Doug Evans committed the change above, but forgot to update one of
the callers of get_raw_print_options, at mt-tdep.c. This commit fixes
this.
2013-07-10 Sergio Durigan Junior <sergiodj@redhat.com>
* mt-tdep.c (mt_registers_info): Call
get_no_prettyformat_print_options instead of
get_raw_print_options (regression by last patch from Doug
Evans).
This patch adds some tests that evidence a regression fixed by
the following patch from Pedro Alves:
[PATCH] ada-lang.c:coerce_unspec_val_to_type: Preserve laziness
http://www.sourceware.org/ml/gdb-patches/2013-07/msg00178.html
gdb/testsuite/ChangeLog:
* gdb.ada/arrayptr/foo.adb: Add some code defining an access
to a packed array.
* gdb.ada/arrayptr.exp: Add a few tests using that new access
to packed array.
ada-lang.c:coerce_unspec_val_to_type does:
if (value_lazy (val)
|| TYPE_LENGTH (type) > TYPE_LENGTH (value_type (val)))
result = allocate_value_lazy (type);
else
{
result = allocate_value (type);
memcpy (value_contents_raw (result), value_contents (val),
TYPE_LENGTH (type));
}
set_value_component_location (result, val);
set_value_bitsize (result, value_bitsize (val));
set_value_bitpos (result, value_bitpos (val));
set_value_address (result, value_address (val));
set_value_optimized_out (result, value_optimized_out (val));
Notice that before value_optimized_out was made to auto-fetch lazy
values, VAL would end up still lazy if it was lazy on entry. It's not
really a problem here if VAL is lazy, and VAL->optimized_out is 0,
because RESULT is also left lazy. IOW, this just wants to copy the
VAL->optimized_out flag to RESULT->optimized_out, nothing else.
As a side-effect of the change in value_optimized_out, the following
testcase now regresses. Consider:
type Small is range -64 .. 63;
for Small'Size use 7;
type Arr is array (1..10) of Small;
pragma Pack (Arr);
type Arr_Ptr is access Arr;
An_Arr_Ptr : Arr_Ptr := new Arr'(10, 20, 30, 40, 50, 60, 62, 63,
-23, 42);
Trying to print one element of An_Arr_Ptr yields:
(gdb) p an_arr_ptr(3)
Cannot access memory at address 0x0
The patch adds the value_optimized_out_const function for that,
allowing us to avoid trying to fetch a value at a dummy address.
(I found this out by grepping for set_value_optimized_out and trying
to convert the uses I found to instead allocate the value with
allocate_optimized_out_value.)
Tested on x86_64 Fedora 17.
gdb/
2013-07-09 Pedro Alves <palves@redhat.com>
* ada-lang.c (coerce_unspec_val_to_type): Use
value_optimized_out_const.
* value.c (value_optimized_out_const): New function.
* value.h (value_optimized_out_const): New declaration.
If enough information is provided by the compiler, the debugger
now prints the entry value of various parameters:
(gdb) continue
Continuing.
Breakpoint 2, pck.call_me (w=w@entry=50) at [...]
20 Last_Word := W;
This patch adjusts the expected output to allow an optional
"w@entry=" in the parameter value.
gdb/testsuite/ChangeLog:
* gdb.ada/small_reg_param.exp: Accept optional entry value
for parameter "w".
Enum values rename as well. All uses updated.
* valprint.h (value_print_options): Rename member pretty to
pretty format. Rename member prettyprint_arrays to
prettyformat_arrays. Rename member prettyprint_structs to
prettyformat_structs. All uses updated.
(get_no_prettyformat_print_options): Renamed from
get_raw_print_options.
* valprint.c (get_no_prettyformat_print_options): Renamed from
get_raw_print_options. All callers updated.
(show_prettyformat_structs): Renamed from show_prettyprint_structs.
All callers updated.
(show_prettyformat_arrays): Renamed from show_prettyprint_arrays.
All callers updated.
(_initialize_valprint): Improve help text for "set print pretty" and
"set print arrays".
testsuite/
* gdb.base/default.exp: Update expected output of "show print array"
and "show print pretty".
