The original problem was fixed (see related PR 22242). But using a typedef
as the declared type for a static member variable, as commented in this PR,
is still causing gdb to get into infinite loop when printing the static
member's value. This problem can be reproduced as follows:
% cat t.cc
class A {
typedef A type;
public:
bool operator==(const type& other) { return true; }
static const type INSTANCE;
};
const A A::INSTANCE;
int main() {
A a;
if (a == A::INSTANCE) {
return -1;
}
return 0;
}
% g++ -g t.cc
% gdb -ex "start" -ex "p a" a.out
The fix is rather trivial - in cp_print_static_field(), should call
check_typedef() to get the static member's real type and use it to
check whether it's a struct or an array.
As Simon suggested, I've added a new test case to the testsuite
and am passing the original type, not the real type, as argument
to both cp_print_value_fields() and val_print().
Re-tested on both aarch64-linux-gnu and amd64-linux-gnu. No regressions.
This is a command we somehow forgot to contribute at the time the Ada
language was first contributed to the FSF. This command allows
the user to change the maximum size we allow when reading memory
from dynamic objects (the default is 65536 bytes).
At the moment, this limit is only used by Ada, and so the implementation
is kept inside ada-lang.c. However, it is conceivable that other language
might want to use it also to handle the same kind of issues; for instance,
this might be useful when handling dynamic types in C. So the name
of the setting was made language-neutral, to allow for this.
Note that an alias for "set var" needs to be introduced as well.
We are not adding a test for that, since this is a feature that is
already exercized by numerous existing tests.
gdb/ChangeLog
* NEWS: Add entry describing new "set|show varsize-limit" command.
* ada-lang.c (_initialize_ada_language): Add "set/show varsize-limit"
command.
* printcmd.c (_initialize_printcmd): Add "set var" alias of
"set variable".
gdb/doc/ChangeLog:
* gdb.texinfo (Ada Settings): New subsubsection.
gdb/testsuite/ChangeLog:
* gdb.ada/varsize_limit: New testcase.
Tested on x86_64-linux.
This patch fixes a known failure in gdb.ada/maint_with_ada.exp
(maintenance check-psymtabs). Another way to witness the same
issue is by considering the following Ada declarations...
type Wrapper is record
A : Integer;
end record;
u00045 : constant Wrapper := (A => 16#060287af#);
pragma Export (C, u00045, "symada__cS");
... which declares a variable name "u00045" but with a linkage
name which is "symada__cS". This variable is a record with one
component, the Ada equivalent of a struct with one field in C.
Trying to print that variable's value currently yields:
(gdb) p /x <symada__cS>
'symada(char, signed)' has unknown type; cast it to its declared type
This indicates that GDB was only able to find the minimal symbol,
but not the full symbol. The expected output is:
(gdb) print /x <symada__cS>
$1 = (a => 0x60287af)
The error message gives a hint about what's happening: We processed
the symbol through gdb_demangle, which in the case of this particular
symbol name, ends up matching the C++ naming scheme. As a result,
the demangler transforms our symbol name into 'symada(char, signed)',
thus breaking Ada lookups.
This patch fixes the issue by first introducing a new language_defn
attribute called la_store_sym_names_in_linkage_form_p, which is a boolean
to be set to true for the few languages that do not want their symbols
to have their names stored in demangled form, and false otherwise.
We then use this language attribute to skip the call to gdb_demangle
for all languages whose la_store_sym_names_in_linkage_form_p is true.
In terms of the selection of languages for which the new attribute
is set to true, the selection errs on the side of preserving the
existing behavior, and only changes the behavior for the languages
where we are certain storing symbol names in demangling form is not
needed. It is conceivable that other languages might be in the same
situation, but I not knowing in detail the symbol name enconding
strategy, I decided to play it safe and let other language maintainers
potentially adjust their language if it makes sense to do so.
gdb/ChangeLog:
PR gdb/22670
* dwarf2read.c (dwarf2_physname): Do not return the demangled
symbol name if the CU's language stores symbol names in linkage
format.
* language.h (struct language_defn)
<la_store_sym_names_in_linkage_form_p>: New field. Adjust
all instances of this struct.
gdb/testsuite/ChangeLog:
* gdb.ada/maint_with_ada.exp: Remove PR gdb/22670 setup_kfail.
* gdb.ada/notcplusplus: New testcase.
* gdb.base/c-linkage-name.c: New file.
* gdb.base/c-linkage-name.exp: New testcase.
Tested on x86_64-linux.
This also passes AdaCore's internal GDB testsuite.
