check/cleanup MI documentation. The list of commands specified in the documentation needs to be checked against the mi-cmds.c table in a mechanical way (so that they two can be kept up-to-date). -- Convert MI into libgdb MI provides a text interface into what should be many of the libgdb functions. The implementation of those functions should be separated into the MI interface and the functions proper. Those functions being moved to gdb/lib say. -- Create libgdb.h The first part can already be found in defs.h. -- MI's input does not use buffering. At present the MI interface reads raw characters of from an unbuffered FD. This is to avoid several nasty buffer/race conditions. That code should be changed so that it registers its self with the event loop (on the input FD) and then push commands up to MI as they arrive. The serial code already does this. -- Make MI interface accessable from existing CLI. -- Add a breakpoint-edit command to MI. It would be similar to MI's breakpoint create but would apply to an existing breakpoint. It saves the need to delete/create breakpoints when ever they are changed. -- Add directory path to MI breakpoint. That way the GUI's task of finding the file within which the breakpoint was set is simplified. -- Add a mechanism to reject certain expression classes to MI There are situtations where you don't want GDB's expression parser/evaluator to perform inferior function calls or variable assignments. A way of restricting the expression parser so that such operations are not accepted would be very helpful. -- Remove sideffects from libgdb breakpoint create function. The user can use the CLI to create a breakpoint with partial information - no file (gdb would use the file from the last breakpoint). The libgdb interface currently affects that environment which can lead to confusion when a user is setting breakpoints via both the MI and the CLI. This is also a good example of how getting the CLI ``right'' will be hard. -- Move gdb_lasterr to ui_out? The way GDB throws errors and records them needs a re-think. ui_out handles the correct output well. It doesn't resolve what to do with output / error-messages when things go wrong. -- do_setshow_command contains a 1024 byte buffer. The function assumes that there will never be any more than 1024 bytes of enum. It should use mem_file. -- Should struct cmd_list_element . completer take the command as an argument? -- Should the bulk of top.c:line_completion_function() be moved to command.[hc]? complete_on_cmdlist() and complete_on_enums() could then be made private. -- top.c (execute_command): Should a command being valid when the target is running be made an attribute (predicate) to the command rather than an explicit set of tests. -- top.c (execute_command): Should the bulk of this function be moved into command.[hc] so that top.c doesn't grub around in the command internals? -- Architectural Change: Async =========================== While GDB uses an event loop when prompting the user for input. That event loop is not exploited by targets when they allow the target program to continue. Typically targets still block in (target_wait()) until the program again halts. The closest a target comes to supporting full asynchronous mode are the remote targets ``async'' and ``extended-async''. -- Asynchronous expression evaluator Inferior function calls hang GDB. -- Fix implementation of ``target xxx''. At present when the user specifies ``target xxxx'', the CLI maps that directly onto a target open method. It is then assumed that the target open method should do all sorts of complicated things as this is the only chance it has. Check how the various remote targets duplicate the target operations. Check also how the various targets behave differently for purely arbitrary reasons. What should happen is that ``target xxxx'' should call a generic ``target'' function and that should then co-ordinate the opening of ``xxxx''. This becomes especially important when you're trying to open an asynchronous target that may need to perform background tasks as part of the ``attach'' phase. Unfortunatly, due to limitations in the old/creaking command.h interface, that isn't possible. The function being called isn't told of the ``xxx'' or any other context information. Consequently a precursor to fixing ``target xxxx'' is to clean up the CLI code so that it passes to the callback function (attatched to a command) useful information such as the actual command and a context for that command. Other changes such as making ``struct command'' opaque may also help. See also: http://sourceware.cygnus.com/ml/gdb-patches/2000-06/msg00062.html -- Make "target xxx" command interruptible. As things become async this becomes possible. A target would start the connect and then return control to the event loop. A cntrl-c would notify the target that the operation is to be abandoned and the target code could respond. -- Add a "suspend" subcommand of the "continue" command to suspend gdb while continuing execution of the subprocess. Useful when you are debugging servers and you want to dodge out and initiate a connection to a server running under gdb. [hey async!!] -- TODO FAQ ======== Frequently requested but not approved requests. -- Eliminate unused argument warnings using ATTRIBUTE_UNUSED. The benefits on this one are thought to be marginal - GDBs