ARM uses @ as a comment character, but % seems to be usable by all
existing ifunc enabled architectures.
gdb/testsuite/ChangeLog:
2013-07-05 Will Newton <will.newton@linaro.org>
* gdb.base/gnu-ifunc-lib.c: Use %function instead of @function
in asm syntax to allow building on ARM.
Allocate the value as optimized out from the start rather than allocating
a value with contents, and then marking it optimized out.
gdb/
2013-07-04 Pedro Alves <palves@redhat.com>
* findvar.c (value_of_register): Use allocate_optimized_out_value
if the register has been optimized out, instead of
set_value_optimized_out.
* frame-unwind.c (frame_unwind_got_optimized): Use
allocate_optimized_out_value.
Doing something else, I factored out the bits of the value_bits_valid
function that actually handle the check_validity hook, and
surprisingly found out that the result was misbehaving. Turns out
value_bits_valid has a latent bug. If the value is not lval_computed,
or doesn't have a check_validity hook, then we should assume the value
is entirely valid, not invalid. This is currently masked by the
value->optimized_out check -- I ran the testsuite with a gdb_assert(0)
inserted in place of that return being touched by the patch, and it
never triggers.
Tested on x86_64 Fedora 17.
gdb/
2013-07-04 Pedro Alves <palves@redhat.com>
* value.c (value_bits_valid): If the value is not lval_computed,
or doesn't have a check_validity hook, assume the value is entirely
valid.
gdb/ChangeLog
* stack.c (read_frame_arg): No longer fetch lazy values.
* value.c (value_optimized_out): If the value is not already
marked optimized out, and is lazy then fetch it.
(value_primitive_field): Move optimized out check to later in the
function, after we have loaded any lazy values.
(value_fetch_lazy): Use optimized out flag directly rather than
calling optimized_out method.
gdb/testsuite/ChangeLog
* gdb.dwarf2/dw2-reg-undefined.exp: New file.
* gdb.dwarf2/dw2-reg-undefined.c: Likewise.
* gdb.dwarf2/dw2-reg-undefined.S: Likewise.
* valops.c: Don't include "user-regs.h".
(value_fetch_lazy): Moved to value.c.
* value.c: Include "user-regs.h".
(value_fetch_lazy): Moved from valops.c.
2013-07-04 Yao Qi <yao@codesourcery.com>
Revert:
2013-06-27 Yao Qi <yao@codesourcery.com>
* common/create-version.sh: Update comments. Handle the case
that TARGET_ALIAS is empty.
gdb/gdbserver/
2013-07-04 Yao Qi <yao@codesourcery.com>
* Makefile.in (host_alias): Use @host_noncanonical@.
(target_alias): Use @target_noncanonical@.
* configure.ac: Use ACX_NONCANONICAL_TARGET and
ACX_NONCANONICAL_HOST.
* configure: Regenerated.
Revert:
2013-06-28 Mircea Gherzan <mircea.gherzan@intel.com>
* configure.ac (version_host, version_target): Set and AC_SUBST them.
* configure: Rebuild.
* Makefile.in (version_host, version_target): Get from configure.
(version.c): Use $(version_host) and $(version_target).
This factors --enable-libmcheck related bits from GDB's configure.ac
and makes GDBserver use them too. Specifically, the 'development'
global is moved to a separate script to it can be sourced by both GDB
and GDBserver, and the --enable-libmcheck/--disable-libmcheck bits
proper are moved to a new m4 file.
I started out by defining 'development' in the m4 file, but in the end
decided against it, as a separate script has the advantage that
changing it in release branches does not require regenerating
configure, unlike today.
I had also started out by making the new GDB_AC_LIBMCHECK itself
handle the yes/no default fallback depending on release/developement,
but since I had split out 'development' to a separate script, and, GDB
needs the python checks anyway (hence we'd need to do the python
checks in gdb's configure.ac, and pass in a 'default lmcheck yes/no'
parameter to GDB_AC_LIBMCHECK anyway), I ended up keeping
GDB_AC_LIBMCHECK isolated from the 'development' global. IOW, it's
the caller's business to handle it.
Tested on x86_64 Fedora 17. Built GDB and GDBserver with and without
--enable-libmcheck, and observed --enable-libmcheck overrides the
disablement of -lmcheck caused by python supporting threads, and that
GDBserver links with -lmcheck when expected. Also observed that
changing the 'development' global, and issuing "make" triggers a
relink, and '-lmcheck' is included or not from the link accordingly.
gdb/
2013-07-03 Pedro Alves <palves@redhat.com>
* Makefile.in (config.status): Depend on development.sh.
(aclocal_m4_deps): Add libmcheck.m4.
* acinclude.m4: Include libmcheck.m4.
* configure.ac: Source development.sh instead of setting
'development' here. --enable-libmcheck/--disable-libmcheck code
factored out to GDB_AC_LIBMCHECK. Run it.
* development.sh: New file.
* libmcheck.m4: New file.
* configure: Regenerate.
gdb/gdbserver/
2013-07-03 Pedro Alves <palves@redhat.com>
* Makefile.in (config.status): Depend on development.sh.
* acinclude.m4: Include libmcheck.m4.
* configure: Regenerate.
The documentation refers to "target nrom", but this target doesn't
appear in the tree. It was zapped here:
2002-12-16 Andrew Cagney <ac131313@redhat.com>
[...]
* remote-nrom.c, remote-os9k.c, remote-vx960.c: Delete.
This patch removes the reference from the documentation.
* gdb.texinfo (Target Commands): Don't mention "target nrom".
2013-06-25 Mircea Gherzan <mircea.gherzan@intel.com>
gdbserver/
* notif.h (notif_event): Add a dummy member to avoid compiler
errors.
Change-Id: I490dbdb70a24f52b3947371f7c0397bf7a18423c
Signed-off-by: Mircea Gherzan <mircea.gherzan@intel.com>
BFD recently got a few functions related to "dwz" files. This patch
changes gdb to use them, just to share a bit more code.
This changes dwarf2_get_dwz_file to possibly return NULL. This
simplified a bit of code elsewhere.
Built and regtested on x86-64 Fedora 18. I specifically regtested it
using my pending dwz test case.
* dwarf2read.c (dwarf2_get_dwz_file): Return NULL if
.gnu_debugaltlink not found. Use bfd_get_alt_debug_link_info.
(dwarf2_read_index, create_all_comp_units): Update.
This adds -Wold-style-definition to gdb's list of warnings. This
found a couple of spots where "()" was used where "(void)" is more
correct.
Tested by rebuilding on x86-64 Fedora 18.
* configure.ac (build_warnings): Add -Wold-style-definition.
* configure: Rebuild.
* machoread.c (_initialize_machoread): Use "(void)".
* macrocmd.c (macro_inform_no_debuginfo): Fix formatting;
use "(void)".
This adds -Wold-style-declaration to gdb's list of warnings.
It turns out that a few places use "const static" rather than
"static const". The former is deprecated according to the C standard.
Tested by rebuilding with --enable-targets=all on x86-64 Fedora 18.
* configure.ac (build_warnings): Add -Wold-style-declaration.
* configure: Rebuild.
* dsrec.c (make_srec): Use "static const", not "const static".
* h8300-tdep.c (h8300_breakpoint_from_pc): Use "static const",
not "const static".
* mi/mi-parse.c (mi_no_values, mi_simple_values, mi_all_values):
Use "static const", not "const static".
* mn10300-tdep.c (mn10300_breakpoint_from_pc): Use "static const",
not "const static".
* moxie-tdep.c (moxie_breakpoint_from_pc): Use "static const",
not "const static".
* rs6000-tdep.c (rs6000_breakpoint_from_pc): Use "static const",
not "const static".
* v850-tdep.c (v850_breakpoint_from_pc): Use "static const",
not "const static".
(v850_dbtrap_breakpoint_from_pc): Likewise.
* xstormy16-tdep.c (xstormy16_breakpoint_from_pc): Use "static const",
not "const static".
This adds -Wmissing-parameter-type to gdb's list of warnings.
This one doesn't happen to trigger for a --enable-targets=all build on
x86-64 Fedora 18.
* configure.ac (build_warnings): Add -Wmissing-parameter-type.
* configure: Rebuild.
PATH_MAX is not defined on systems which have no limit on filename
length, such as GNU/Hurd. As designed, the hostio RSP packets carry
the paths as parameters in the request/reply packets, which themselves
have an upper size limit, so lifting the filename limit completely
would require a redesign with new hostio packets. While that doesn't
happen, we can at least support filename lengths as long as the packet
buffer can fit.
gdb/gdbserver/
2013-07-01 Pedro Alves <palves@redhat.com>
* hostio.c (HOSTIO_PATH_MAX): Define.
(require_filename, handle_open, handle_unlink, handle_readlink):
Use it.
With the pathmax gnulib module in place, we can use PATH_MAX
consistently throughout, instead of the current mixbag of PATH_MAX and
MAXPATHLEN uses. It's no longer necessary to include sys/param.h
(supposedly, I can't check all ports touched here) for MAXPATHLEN.
Don't remove sys/param.h from GDB's configure.ac, as later tests in
the file use HAVE_SYS_PARAM_H checks.
Tested on x86_64 Fedora 17.
Also cross-built for --host=i686-w64-mingw32, and --host=i586-pc-msdosdjgpp.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* defs.h: Include "pathmax.h".
* utils.c: Don't include sys/param.h.
(gdb_realpath): Remove code that checks for MAXPATHLEN.
* solib-ia64-hpux.c (ia64_hpux_handle_load_event): Use PATH_MAX
instead of MAXPATHLEN.
* solib-sunos.c: Don't include sys/param.h.
* xcoffread.c: Don't include sys/param.h.
* bsd-kvm.c: Don't include sys/param.h.
* darwin-nat.c: Don't include sys/param.h.
(darwin_pid_to_exec_file): Use PATH_MAX instead of MAXPATHLEN.
* darwin-nat-info.c: Don't include sys/param.h.
* fbsd-nat.c (fbsd_pid_to_exec_file): Use PATH_MAX instead of
MAXPATHLEN.
* i386obsd-nat.c: Don't include sys/param.h.
* inf-child.c: Don't include sys/param.h.
(inf_child_fileio_readlink): Use PATH_MAX instead of MAXPATHLEN.
* linux-fork.c: Don't include sys/param.h.
(fork_save_infrun_state): Use PATH_MAX instead of MAXPATHLEN.
* linux-nat.c: Don't include sys/param.h.
(linux_child_pid_to_exec_file, linux_proc_pending_signals)
(linux_proc_pending_signals): Use PATH_MAX instead of MAXPATHLEN.
* m68klinux-nat.c: Don't include sys/param.h.
* nbsd-nat.c: Don't include sys/param.h.
(nbsd_pid_to_exec_file): Use PATH_MAX instead of MAXPATHLEN.
* ppc-linux-nat.c: Don't include sys/param.h.
* rs6000-nat.c: Don't include sys/param.h.
* spu-linux-nat.c. Don't include sys/param.h.
* windows-nat.c: Don't include sys/param.h.
* xtensa-linux-nat.c: Don't include sys/param.h.
* config/i386/nm-fbsd.h: Don't include sys/param.h.
gdb/gdbserver/
2013-07-01 Pedro Alves <palves@redhat.com>
* server.h: Include "pathmax.h".
* linux-low.c: Don't include sys/param.h.
(linux_pid_exe_is_elf_64_file): Use PATH_MAX instead of
MAXPATHLEN.
* win32-low.c: Don't include sys/param.h.
(win32_create_inferior): Use PATH_MAX instead of MAXPATHLEN.
This imports the "pathmax" gnulib module, making PATH_MAX always
available on systems that have the notion of a constant max path limit
(i.e., practically everywhere except the Hurd), along with fixing a
couple broken systems --- see pathmax.h in the patch.
This means we can normalize on PATH_MAX throughout the tree (instead
of some places using MAXPATHLEN, or defining fallback constants
ourselves).
This is the just the importing step.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* gnulib/update-gnulib.sh (IMPORTED_GNULIB_MODULES): Add pathmax.
* gnulib/Makefile.in (aclocal_m4_deps): Add import/m4/pathmax.m4.
* gnulib/aclocal.m4: Regenerate.
* gnulib/config.in: Regenerate.
* gnulib/configure: Regenerate.
* gnulib/import/pathmax.h: New file.
* gnulib/import/Makefile.am: Regenerate.
* gnulib/import/Makefile.in: Regenerate.
* gnulib/import/m4/gnulib-cache.m4: Regenerate.
* gnulib/import/m4/gnulib-comp.m4: Regenerate.
* gnulib/import/m4/pathmax.m4: New file.
This simplifies the .gdbinit filename selection logic.
We have a GDBINIT_FILENAME define that supposedly configurations would
override, but none do so. Instead, the only configuration that wants
a different file name instead of ".gdbinit", djgpp, does a strcpy over
the gdbinit global array. This means the array needs to be sized, and
the code that does that is doing the usual
'PATH_MAX/FILENAME_MAX/fallback constant/etc.' mess.
Instead of all that, it's much simpler to have configure specificy the
.gdbinit filename. As bonus, we can then make the "gdbinit" global
array const.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* configure.ac (GDBINIT): Define, depending on host.
* go32-nat.c (init_go32_ops): Don't override gdbinit here.
* top.c (PATH_MAX): Delete fallback definition.
(GDBINIT_FILENAME): Delete.
(gdbinit): Reimplement as const char array set to the GDBINIT
string constant.
* top.h (gdbinit): Make const.
A following patch will want to make the "gdbinit" global array const.
As usual, that forces in a cascading series of const additions. This
patch preemptively does those. I went all the way up to constifying
catch_command_errors, but then that would require constifying
execute_command as well (which is a much more significant effort). So
as stop point, I found the cleanest would be to add a variant of
catch_command_errors that takes const args, and use that in the few
spots that needed it due to the the get_init_files constification.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* cli/cli-cmds.c (source_script): Make 'file' parameter const.
* cli/cli-cmds.h (source_script): Likewise.
* exceptions.c (catch_command_errors_const): New function.
* exceptions.h (catch_command_errors_const): Declare.
* main.c (get_init_files): Make parameters const, and adjust.
(captured_main): Make 'system_gdbinit', 'home_gdbinit' and
'local_gdbinit' locals const. Adjust to use
catch_command_errors_const.
(print_gdb_help): Make 'system_gdbinit', 'home_gdbinit' and
'local_gdbinit' locals const.
With gnulib's unistd module, we can assume unistd.h is always present, and that
STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO are always defined.
Don't remove unistd.h from GDB's configure.ac, as later tests in the
file use HAVE_UNISTD_H checks.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* defs.h: Don't check HAVE_UNISTD_H before including <unistd.h>.
(STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO): Delete.
* tracepoint.c: Don't check HAVE_UNISTD_H before including
<unistd.h>.
gdb/gdbserver/
2013-07-01 Pedro Alves <palves@redhat.com>
* event-loop.c: Don't check HAVE_UNISTD_H before including
<unistd.h>.
* gdbreplay.c: Likewise.
* remote-utils.c: Likewise.
* server.c: Likewise.
* configure.ac: Don't check for unistd.h.
* configure: Regenerate.
I wanted to import the pathmax module, and that pulls in the unistd
module as dependency. The unistd module is actually bigger than the
pathmax module. If we're going to end up with it, might as well
import it explicitly, and make use of it throughout.
The "unistd" module makes a GNU-like <unistd.h> always available.
This means we no longer need to do:
+#ifdef HAVE_UNISTD_H
#include <unistd.h>
+#endif
and we can remove a few constants from defs.h.
This is just the importing step.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
Import the "unistd" gnulib module.
* gnulib/update-gnulib.sh (IMPORTED_GNULIB_MODULES): Add "unistd".
* gnulib/Makefile.in (aclocal_m4_deps): Add import/m4/off_t.m4,
import/m4/ssize_t.m4, import/m4/sys_types_h.m4 and
import/m4/unistd_h.m4.
* gnulib/aclocal.m4: Renenerate.
* gnulib/config.in: Renenerate.
* gnulib/configure: Renenerate.
* gnulib/import/Makefile.am: Renenerate.
* gnulib/import/Makefile.in: Renenerate.
* gnulib/import/m4/gnulib-cache.m4: Renenerate.
* gnulib/import/m4/gnulib-comp.m4: Renenerate.
* gnulib/import/m4/off_t.m4: New file.
* gnulib/import/m4/ssize_t.m4: New file.
* gnulib/import/m4/sys_types_h.m4: New file.
* gnulib/import/m4/unistd_h.m4: New file.
* gnulib/import/sys_types.in.h: New file.
* gnulib/import/unistd.c: New file.
* gnulib/import/unistd.in.h: New file.
This check in utils.c for HAVE_UNISTD_H is being used as proxy for
"HAVE_PATHCONF", as pathconf is supposed to be declared in unistd.h.
It's possible that there are systems out there that have realpath,
unistd.h and alloca, but not pathconf+_PC_PATH_MAX. I don't know of
any by heart, but if we import gnulib's unistd module (which a
following patch will do), then unistd.h ends up always available, so
the check ends up incorrect. As pathconf is being called with
_PC_PATH_MAX, check for that instead.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
* utils.c <pathconf/_PC_PATH_MAX use>: Check if _PC_PATH_MAX is
defined instead of checking HAVE_UNISTD_H.
Moving aside gnulib/import/, and re-running our
gnulib/update-gnulib.sh script, surprisingly, one gets a different
result compared to what's in the tree. This is with pristine FSF
autoconf and FSF automake, at the versions required by
update-gnulib.sh. However, if one just runs the update-gnulib.sh
scripts against the _existing_ tree, then nothing changes... I
suspect gnulib-tool's merge logic might be preserving some things by
design. This gets rid of cruft that might have accumulated over
gnulib updates. onceonly.m4 seems to fit in that category.
gdb/
2013-07-01 Pedro Alves <palves@redhat.com>
Reimport gnulib from scratch.
* gnulib/Makefile.in (aclocal_m4_deps): Remove reference to
import/m4/onceonly.m4.
* gnulib/aclocal.m4: Renegerate.
* gnulib/config.in: Renegerate.
* gnulib/configure: Renegerate.
* gnulib/import/Makefile.in: Renegerate.
* gnulib/import/extra/update-copyright: Renegerate.
* gnulib/import/m4/onceonly.m4: Delete.
Joel Brobecker reported a bug which was happening because of the
"Improved linker-debugger interface", by Gary Benson, when tested on
IA-64 with rhES5. The message is:
<http://sourceware.org/ml/gdb-patches/2013-06/msg00745.html>
The failure happened because GDB was trying to parse the probes'
operands using only the generic parser (implemented on stap-probe.c),
because the arch-specific parser was not implemented. Thus, I am
committing the following changes which solve the bug.
2013-06-29 Sergio Durigan Junior <sergiodj@redhat.com>
* ia64-linux-tdep.c: Include <ctype.h>.
(ia64_linux_stap_is_single_operand): New function.
(ia64_linux_init_abi): Initialize SystemTap related attributes.
This reverts part of the earlier version.in change. It moves
version.in back to the gdb directory. This works around the CVS bug
we've found.
gdb
* Makefile.in (version.c): Use version.in, not
common/version.in.
* common/create-version.sh: Likewise.
* common/version.in: Move...
* version.in: ...here.
gdb/doc
* Makefile.in (version.subst): Use version.in, not
common/version.in.
* gdbint.texinfo (Versions and Branches, Releasing GDB):
Likewise.
gdb/gdbserver
* Makefile.in (version.c): Use version.in, not
common/version.in.
sim/common
* Make-common.in (version.c): Use version.in, not
common/version.in.
* create-version.sh: Likewise.
sim/ppc:
* Make-common.in (version.c): Use version.in, not
common/version.in.