If a C-c comes while the Python code for a frame filter is running, it
will be turned into a Python KeyboardException. It seems good for
this to be treated like a GDB quit, so this patch changes
py-framefilter.c to notice this situation and call throw_quit in this
case.
gdb/ChangeLog
2018-03-26 Tom Tromey <tom@tromey.com>
* python/py-framefilter.c (throw_quit_or_print_exception): New
function.
(gdbpy_apply_frame_filter): Use it.
gdb/testsuite/ChangeLog
2018-03-26 Tom Tromey <tom@tromey.com>
* gdb.python/py-framefilter.exp: Add test for KeyboardInterrupt.
* gdb.python/py-framefilter.py (name_error): New global.
(ErrorInName.function): Use name_error.
When a frame filter elides some frames, they are still printed by
"bt", indented a few spaces. PR backtrace/15582 notes that it would
be nice for users if elided frames could simply be dropped. This
patch adds this capability.
gdb/ChangeLog
2018-03-26 Tom Tromey <tom@tromey.com>
PR backtrace/15582:
* stack.c (backtrace_command): Parse "hide" argument.
* python/py-framefilter.c (py_print_frame): Handle PRINT_HIDE.
* extension.h (enum frame_filter_flags) <PRINT_HIDE>: New
constant.
gdb/doc/ChangeLog
2018-03-26 Tom Tromey <tom@tromey.com>
PR backtrace/15582:
* gdb.texinfo (Backtrace): Mention "hide" argument.
gdb/testsuite/ChangeLog
2018-03-26 Tom Tromey <tom@tromey.com>
PR backtrace/15582:
* gdb.python/py-framefilter.exp: Add "bt hide" test.
c++/22968 involves the inability of ptype to find a type definition for
a type defined inside another type. I recently added some additional
support for nested type definitions, but I apparently overlooked psymbols.
The user reports that using -readnow fixes the problem:
$ gdb 22968 -ex "ptype Outer::Inner"
There is no field named Inner
$ gdb -readnow 22968 -ex "ptype Outer::Inner"
type = struct Outer::Inner {
<no data field>
}
We clearly did not find a psymbol for Outer::Inner because it was located
in another CU. This patch addresses this problem by scanning structs
for additional psymbols. Rust is already doing this.
With this patch, the identical result to "-readnow" is given (without
using `-readnow', of course).
gdb/ChangeLog:
PR c++/22968
* dwarf2read.c (scan_partial_symbols): Scan structs/classes for
nested type definitions for C++, too.
gdb/testsuite/ChangeLog:
PR c++/22968
* gdb.cp/subtypes.exp: New file.
* gdb.cp/subtypes.h: New file.
* gdb.cp/subtypes.cc: New file.
* gdb.cp/subtypes-2.cc: New file.
This test starts up and confirms that $xmm0 has the value 0, it then
modifies $xmm0 (in the inferior) and confirms that the new value can
be read (in GDB).
On some machines I was noticing that this test would occasionally
fail, and on investigation I believe that the reason for this is that
the test is linked as a dynamically linked executable and makes use of
the system libraries during startup. The reason that this causes
problems is that both the runtime linker and the startup code run
before main can, and do (on at least some platforms) make use of the
XMM registers.
In this commit I modify the test program slightly to allow it to be
linked statically, without using the startup libraries. Now by the
time GDB reaches the symbol main we have only executed one 'nop'
instruction, and the XMM registers should all have the value 0. I've
extended the test script to confirm that $xmm0 to $xmm15 are all
initially 0, and I also check that at the point after $xmm0 has been
modified, all the other XMM registers ($xmm1 to $xmm15) are still 0.
The test program is still linked against libc in order that we can
call the exit function, however, we now call _exit rather than exit in
order to avoid all of the usual cleanup that exit does. This clean up
tries to tear down things that are usually setup during the startup
code, but now this isn't called calling exit will just result in a
crash.
gdb/testsuite/ChangeLog:
* gdb.arch/amd64-disp-step-avx.S: Add '_start' label.
(done): Call '_exit' not 'exit' to avoid atexit handlers.
* gdb.arch/amd64-disp-step-avx.exp: Pass -static, and
-nostartfiles when compiling the test. Confirm that all registers
xmm0 to xmm15 are initially 0, and that xmm1 to xmm15 are 0 after.
A small number of tests incorrectly tried to pass -Wa,-g through to
GCC as an extra compile time flag, either to gdb_compile or
prepare_for_testing.
The problem is that the syntax used for passing the flags was
incorrect, and as a result these extra flags were being ignored.
Luckily, the 'debug' flag was being passed in each case anyway, which
means that the '-g' flag would already be added.
Given that all these tests pass 'debug', and the invalid flag has been
ignored for some time, I'm just removing the flags in this commit.
I've also changed the tests from using gdb_compile to
prepare_for_testing, which allows some extra code to be removed from a
couple of tests scripts.