design means that unused parameters are very common. GCC 3.0 will also include the option -Wno-unused-parameter which means that ``-Wall -Wno-unused-parameters -Werror'' can be specified. -- Legacy Wish List ================ This list is not up to date, and opinions vary about the importance or even desirability of some of the items. If you do fix something, it always pays to check the below. -- @c This does not work (yet if ever). FIXME. @c @item --parse=@var{lang} @dots{} @c Configure the @value{GDBN} expression parser to parse the listed languages. @c @samp{all} configures @value{GDBN} for all supported languages. To get a @c list of all supported languages, omit the argument. Without this @c option, @value{GDBN} is configured to parse all supported languages. -- START_INFERIOR_TRAPS_EXPECTED need never be defined to 2, since that is its default value. Clean this up. -- It should be possible to use symbols from shared libraries before we know exactly where the libraries will be loaded. E.g. "b perror" before running the program. This could maybe be done as an extension of the "breakpoint re-evaluation" after new symbols are loaded. -- Make single_step() insert and remove breakpoints in one operation. [If this is talking about having single_step() insert the breakpoints, run the target then pull the breakpoints then it is wrong. The function has to return as control has to eventually be passed back to the main event loop.] -- Speed up single stepping by avoiding extraneous ptrace calls. -- Speed up single stepping by not inserting and removing breakpoints each time the inferior starts and stops. Breakpoints should not be inserted and deleted all the time. Only the one(s) there should be removed when we have to step over one. Support breakpoints that don't have to be removed to step over them. [this has resulted in numerous debates. The issue isn't clear cut] -- Provide "voodoo" debugging of core files. This creates a zombie process as a child of the debugger, and loads it up with the data, stack, and regs of the core file. This allows you to call functions in the executable, to manipulate the data in the core file. [you wish] -- GDB reopens the source file on every line, as you "next" through it. [still true? I've a memory of this being fixed] -- Perhaps "i source" should take an argument like that of "list". -- Remove "at 0xnnnn" from the "b foo" response, if `print address off' and if it matches the source line indicated. -- The prompt at end of screen should accept space as well as CR. -- Backtrace should point out what the currently selected frame is, in its display, perhaps showing "@3 foo (bar, ...)" or ">3 foo (bar, ...)" rather than "#3 foo (bar, ...)". -- "i program" should work for core files, and display more info, like what actually caused it to die. -- "x/10i" should shorten the long name, if any, on subsequent lines. -- "next" over a function that longjumps, never stops until next time you happen to get to that spot by accident. E.g. "n" over execute_command which has an error. -- "set zeroprint off", don't bother printing members of structs which are entirely zero. Useful for those big structs with few useful members. -- GDB does four ioctl's for every command, probably switching terminal modes to/from inferior or for readline or something. -- terminal_ours versus terminal_inferior: cache state. Switch should be a noop if the state is the same, too. -- "i frame" shows wrong "arglist at" location, doesn't show where the args should be found, only their actual values. -- There should be a way for "set" commands to validate the new setting before it takes effect. -- "ena d" is ambiguous, why? "ena delete" seems to think it is a command! -- i line VAR produces "Line number not known for symbol ``var''.". I thought we were stashing that info now! -- We should be able to write to random files at hex offsets like adb. -- [elena - delete this] Handle add_file with separate text, data, and bss addresses. Maybe handle separate addresses for each segment in the object file? -- [Jimb/Elena delete this one] Handle free_named_symtab to cope with multiply-loaded object files in a dynamic linking environment. Should remember the last copy loaded, but not get too snowed if it finds references to the older copy. -- [elena delete this also] Remove all references to: text_offset data_offset text_data_start text_end exec_data_offset ... now that we have BFD. All remaining are in machine dependent files. -- Re-organize help categories into things that tend to fit on a screen and hang together. -- Add in commands like ADB's for searching for patterns, etc. We should be able to examine and patch raw unsymboled binaries as well in gdb as we can in adb. (E.g. increase the timeout in /bin/login without source). [actually, add ADB interface :-] -- When doing "step" or "next", if a few lines of source are skipped between the previous line and the current one, print those lines, not just the last line of a multiline statement. -- Handling of "&" address-of operator needs some serious overhaul for ANSI C and consistency on arrays and functions. For "float point[15];": ptype &point[4] ==> Attempt to take address of non-lvalue. For "char *malloc();": ptype malloc ==> "char *()"; should be same as ptype &malloc ==> "char *(*)()" call printf ("%x\n", malloc) ==> weird value, should be same as