and stack pointer during stepping, to speed things up.
A. Changes to not select a frame until we need a selected frame:
* blockframe.c (flush_cached_frames): Call select_frame (NULL, -1).
* infrun.c (wait_for_inferior): Move call to select_frame back to
normal_stop. This reverts a change of 13 Apr 94 (it says Jeff
Law, but the change was my idea); the only reason for that change
was so we could save and restore the selected frame in
wait_for_inferior, and now that flush_cached frames clears the
selected frame, that should work OK now.
B. Changes to not create a current_frame until we need one:
* blockframe.c (get_current_frame): If current_frame is NULL, try
to create an innermost frame.
* sparc-tdep.c (sparc_pop_frame), infcmd.c (run-stack_dummy),
infrun.c (wait_for_inferior), thread.c (thread_switch),
convex-tdep.c (set_thread_command), a29k-tdep.c (pop_frame),
alpha-tdep.c (alpha_pop_frame), convex-xdep.c (core_file_command),
h8300-tdep.c (h8300_pop_frame), h8500-tdep.c (h8300_pop_frame),
hppa-tdep.c (hppa_pop_frame), i386-tdep.c (i386_pop_frame),
i960-tdep.c (pop_frame), m68k-tdep.c
(m68k_pop_frame), mips-tdep.c (mips_pop_frame), rs6000-tdep.c
(push_dummy_frame, pop_dummy_frame, pop_frame), sh-tdep.c
(pop_frame), config/arm/tm-arm.h (POP_FRAME),
config/convex/tm-convex.h (POP_FRAME), config/gould/tm-pn.h
(POP_FRAME), config/ns32k/tm-merlin.h (POP_FRAME),
config/ns32k/tm-umax.h (POP_FRAME), config/tahoe/tm-tahoe.h
(POP_FRAME), config/vax/tm-vax.h (POP_FRAME): Don't
call create_new_frame.
* corelow.c (core_open), altos-xdep.c (core_file_command),
arm-xdep.c (core_file_command), gould-xdep.c (core_file_command),
m3-nat.c (select_thread), sun386-nat.c (core_file_command),
umax-xdep.c (core_file_command): Don't call create_new_frame; do
call flush_cached_frames.
* blockframe.c (reinit_frame_cache): Don't call create_new_frame
or select_frame.
C. Changes to get rid of stop_frame_address and instead only
fetch the frame pointer when we need it.
* breakpoint.c (bpstat_stop_status): Remove argument
frame_address; use FRAME_FP (get_current_frame ()).
* infrun.c (wait_for_inferior): Don't pass frame pointer to
bpstat_stop_status.
* infrun.c (wait_for_inferior): Use FRAME_FP (get_current_frame
()) instead of stop_frame_address.
* infrun.c (save_inferior_status, restore_inferior_status),
inferior.h (struct inferior_status): Don't save and restore
stop_frame_address.
* inferior.h, infcmd.c, thread.c (thread_switch), m3-nat.c
(select_thread): Remove stop_frame_address and uses thereof.
D. Same thing for the stack pointer.
* infrun.c (wait_for_inferior): Remove stop_sp and replace
uses thereof with read_sp ().
E. Change to eliminate one nasty little spot where we were
wanting to know the frame pointer from before the current step
(idea from GDB 3.5, which saved my ass, because my other ideas of
how to fix it were very baroque).
* infrun.c: Remove prev_frame_address.
* infrun.c (wait_for_inferior, step_over_function): Use
step_frame_address instead of prev_frame_address.
F. Same basic idea for the stack pointer.
* inferior.h, infcmd.c: New variable step_sp.
* infcmd.c (step_1, until_next_command): Set it.
* infrun.c: Remove prev_sp and replace uses by step_sp.
* infrun.c (wait_for_inferior): If we get out of the step
range, then set step_sp to the current stack pointer before we
start going again.
Change all references to stdout/stderr to gdb_stdout/gdb_stderr.
Replace all calls to stdio output functions with calls to
corresponding _unfiltered functions (`fprintf_unfiltered')
Replaced calls to fopen for output to gdb_fopen.
Added sufficient goo to utils.c and defs.h to make the above work.
The net effect is that stdio output functions are only directly used
in utils.c. Elsewhere, the _unfiltered and _filtered functions and
GDB_FILE type are used.
In the near future, GDB_FILE will stop being equivalant to FILE.
The semantics of some commands has changed in a very subtle way:
called in the right context, they may cause new occurences of
prompt_for_continue() behavior. The testsuite doesn't notice anything
like this, though.
Please respect this change by not reintroducing stdio output
dependencies in the main body of gdb code. All output from commands
should go to a GDB_FILE.
Target-specific code can still use stdio directly to communicate with
targets.
* All GDB files that #include defs.h: Removed stdio.h.
(defs.h): #include stdio.h.
This has been tested by building GDBs for all targets hosted on Sun4.
None of the build problems were related to stdio.h inclusion. (n.b.
many configurations don't build for other reasons.)
