2002-03-21 Klee Dienes <klee@apple.com> * nextstep-nat-dyld.c (dyld_info_process_raw): The issue is that when dyld goes to process DYLD_INSERT_LIBRARIES, it removes the colons from the DYLD_INSERT_LIBRARIES string while parsing, loads the libraries, then replaces the colons. The problem here is that when GDB goes to read the namestring out of the inferior memory, the colons have been replaced, and therefore GDB ends up with multiple strings instead of just the one. I've changed GDB to treat ':' as a string terminator when processing all shared library pathnames returned by DYLD. I'm not sure this is the ideal long-term solution though, since I suppose some day some masochist will try to debug a program that uses a shared library with a ':' in the name. 2002-03-21 Klee Dienes <klee@apple.com> * nextstep-nat-inferior.c (next_handle_exception): Check incoming exceptions --- if the task receiving the exception is not the same as the one being debugged by GDB, it resets the exception ports of the task back to the ones GDB saved for the target when it first started debugging; then resumes the task. This isn't quite perfect, since ideally GDB would allow changes to the exception ports made by the target be reflected in children of the target as well; and also because it relies on being able to repeat an exception by simply continuing the child without modifying the program state (I think this should always work, but there may be cases I haven't thought of). (next_wait): If control returns due to an event of type WAITKIND_SPURIOUS, continue waiting for a real event. 2002-03-19 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-process.c: rearrange the include order to remove a few compiler warnings. * ppc-tdep.c (ppc_fast_show_stack_helper): Return 1 when you find only 1 frame, not 0 dummy... * ppc-frameinfo.c (ppc_parse_instructions): Fix a bug with handling the saveFP builtin function in prologues. 2002-03-14 James Ingham <jingham@inghji.apple.com> * ppc-tdep.c (ppc_fast_show_stack_helper): Use lookup_minimal_symbol to find _sigtramp, it is much more efficient than parse_and_eval_address... 2002-03-05 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld.c (dyld_starts_here_p): New function, does the memory passed in actually contain dyld? (next_locate_dyld): New function, find where dyld got loaded in this executable. (next_set_start_breakpoint): call next_locate_dyld. (dyld_lookup_address): use the slide set by next_locate_dyld. * nextstep-nat-cfm.h: add cfm_update to silence compiler warnings. * nextstep-nat-dyld-info.h: add dyld_objfile_info_compare. * nextstep-nat-dyld-process.h: add dyld_parse_load_level and dyld_minimal_load_flag. * nextstep-nat-dyld.h: remove next_mach_start_dyld - it is never used or defined. 2002-02-26 James Ingham <jingham@inghji.apple.com> * nextstep-nat-sigthread.c (next_signal_thread): Don't call pause in between pthread_testcancel calls, we never get a signal to wake us up... * nextstep-nat-inferior.c (next_clear_pending_events): New function, clears out the pending event chain. (next_mourn_inferior): Call next_clear_pending_events. 2002-02-21 James Ingham <jingham@inghji.apple.com> * remote-kdp.c (kdp_resume): Fix signature of kdp_resume - should be a ptid_t for first argument, not an int... 2002-02-06 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-process.c (dyld_libraries_compatible): The prefix may be NULL, check this before comparing the contents. (dyld_objfile_move_load_data): copy the prefix over as well. * nextstep-nat-dyld-info.c (dyld_objfile_entry_clear): The unset value for the prefix should be NULL, not "". 2002-01-25 James Ingham <jingham@inghji.apple.com> Create a queue of "pending events" to handle the case where we get more than one event from either Mach or BSD when the inferior stops. If we get more than one event, then we report them to the user serially, and each call to next_child_resume will fetch the next pending event till the queue is exhausted. Only then do we really resume the inferior. * nextstep-nat-inferior-stubs.c: Fix child_wait to pass the client_data field. * nextstep-nat-inferior-util.c (next_inferior_suspend_ptrace): Fix call to next_wait. * nextstep-nat-inferior.h: Change declaration of next_wait. * nextstep-nat-inferior.c: Include regcache.h to remove compiler warnings. Generalize the signal & exception handling functions so we can share them between the immediate event handling and the deferred servicing routines. Add the next_pending_event struct, queue, and handling functions. * kdp.c: change declaration of kdp_wait, and since the gdb_queue_event function is now exported, remove the copy of it from here. 2002-01-18 James Ingham <jingham@inghji.apple.com> * i386-next-nat-exec.c (fetch_inferior_registers): Fix a reference to the obsolete inferior_pid. (store_inferior_registers): ditto 2002-01-11 James Ingham <jingham@inghji.apple.com> * ppc-tdep.c (ppc_breakpoint_from_pc): Fixed a couple of places where we needed to switch BIG_ENDIAN for BFD_BIG_ENDIAN. 2001-12-07 James Ingham <jingham@inghji.apple.com> (From Bernard Semeria) * remote-kdp.c (kdp_attach): Use the version packet rather than hijacking the region packet to get the kdp version. * kdp-protocol.c (kdp_marshall, kdp_unmarshall): ditto. * kdp-protocol.h: ditto. 2001-12-03 James Ingham <jingham@inghji.apple.com> * nextstep-nat-inferior.c: Format changes, more functions comply with the GNU coding conventions. Also added a few prototypes, and extern'ed some functions from exec.c to remove compiler warnings. * nextstep-nat-inferior.c (next_set_auto_start_dyld): New function, used to dynamically switch the dyld breakpoint on and off. (_initialize_next_inferior): use next_set_auto_start_dyld. (next_process_events): drain both event sources here, not just the one you got the first hit on. We have more chance of handling extraneous events here than in next_child_resume... 2001-11-01 James Ingham <jingham@inghji.apple.com> * ppc-tdep.c (ppc_fast_show_stack_helper): Move most of the functionality into this function, and set it as the FAST_COUNT_STACK_DEPTH function, which is used in mi_cmd_info_stack_depth. * tm-ppc.h: Add define for FAST_COUNT_STACK_DEPTH 2001-10-31 James Ingham <jingham@inghji.apple.com> * ppc-tdep.c (ppc_fast_show_stack): New function. Run up the stack and return pc & fp for each frame, and a frame count. Since this doesn't build up the frame cache, it is much faster. (ppc_frame_find_pc): We were looking in the wrong place for the frame stored PC for signal frames. 2001-10-30 Klee Dienes <kdienes@apple.com> * nextstep-nat-dyld-process.c (dyld_add_image_libraries): recognize BFD_MACH_O_LC_LOAD_WEAK_DYLIB; treat it like LC_LOAD_DYLIB for now. 2001-10-26 Klee Dienes <kdienes@apple.com> * nextstep-nat-dyld-info.c (dyld_entry_out): Send the basename as the "path" field when talking via MI. The ProjectBuilder shared library dialog uses the pathname as the key for the table, so all entries without a "path" field were getting collapsed into one line in the table. The correct fix for this is to have PB used a different algorithm for building the table, but this fix should work well for now. * nextstep-nat-dyld-process.c (dyld_load_library): Automatically load CFM fragments from memory, rather than trying to load from disk first. The only time we should be relying on the fragment name to match a disk file is when looking for xSym files. Trying to read from disk was taking extra time in the cases where the disk access failed, and the warning messages when it failed were distracting. * nextstep-nat-dyld-.c (inferior_bfd_generic): Handle the case where a memory object is provided without a text string for the name (previously it would crash). 2001-10-23 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld.c (next_dyld_update): If you have turned off cfm with the inferior-auto-start-cfm variable, then don't call update-cfm when you update the libraries. 2001-10-17 James Ingham <jingham@inghji.apple.com> * ppc-tdep.c (ppc_frame_find_pc): Actually return the signal frame value. Dopey bug... (ppc_frame_saved_pc_after_call): remove spurious warning. This is always supposed to return the link register, and looking for the frame saved value is not relevant. (ppc_frame_chain_valid): Guard the call that reads the stack from errors. Callers don't expect ppc_frame_chain_valid to error out, but it can if the stack is mauled so that the fp is an unreadable address. * ppc-frameinfo.h: define safe_read_memory_unsigned_integer, so that ppc-tdep can use is. * ppc-frameinfo.c (ppc_parse_instructions): Don't assume that the pic base will use register 31. The dyld stub code uses 11, for instance. 2001-10-02 Jim Ingham (jingham@apple.com) and Jason Molenda (jmolenda@apple.com) * nextstep-nat-inferior.c (next_fetch_event): Accept a third type of timeout: Check for an event to read, but don't block if there's nothing there. (next_process_events, next_child_resume): Update callers. 2001-08-06 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (safe_read_memory_unsigned_integer): The wrapped function has to return a value, or bad things happen... (ppc_frame_cache_properties): pass the function_end as the stop for the parsing, not the pc. This way we can correctly parse the prologue & figure out where we are EVEN IF the pc is in the prologue. 2001-08-03 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-info.c (dyld_entry_info): Check for e->loaded_name not null before passing to strrchr. 2001-08-02 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld.c (_initialize_nextstep_nat_dyld): fix up doc strings. (next_init_dyld): Add a missing check for the ofile being NULL. 2001-07-31 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-process.c: The dyld code doesn't own the objfile for the executable. The generic gdb code keeps a pointer to this through the symfile_objfile pointer. So let the generic code handle the lifetime of the executable objfile. 2001-07-27 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (safe_read_memory_unsigned_integer): New function, safe version of read_memory_unsigned_int. (wrap_read_memory_unsigned_integer): This is the wrapper function. (ppc_parse_instructions): catch errors from read_memory_unsigned_integer, this gives us a much better chance of cleaning up from cases where the pc is off in unreadable memory. 2001-06-21 Klee Dienes <kdienes@localhost> * nextstep-nat-dyld.c (section_info_helper): protect the CFM section of the code with #if WITH_CFM . 2001-06-14 James Ingham <jingham@gdbrulez.apple.com> * ppc-frameinfo.c (ppc_parse_instructions): properly sign extend the instruction offset from the "bl" instruction - the old way didn't work for CORE_ADDR = unsigned long long. * nextstep-nat-dyld-path.c (dyld_resolve_image): check for name being null and don't crash. * nextstep-nat-dyld-process.c (dyld_resolve_load_flag): Ditto... 2001-06-07 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (ppc_frame_cache_properties): protect against the case where we think the lr wasn't saved in a non-leaf function. * ppc-tdep.c (ppc_frame_find_pc): ditto. 2001-05-18 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (ppc_parse_instructions): Add a fallback if we don't recognize ANY valid prologue instructions. In that case, we are just guessing, but guess that the pc is stored on the stack, since this is usually the case. 2001-05-15 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-info.c (dyld_print_shlib_info): use dyld_entry_out to output the path, offset & slide in ui-out'ified form. (dyld_entry_string): Use dyld_entry_info. (dyld_entry_out): New function, format the path, offset & slide in ui-out. (dyld_entry_info): Generic function, returns path, offset & slide in separate fields, for use in dyld_entry_string & dyld_entry_out. 2001-05-10 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld.c (dyld_print_status_info): ui_out'ify the info messages so we don't spam the mi... * nextstep-nat-dyld-info.c (dyld_print_shlib_info): ui_out'ify the output from info sharedlibrary. 2001-04-18 Jim Ingham <jingham@localhost.apple.com> * ppc-tdep.c (ppc_get_unsaved_pc): New function. How to get the frame's pc if it hasn't been saved in the frame. (ppc_frame_find_pc): Reworked the logic to better handle cases where the of the parent frame is in its prologue. (ppc_frame_chain_valid): If the frame pointer is null, we are at the end of the stack. Re-enable the code that does this check. Klee thinks this might cause trouble somewhere but can't remember where, and it is the right thing to do, so... * ppc-frameinfo.h: Removed the offset2 field, it was just for use IN ppc_parse_instructions. Added lr_invalid to handle prologues that temporarily invalidate the link register. * ppc-frameinfo.c (ppc_print_properties): mutatis mutandi for changes to the properties structure. (ppc_parse_instructions): Add support for the saveFP and saveVec gcc prolog setup routines. Also record WHEN the link register is stored in the frame so we can do a better job of backtraces while in the prolog. (ppc_clear_function_properties): mutatis mutandi for changes to the properties structure. 2001-02-21 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-process.c (dyld_load_library): If you get an error in getting the TEXT segment for a shlib, mark the entry as invalid. You can't free it, it wasn't malloc'ed... 2001-01-24 James Ingham <jingham@inghji.apple.com> * nextstep-nat-inferior.c (next_fetch_task_info): Comment out the call to next_init_dyld_symfile. This caused gdb to reset shared library info when quitting. (next_child_stop): New function: don't use child_stop, it sends a kill (-pid, SIGINT) which doesn't work on MacOS X unless the inferior actually IS a process group leader. But apps in MacOS X almost never are (they are children of the WindowServer). initialize remote_async_terminal_ours_p to 1 - this reflects truth: gdb at startup DOES own the terminal. * remote-kdp.c: ditto. 2001-01-22 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (ppc_frame_cache_properties): Check that the frame's extra_info is set. 2000-11-07 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (ppc_parse_instructions): If the FIRST instruction of the prologue is a bl, we are probably NOT in the prologue. * ppc-tdep.c (ppc_frame_find_pc): In the case where we have no symbols, use the instruction parser to tell us whether the return address has been put on the stack yet or not. 2000-09-06 Jim Ingham <jingham@localhost> * nextstep-nat-inferior-util.c: add WITH_CFM guards in the appropriate places. * nextstep-nat-inferior.c: ditto * nextstep-nat-inferior.h: ditto. 2000-09-05 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.c (ppc_parse_instructions): Cleanup a comment & remove an unused variable. 2000-09-01 James Ingham <jingham@inghji.apple.com> * ppc-frameinfo.h: Added a few comments. * ppc-frameinfo.c (ppc_parse_instructions): Detect the "save_world" stack setup routine that gcc uses for C++, and set up the props structure based on what it does. NB. This will break if gcc changes this routine, but there's not much I can do about this... 2000-08-09 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-path.c (build_suffix_name): new function, find full name including the DYLD_IMAGE_SUFFIX. (search_for_name_in_path): Pass in the suffix that comes from DYLD_IMAGE_SUFFIX, and use in search. (get_framework_pathname): pass in flag telling whether to use the image suffix, and use it. (dyld_resolve_image): use the image suffix in resolving the image path. (dyld_init_paths): Initialize the image_suffix from the environment. * nextstep-nat-dyld-path.c: Add image_suffix to dyld_path_info struct. 2000-07-20 James Ingham <jingham@inghji.apple.com> * nextstep-nat-inferior-debug.c: Add fflush to the inferior_stderr. There may be other places where this is needed as well, but we will chase them down as we find them. * nextstep-nat-inferior.c: Open the inferior_stderr "w", not "w+", so the debugging output will work when gdb is running as pipe under another program. 2000-06-23 James Ingham <jingham@inghji.apple.com> * nextstep-nat-dyld-process.c: Take out the call to build_objfile_section_table, since it only screwed up the section table, and had no other purpose...