call printf ("%x\n", &malloc) ==> correct value -- Fix dbxread.c symbol reading in the presence of interrupts. It currently leaves a cleanup to blow away the entire symbol table when a QUIT occurs. (What's wrong with that? -kingdon, 28 Oct 1993). [I suspect that the grype was that, on a slow system, you might want to cntrl-c and get just half the symbols and then load the rest later - scary to be honest] -- Mipsread.c reads include files depth-first, because the dependencies in the psymtabs are way too inclusive (it seems to me). Figure out what really depends on what, to avoid recursing 20 or 30 times while reading real symtabs. -- value_add() should be subtracting the lower bound of arrays, if known, and possibly checking against the upper bound for error reporting. -- When listing source lines, check for a preceding \n, to verify that the file hasn't changed out from under us. [fixed by some other means I think. That hack wouldn't actually work reliably - the file might move such that another \n appears. ] -- Get all the remote systems (where the protocol allows it) to be able to stop the remote system when the GDB user types ^C (like remote.c does). For ebmon, use ^Ak. -- Possible feature: A version of the "disassemble" command which shows both source and assembly code ("set symbol-filename on" is a partial solution). [has this been done? It was certainly done for MI and GDBtk] -- investigate "x/s 0" (right now stops early) (I think maybe GDB is using a 0 address for bad purposes internally). -- Make "info path" and path_command work again (but independent of the environment either of gdb or that we'll pass to the inferior). -- Make GDB understand the GCC feature for putting octal constants in enums. Make it so overflow on an enum constant does not error_type the whole type. Allow arbitrarily large enums with type attributes. Put all this stuff in the testsuite. -- Make TYPE_CODE_ERROR with a non-zero TYPE_LENGTH more useful (print the value in hex; process type attributes). Add this to the testsuite. This way future compilers can add new types and old versions of GDB can do something halfway reasonable. -- Fix mdebugread.c:parse_type to do fundamental types right (see rs6000_builtin_type in stabsread.c for what "right" is--the point is that the debug format fixes the sizes of these things and it shouldn't depend on stuff like TARGET_PTR_BIT and so on. For mdebug, there seem to be separate bt* codes for 64 bit and 32 bit things, and GDB should be aware of that). Also use a switch statement for clarity and speed. -- Investigate adding symbols in target_load--some targets do, some don't. -- Put dirname in psymtabs and change lookup*symtab to use dirname (so /foo/bar.c works whether compiled by cc /foo/bar.c, or cd /foo; cc bar.c). -- Merge xcoffread.c and coffread.c. Use breakpoint_re_set instead of fixup_breakpoints. -- Make a watchpoint which contains a function call an error (it is broken now, making it work is probably not worth the effort). -- New test case based on weird.exp but in which type numbers are not renumbered (thus multiply defining a type). This currently causes an infinite loop on "p v_comb". -- [Hey! Hint Hint Delete Delete!!!] Fix 386 floating point so that floating point registers are real registers (but code can deal at run-time if they are missing, like mips and 68k). This would clean up "info float" and related stuff. -- gcc -g -c enummask.c then gdb enummask.o, then "p v". GDB complains about not being able to access memory location 0. -------------------- enummask.c enum mask { ANIMAL = 0, VEGETABLE = 1, MINERAL = 2, BASIC_CATEGORY = 3, WHITE = 0, BLUE = 4, GREEN = 8, BLACK = 0xc, COLOR = 0xc, ALIVE = 0x10, LARGE = 0x20 } v; -- If try to modify value in file with "set write off" should give appropriate error not "cannot access memory at address 0x65e0". -- Allow core file without exec file on RS/6000. -- Make sure "shell" with no arguments works right on DOS. -- Make gdb.ini (as well as .gdbinit) be checked on all platforms, so the same directory can be NFS-mounted on unix or DOS, and work the same way. -- [Is this another delete???] Get SECT_OFF_TEXT stuff out of objfile_relocate (might be needed to get RS/6000 to work right, might not be immediately relevant). -- Work out some kind of way to allow running the inferior to be done as a sub-execution of, eg. breakpoint command lists. Currently running the inferior interupts any command list execution. This would require some rewriting of wait_for_inferior & friends, and hence should probably be done in concert with the above. -- Add function arguments to gdb user defined functions. -- Add convenience variables that refer to exec file, symbol file, selected frame source file, selected frame function, selected frame line number, etc. -- Modify the handling of symbols grouped through BINCL/EINCL stabs to allocate a partial symtab for each BINCL/EINCL grouping. This will seriously decrease the size of inter-psymtab dependencies and hence lessen the amount that needs to be read in when a new source file is accessed. -- Add a command for searching memory, a la adb. It specifies size, mask, value, start address. ADB searches until it finds it or hits an error (or is interrupted). -- Remove the range and type checking code and documentation, if not going to implement. # Local Variables: # mode: text # End: