Tim Wiederhake [Thu, 4 May 2017 07:05:00 +0000 (09:05 +0200)]
Python: Introduce gdb.RecordGap class
As discussed here: https://sourceware.org/ml/gdb-patches/2017-04/msg00157.html
A gap is not an instruction and it should not pretend to be one.
gdb.Record.instruction_history is now a list of gdb.RecordInstruction and
gdb.RecordGap objects. This allows the user to deal with Gaps in the record
in a more sane way.
Tim Wiederhake [Thu, 4 May 2017 07:05:00 +0000 (09:05 +0200)]
Python: Use correct ptid in btrace recording
The user would always get the instruction_history and function_call_history
objects of the current thread, not the thread for which the gdb.Record object
was created.
The attached testcase fails without this patch and passes with the patch.
Keith Seitz [Wed, 3 May 2017 19:40:01 +0000 (12:40 -0700)]
Make sure malloc is linked into gdb.cp/oranking.cc.
On some platforms, e.g., arm-eabi-none, we need to make certain that
malloc is linked into the program because the test suite uses function
calls requiring malloc:
(gdb) p foo101("abc")
evaluation of this expression requires the program to have a function "malloc".
gdb/testsuite/ChangeLog
* gdb.cp/oranking.cc (dummy): New function to grab malloc.
(main): Call it.
Keith Seitz [Thu, 27 Apr 2017 23:04:33 +0000 (16:04 -0700)]
Fix overload resolution involving rvalue references and cv qualifiers.
The following patch fixes several outstanding overload resolution problems
with rvalue references and cv qualifiers in the test suite. The tests for
these problems typically passed with one compiler version and failed with
another. This behavior occurs because of the ordering of the overloaded
functions in the debug info. So the first best match "won out" over the
a subsequent better match.
One of the bugs addressed by this patch is the failure of rank_one_type to
account for type equality of two overloads based on CV qualifiers. This was
leading directly to problems evaluating rvalue reference overload quality,
but it is also highlighted in gdb.cp/oranking.exp, where two test KFAIL as
a result of this shortcoming.
I found the overload resolution code committed with the rvalue reference
patch (f9aeb8d49) needlessly over-complicated, and I have greatly simplified
it. This fixes some KFAILing tests in gdb.exp/rvalue-ref-overload.exp.
* gdb.cp/oranking.cc (test15): New function.
(main): Call test15 and declare additional variables for testing.
* gdb.cp/oranking.exp: Remove kfail status for "p foo4(&a)" and
"p foo101('abc')" tests.
* gdb.cp/rvalue-ref-overloads.exp: Remove kfail status for
"lvalue reference overload" test.
* gdb.cp/rvalue-ref-params.exp: Remove kfail status for
"print value of f1 on Child&& in f2" test.
Simon Marchi [Thu, 27 Apr 2017 21:03:20 +0000 (17:03 -0400)]
Add missing incref when creating Inferior Python object
The test py-inferior.exp fails when using a debug build of Python 3.6. I don't
see it failing with my system's default Python, but it might be related to the
different memory allocation scheme used when doing a build with pydebug.
The issue is that we are missing a Py_INCREF in
inferior_to_inferior_object. The PyObject_New function initializes the
object with a refcount of 1. If we assume that this refcount
corresponds to the reference we are returning, then we are missing an
incref for the reference in the inferior data.
The counterpart for the incref that corresponds to the reference in the
inferior data is in py_free_inferior, in the form the gdbpy_ref instance.
Here's how I can get it to crash (with some debug output):
Current thread 0x00007f9cf1b68780 (most recent call first):
File "<string>", line 1 in <module>
[1] 408 abort (core dumped) ./gdb -nx -ex "set debug python 1"
After having created the inferiors object, their refcount is 1 (which
comes from PyObject_New), but it should be two. The gdb inferior object
has a reference and the "infs" list has a reference.
When invoking remove-inferiors, py_free_inferior gets called. It does
the decref that corresponds to the reference that the gdb inferior
object kept. At this moment, the refcount drops to 0 and the object
gets deallocated, even though the "infs" list still has a reference.
When we set "infs" to None, Python tries to decref the already zero
refcount and the assert triggers.
Jan Kratochvil [Fri, 21 Apr 2017 14:14:37 +0000 (16:14 +0200)]
release branch: Fix: --enable-werror
gdb-8.0-branch
./configure --enable-werror --enable-targets=all
aarch64-tdep.c:3045:13: error: ‘void selftests::aarch64_process_record_test()’ declared ‘static’ but never defined [-Werror=unused-function]
arm-tdep.c:9601:13: error: ‘void selftests::arm_record_test()’ declared ‘static’ but never defined [-Werror=unused-function]
gdb/ChangeLog
2017-04-21 Jan Kratochvil <jan.kratochvil@redhat.com>
* aarch64-tdep.c (selftests::aarch64_process_record_test): Make it #if
GDB_SELF_TEST.
* arm-tdep.c (selftests::arm_record_test): Likewise.
On gdb/windows-nat.c:windows_create_inferior, ALLARGS needs to be
declared independently of the host that we're building for. This
fixes a build breakage on Cygwin.
Pedro Alves [Thu, 13 Apr 2017 15:15:35 +0000 (16:15 +0100)]
Make inferior a class with cdtors, and use new/delete
struct inferior became a non-POD when enum_flags was made a non-POD,
so we should be allocating/destroying inferiors with new/delete, etc.
That's what this commit does.
Note: this commit makes all boolean fields of inferior be "bool",
except the "detaching" field. That'll require more work, so I split
it to a separate patch.
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* inferior.c (free_inferior): Convert to ...
(inferior::~inferior): ... this dtor.
(inferior::inferior): New ctor, factored out from ...
(add_inferior_silent): ... here. Allocate the inferior with a new
expression.
(delete_inferior): Call delete instead of free_inferior.
* inferior.h (gdb_environ, continuation): Forward declare.
(inferior): Now a class. Add in-class initialization to all
members. Make boolean fields bool, except 'detaching'.
(inferior::inferior): New explicit ctor.
(inferior::~inferior): New.
Pedro Alves [Thu, 13 Apr 2017 15:15:34 +0000 (16:15 +0100)]
GC inferior.c:init_inferior_list
Not used anywhere. This was actually never used. It came in because
I originally created inferior.c by copying thread.c, and doing
s/thread/inferior/g, and missed that nothing needs this. :-)
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
Pedro Alves [Thu, 13 Apr 2017 15:15:34 +0000 (16:15 +0100)]
C++fy thread_apply_all_command
This eliminates a couple cleanups.
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* thread.c: Include <algorithm>.
(thread_array_cleanup): Delete.
(scoped_inc_dec_ref): New class.
(live_threads_count): New function.
(set_thread_refcount): Delete.
(tp_array_compar_ascending): Now a bool.
(tp_array_compar): Convert to a std::sort comparison function.
(thread_apply_all_command): Use std::vector and scoped_inc_dec_ref
and live_threads_count.
Pedro Alves [Thu, 13 Apr 2017 15:15:34 +0000 (16:15 +0100)]
Fix follow-fork latent bug
A later patch in the series adds an assertion to switch_to_thread that
the resulting inferior_ptid always matches the "current_inferior()"
inferior. This exposed a latent bug in the follow-fork code, where
we're building the fork child inferior. We're switching
inferior_ptid, but not the current inferior object...
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* infrun.c (follow_fork_inferior): Also switch the current
inferior.
Pedro Alves [Thu, 13 Apr 2017 15:15:34 +0000 (16:15 +0100)]
watch_command_1: Fix dangling frame access
While working on some changes to switch_to_thread, I inadvertently
make switch_to_thread call reinit_frame_cache more frequently, even
when the thread didn't change. This exposed a latent bug in
watch_command_1, where we're referencing a frame after
creating/inserting breakpoints, which potentially calls
reinit_frame_cache if it needs to install breakpoints with a different
thread selected.
Handle this similarly to how it's already handled in other similar
cases. I.e., save any frame-related information we might need before
creating a breakpoint.
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* breakpoint.c (watch_command_1): Save watchpoint-frame info
before calling create_internal_breakpoint.
Nick Clifton [Thu, 13 Apr 2017 13:50:56 +0000 (14:50 +0100)]
Add note merging to strip and add code to merge stack size notes.
* objcopy.c: Add --no-merge-notes option to disable note merging.
Add --[no-]merge-notes option to strip, and enable it by default.
(num_bytes): New function.
(merge_gnu_build_notes): Add code to merge stack size notes.
* binutils.texi: Update strip and objcopy documentation.
* readelf.c (print_gnu_build_attribute_name): Use defined
constants for note types.
The previous change to fork-child.c converted the argv building from
an alloca-allocated array of non-owning arg pointers, to a std::vector
of owning pointers, which results in N string dups, with N being the
number of arguments in the vector, and then requires manually
releasing the pointers owned by the vector.
This patch makes the vector hold non-owning pointers, and avoids the
string dups, by doing one single string copy of the arguments upfront,
and replacing separators with NULL terminators in place, like we used
to. All the logic to do that is encapsulated in a new class.
With this, there's no need to remember to manually release the argv
elements with free_vector_argv either.
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* fork-child.c (execv_argv): New class.
(breakup_args): Refactored as ...
(execv_argv::init_for_no_shell): .. this method of execv_argv.
Copy arguments to storage and replace separators with NULL
terminators in place.
(escape_bang_in_quoted_argument): Adjust to return bool.
(execv_argv::execv_argv): New ctor.
(execv_argv::init_for_shell): New method, factored out from
fork_inferior. Don't strdup strings into the vector.
(fork_inferior): Eliminate "shell" local and use execv_argv. Use
Remove free_vector_argv call.
Alan Modra [Thu, 13 Apr 2017 01:28:40 +0000 (10:58 +0930)]
Rewrite bfd error handler
This steals _doprnt from libiberty, extended to handle %A and %B.
Which lets us do away with the current horrible %A and %B handling
that requires all %A and %B arguments to be passed first, rather than
in the natural order.
* bfd.c (PRINT_TYPE): Define.
(_doprnt): New function.
(error_handler_internal): Use _doprnt.
* coff-arm.c: Put %A and %B arguments to _bfd_error_handler
calls in their natural order, throughout file.
* coff-mcore.c: Likewise.
* coff-ppc.c: Likewise.
* coff-tic80.c: Likewise.
* cofflink.c: Likewise.
* elf-s390-common.c: Likewise.
* elf.c: Likewise.
* elf32-arm.c: Likewise.
* elf32-i386.c: Likewise.
* elf32-m32r.c: Likewise.
* elf32-msp430.c: Likewise.
* elf32-spu.c: Likewise.
* elf64-ia64-vms.c: Likewise.
* elf64-sparc.c: Likewise.
* elf64-x86-64.c: Likewise.
* elflink.c: Likewise.
* elfnn-aarch64.c: Likewise.
* elfnn-ia64.c: Likewise.
* elfxx-mips.c: Likewise.
Alan Modra [Thu, 13 Apr 2017 01:14:20 +0000 (10:44 +0930)]
Missing _bfd_error_handler args
* elf32-arm.c (arm_type_of_stub): Supply missing args to "long
branch veneers" error. Fix double space and format message.
* elf32-avr.c (avr_add_stub): Do not pass NULL as %B arg.
* elf64-ppc.c (tocsave_find): Supply missing %B arg.
Pedro Alves [Thu, 13 Apr 2017 01:12:53 +0000 (02:12 +0100)]
breakpoint.c: bp_location (the array) shadows bp_location (the type)
The bp_location array has the same name as the "struct bp_location",
type preventing refering to the structure without the "struct" inside
breakpoint.c. I.e., we must write:
"new struct bp_location;"
instead of:
"new bp_location"
Rename the array and the associated variables/functions to avoid the
shadowing.
gdb/ChangeLog:
2017-04-13 Pedro Alves <palves@redhat.com>
* breakpoint.c (bp_location): Rename to ...
(bp_locations): ... this. All references updated.
(bp_location_count): Rename to ...
(bp_locations_count): ... this. All references updated.
(bp_location_placed_address_before_address_max): Rename to ...
(bp_locations_placed_address_before_address_max): ... this. All
references updated.
(bp_location_shadow_len_after_address_max): Rename to ...
(bp_locations_shadow_len_after_address_max): ... this. All
references updated.
(bp_location_compare_addrs): Rename to ...
(bp_locations_compare_addrs): ... this. All references updated.
(bp_location_compare):Rename to ...
(bp_locations_compare): ... this. All references updated.
(bp_location_target_extensions_update): Rename to ...
(bp_locations_target_extensions_update): ... this. All references
updated.
As requested, I'm sending this as a separate patch because it is ready
to be included as-is.
The idea here is that both gdb/terminal.h and gdb/gdbserver/terminal.h
share the same code, which is responsible for setting a bunch of
defines on based on the presence of termios.h and a few other headers.
This simple patch just moves this common code to common/gdb_termios.h
and makes the necessary adjustments on both GDB and gdbserver so that
they can use this new header. It also implements the some header
checks on common/common.m4.
As a bonus, gdb/gdbserver/terminal.h can be removed because it's now
empty.
* Makefile.in (HFILES_NO_SRCDIR): Add "common/gdb_termios.h".
* common/common.m4: Check headers 'termios.h', 'termio.h' and
'sgtty.h'.
* common/gdb_termios.h: New file, with parts of "terminal.h".
* inflow.c: Include "gdb_termios.h".
* ser-unix.c: Include "gdb_termios.h".
* terminal.h: Move terminal-related defines to
"common/gdb_termios.h".
Tom Tromey [Mon, 10 Apr 2017 21:47:21 +0000 (15:47 -0600)]
Add a constructor and destructor to linespec_result
linespec_result is only ever allocated on the stack, so it's
relatively easy to convert to having a constructor and a destructor.
This patch makes this change. This removes some cleanups.
gdb/ChangeLog
2017-04-12 Tom Tromey <tom@tromey.com>
Tom Tromey [Mon, 10 Apr 2017 21:41:25 +0000 (15:41 -0600)]
Change breakpoint event locations to event_location_up
This is a follow-up to an earlier patch. It changes breakpoint's
location and location_range_end members to be of type
event_location_up, then fixes up the users.
gdb/ChangeLog
2017-04-12 Tom Tromey <tom@tromey.com>
Tom Tromey [Fri, 7 Apr 2017 02:09:42 +0000 (20:09 -0600)]
Introduce gdb_dlhandle_up
This introduces gdb_dlhandle_up, a unique_ptr that can close a
dlopen'd library. All the functions working with dlopen handles are
updated to use this new type.
I did not try to build this on Windows.
gdb/ChangeLog
2017-04-12 Tom Tromey <tom@tromey.com>
* jit.c (struct jit_reader): Declare separately. Add constructor
and destructor. Change type of "handle".
(loaded_jit_reader): Define separately.
(jit_reader_load): Update. New "new".
(jit_reader_unload_command): Use "delete".
* gdb-dlfcn.h (struct dlclose_deleter): New.
(gdb_dlhandle_up): New typedef.
(gdb_dlopen, gdb_dlsym): Update types.
(gdb_dlclose): Remove.
* gdb-dlfcn.c (gdb_dlopen): Return a gdb_dlhandle_up.
(gdb_dlsym): Change type of "handle".
(make_cleanup_dlclose): Remove.
(dlclose_deleter::operator()): Rename from gdb_dlclose.
* compile/compile-c-support.c (load_libcc): Update.
Tom Tromey [Thu, 6 Apr 2017 03:14:09 +0000 (21:14 -0600)]
Introduce command_line_up
This introduces command_line_up, a unique_ptr for command_line
objects, and changes many places to use it. This removes a number of
cleanups.
Command lines are funny in that sometimes they are reference counted.
Once there is more C++-ification of some of the users, perhaps all of
these can be changed to use shared_ptr instead.
gdb/ChangeLog
2017-04-12 Tom Tromey <tom@tromey.com>
Tom Tromey [Thu, 6 Apr 2017 02:44:01 +0000 (20:44 -0600)]
Introduce event_location_up
This removes make_cleanup_delete_event_location and instead changes
the various location functions to return an event_location_up, a new
unique_ptr typedef.
This is largely straightforward, but be sure to examine the
init_breakpoint_sal change. I believe the code I deleted there is
dead, because "location != NULL" can never be true in that branch; but
you should double-check.
gdb/ChangeLog
2017-04-12 Tom Tromey <tom@tromey.com>
Pedro Alves [Wed, 12 Apr 2017 15:00:04 +0000 (16:00 +0100)]
gdb: Move DJGPP/go32 bits to their own tdep file
I posit that this makes them easier to find.
The other day while working on the wchar_t patch, I had a bit of
trouble finding the DJGPP/go32 tdep bits. My initial reaction was
looking for a go32-specific tdep file, but there's none.
Confirmed that a --host=i586-pc-msdosdjgpp GDB still builds
successfully and includes the i386-go32-tdep.o object.
Confirmed that an --enable-targets=all build of GDB on x86-64
GNU/Linux includes the DJGPP/go32 bits too.
gdb/ChangeLog:
2017-04-12 Pedro Alves <palves@redhat.com>
* Makefile.in (ALL_TARGET_OBS): Add i386-go32-tdep.o.
* configure.tgt: Handle i[34567]86-*-go32* and
i[34567]86-*-msdosdjgpp*.
* i386-tdep.c (i386_svr4_reg_to_regnum):
Make extern.
(i386_go32_init_abi, i386_coff_osabi_sniffer): Moved to
i386-go32-tdep.c.
(_initialize_i386_tdep): DJGPP bits moved to i386-go32-tdep.c.
* i386-go32-tdep.c: New file.
* i386-tdep.h (tdesc_i386_mmx, i386_svr4_reg_to_regnum): New
declarations.
Simon Marchi [Wed, 12 Apr 2017 14:46:13 +0000 (10:46 -0400)]
Fix build error in aix-thread.c
Obvious fix for:
aix-thread.c: In function 'char* pd_status2str(int)':
aix-thread.c:163:33: error: deprecated conversion from string constant to 'char*' [-Werror=write-strings]
case PTHDB_SUCCESS: return "SUCCESS";
^
gdb/ChangeLog:
* aix-thread.c (pd_status2str): Change return type to const char *.
Pedro Alves [Wed, 12 Apr 2017 13:00:49 +0000 (14:00 +0100)]
Teach GDB that wchar_t is a built-in type in C++ mode
GDB is currently not aware that wchar_t is a built-in type in C++
mode. This is usually not a problem because the debug info describes
the type, so when you have a program loaded, you don't notice this.
However, if you try expressions involving wchar_t before a program is
loaded, gdb errors out:
(gdb) p (wchar_t)-1
No symbol table is loaded. Use the "file" command.
(gdb) p L"hello"
No type named wchar_t.
(gdb) ptype L"hello"
No type named wchar_t.
This commit teaches gdb about the type. After:
(gdb) p (wchar_t)-1
$1 = -1 L'\xffffffff'
(gdb) p L"hello"
$2 = L"hello"
(gdb) ptype L"hello"
type = wchar_t [6]
Unlike char16_t/char32_t, unfortunately, the underlying type of
wchar_t is implementation dependent, both size and signness. So this
requires adding a couple new gdbarch hooks.
I grepped the GCC code base for WCHAR_TYPE and WCHAR_TYPE_SIZE, and it
seems to me that the majority of the ABIs have a 4-byte signed
wchar_t, so that's what I made the default for GDB too. And then I
looked for which ports have a 16-bit and/or unsigned wchar_t, and made
GDB follow suit.
gdb/ChangeLog:
2017-04-12 Pedro Alves <palves@redhat.com>
Pedro Alves [Wed, 12 Apr 2017 13:00:49 +0000 (14:00 +0100)]
Fix PR c++/21323: GDB thinks char16_t and char32_t are signed in C++
While the C++ standard says that char16_t and char32_t are unsigned types:
Types char16_t and char32_t denote distinct types with the same size,
signedness, and alignment as uint_least16_t and uint_least32_t,
respectively, in <cstdint>, called the underlying types.
... gdb treats them as signed currently:
(gdb) p (char16_t)-1
$1 = -1 u'\xffff'
There are actually two places in gdb that hardcode these types:
- gdbtypes.c:gdbtypes_post_init, when creating the built-in types,
seemingly used by the "x /s" command (judging from commit 9a22f0d0).
- dwarf2read.c, when reading base types with DW_ATE_UTF encoding
(which is what is used for these types, when compiling for C++11 and
up). Despite the comment, the type created does end up used.
Both places need fixing. But since I couldn't tell why dwarf2read.c
needs to create a new type, I've made it use the per-arch built-in
types instead, so that the types are only created once per arch
instead of once per objfile. That seems to work fine.
While writting the test, I noticed that the C++ language parser isn't
actually aware of these built-in types, so if you try to use them
without a program that uses them, you get:
(gdb) set language c++
(gdb) ptype char16_t
No symbol table is loaded. Use the "file" command.
(gdb) ptype u"hello"
No type named char16_t.
(gdb) p u"hello"
No type named char16_t.
That's fixed by simply adding a couple entries to C++'s built-in types
array in c-lang.c. With that, we get the expected:
(gdb) ptype char16_t
type = char16_t
(gdb) ptype u"hello"
type = char16_t [6]
(gdb) p u"hello"
$1 = u"hello"
gdb/ChangeLog:
2017-04-12 Pedro Alves <palves@redhat.com>
PR c++/21323
* c-lang.c (cplus_primitive_types) <cplus_primitive_type_char16_t,
cplus_primitive_type_char32_t>: New enum values.
(cplus_language_arch_info): Register cplus_primitive_type_char16_t
and cplus_primitive_type_char32_t.
* dwarf2read.c (read_base_type) <DW_ATE_UTF>: If bit size is 16 or
32, use the archtecture's built-in type for char16_t and char32_t,
respectively. Otherwise, fallback to init_integer_type as before,
but make the type unsigned, and issue a complaint.
* gdbtypes.c (gdbtypes_post_init): Make char16_t and char32_t unsigned.
gdb/testsuite/ChangeLog:
2017-04-12 Pedro Alves <palves@redhat.com>
PR c++/21323
* gdb.cp/wide_char_types.c: New file.
* gdb.cp/wide_char_types.exp: New file.