There should be no change in the test results after this commit.
gdb/testsuite/ChangeLog:
* gdb.arch/amd64-disp-step-avx.exp: Remove unneeded assembler flag
option, syntax was wrong anyway.
* gdb.arch/arm-disp-step.exp: Likewise.
* gdb.arch/sparc64-regs.exp: Likewise.
* gdb.arch/amd64-disp-step.exp: Remove unneeded assembler flag
option, syntax was wrong anyway, switch to use
prepare_for_testing.
* gdb.arch/i386-disp-step.exp: Likewise.
Currently "info proc cmdline" on GNU/Linux does not show the full command
line, but only argument 0. And even a warning is shown if there are more.
This was discussed in 2014 already:
https://sourceware.org/ml/gdb-patches/2014-04/msg00212.html
Follow the advice there and avoid target_fileio_read_stralloc. Instead,
use target_fileio_read_alloc to read the whole command line and then
replace NUL characters by spaces. Also add an appropriate test case.
Note that gdbserver already handles this correctly.
gdb/ChangeLog:
* linux-tdep.c (linux_info_proc): For "info proc cmdline", print
command line args instead of emitting a warning.
gdb/testsuite/ChangeLog:
* gdb.base/info-proc.exp: Add test for "info proc cmdline".
This patch addresses slowness when setting breakpoints, especially in
heavily templatized code. Profiling showed that find_pc_sect_line in
symtab.c was the performance bottleneck. The original logic performed a
linear search over ordered data. This patch uses a binary search, as
suggested by comments around the function. There are no behavioural
changes, but gdb is now faster at setting breakpoints in template code.
Tested using on make check on an x86 target. The optimisation speeds up
the included template-breakpoints.py performance test by a factor of 7
on my machine.
ChangeLog:
2018-03-20 Stephen Roberts <stephen.roberts@arm.com>
* gdb/symtab.c (find_pc_sect_line): now uses binary search.
gdb/testsuite/
* gdb.perf/template-breakpoints.cc: New file.
* gdb.perf/template-breakpoints.exp: New file.
* gdb.perf/template-breakpoints.py: New file.
In Rust one can initialize a struct member from an identically-named
local variable by simply mentioning the member name in the
initializer, like:
let x = 0;
let y = Struct { x };
This initializes "Struct::x" from "x".
This patch adds this form of initializer to the Rust expression parser
and adds a test.
Tested on x86-64 Fedora 26 using rustc 1.23.
2018-03-19 Tom Tromey <tom@tromey.com>
* rust-exp.y (struct_expr_tail, struct_expr_list): Add plain
"IDENT" production.
2018-03-19 Tom Tromey <tom@tromey.com>
* gdb.rust/simple.rs (main): Add local variables field1, field2,
y0.
* gdb.rust/simple.exp: Test bare identifier form of struct
initializer.
This converts observers from using a special source-generating script
to be plain C++. This version of the patch takes advantage of C++11
by using std::function and variadic templates; incorporates Pedro's
patches; and renames the header file to "observable.h" (this change
eliminates the need for a clean rebuild).
Note that Pedro's patches used a template lambda in tui-hooks.c, but
this failed to compile on some buildbot instances (presumably due to
differing C++ versions); I replaced this with an ordinary template
function.
Regression tested on the buildbot.
gdb/ChangeLog
2018-03-19 Pedro Alves <palves@redhat.com>
Tom Tromey <tom@tromey.com>
* unittests/observable-selftests.c: New file.
* common/observable.h: New file.
* observable.h: New file.
* ada-lang.c, ada-tasks.c, agent.c, aix-thread.c, annotate.c,
arm-tdep.c, auto-load.c, auxv.c, break-catch-syscall.c,
breakpoint.c, bsd-uthread.c, cli/cli-interp.c, cli/cli-setshow.c,
corefile.c, dummy-frame.c, event-loop.c, event-top.c, exec.c,
extension.c, frame.c, gdbarch.c, guile/scm-breakpoint.c,
infcall.c, infcmd.c, inferior.c, inflow.c, infrun.c, jit.c,
linux-tdep.c, linux-thread-db.c, m68klinux-tdep.c,
mi/mi-cmd-break.c, mi/mi-interp.c, mi/mi-main.c, objfiles.c,
ppc-linux-nat.c, ppc-linux-tdep.c, printcmd.c, procfs.c,
python/py-breakpoint.c, python/py-finishbreakpoint.c,
python/py-inferior.c, python/py-unwind.c, ravenscar-thread.c,
record-btrace.c, record-full.c, record.c, regcache.c, remote.c,
riscv-tdep.c, sol-thread.c, solib-aix.c, solib-spu.c, solib.c,
spu-multiarch.c, spu-tdep.c, stack.c, symfile-mem.c, symfile.c,
symtab.c, thread.c, top.c, tracepoint.c, tui/tui-hooks.c,
tui/tui-interp.c, valops.c: Update all users.
* tui/tui-hooks.c (tui_bp_created_observer)
(tui_bp_deleted_observer, tui_bp_modified_observer)
(tui_inferior_exit_observer, tui_before_prompt_observer)
(tui_normal_stop_observer, tui_register_changed_observer):
Remove.
(tui_observers_token): New global.
(attach_or_detach, tui_attach_detach_observers): New functions.
(tui_install_hooks, tui_remove_hooks): Use
tui_attach_detach_observers.
* record-btrace.c (record_btrace_thread_observer): Remove.
(record_btrace_thread_observer_token): New global.
* observer.sh: Remove.
* observer.c: Rename to observable.c.
* observable.c (namespace gdb_observers): Define new objects.
(observer_debug): Move into gdb_observers namespace.
(struct observer, struct observer_list, xalloc_observer_list_node)
(xfree_observer_list_node, generic_observer_attach)
(generic_observer_detach, generic_observer_notify): Remove.
(_initialize_observer): Update.
Don't include observer.inc.
* Makefile.in (generated_files): Remove observer.h, observer.inc.
(clean mostlyclean): Likewise.
(observer.h, observer.inc): Remove targets.
(SUBDIR_UNITTESTS_SRCS): Add observable-selftests.c.
(COMMON_SFILES): Use observable.c, not observer.c.
* .gitignore: Remove observer.h.
gdb/doc/ChangeLog
2018-03-19 Tom Tromey <tom@tromey.com>
* observer.texi: Remove.
gdb/testsuite/ChangeLog
2018-03-19 Tom Tromey <tom@tromey.com>
* gdb.gdb/observer.exp: Remove.
Some of GDB's trace test cases define a function end() and place a
breakpoint there with "break end". However, when libinproctrace is linked
to the binary, there are multiple methods named "end", such as
std::string::end() from the C++ library or format_pieces::end() from
common/format.h. GDB then creates multiple breakpoints instead of just a
single one, and some FAILs result, such as these:
FAIL: gdb.trace/trace-mt.exp: ftrace on: break end
FAIL: gdb.trace/trace-mt.exp: ftrace off: break end
Fix this by adding the "-qualified" option to the break commands. For
consistency, change all occurrences of "break end" (and similar) in all
trace test cases, even if the current behavior does not cause problems.
Also, consequently use the gdb_breakpoint convenience proc.
gdb/testsuite/ChangeLog:
* gdb.trace/actions-changed.exp: Call gdb_breakpoint with the
"qualified" option when setting breakpoints.
* gdb.trace/backtrace.exp: Likewise.
* gdb.trace/circ.exp: Likewise.
* gdb.trace/collection.exp: Likewise.
* gdb.trace/disconnected-tracing.exp: Likewise.
* gdb.trace/ftrace-lock.exp: Likewise.
* gdb.trace/ftrace.exp: Likewise.
* gdb.trace/infotrace.exp: Likewise.
* gdb.trace/packetlen.exp: Likewise.
* gdb.trace/passc-dyn.exp: Likewise.
* gdb.trace/qtro.exp: Likewise.
* gdb.trace/read-memory.exp: Likewise.
* gdb.trace/report.exp: Likewise.
* gdb.trace/signal.exp: Likewise.
* gdb.trace/status-stop.exp: Likewise.
* gdb.trace/strace.exp: Likewise.
* gdb.trace/tfind.exp: Likewise.
* gdb.trace/trace-break.exp: Likewise.
* gdb.trace/trace-condition.exp: Likewise.
* gdb.trace/trace-mt.exp: Likewise.
* gdb.trace/tstatus.exp: Likewise.
* gdb.trace/tsv.exp: Likewise.
* gdb.trace/unavailable-dwarf-piece.exp: Likewise.
* gdb.trace/unavailable.exp: Likewise.
* gdb.trace/while-dyn.exp: Likewise.
The tspeed test case does not execute correctly because libinproctrace.so
is not copied to the target. This is fixed.
gdb/testsuite/ChangeLog:
* gdb.trace/tspeed.exp: Add invocation of gdb_load_shlib to ensure
that libinproctrace is copied to the target.
This changes the printf command's %s and %ls formats to special-case
NULL, and print "(null)" for these. This is PR cli/14977. This
behavior seems a bit friendlier; I was undecided on whether other
invalid pointers should be handled specially somehow, so for the time
being I've left those out.
gdb/ChangeLog
2018-03-14 Tom Tromey <tom@tromey.com>
PR cli/14977:
* printcmd.c (printf_c_string, printf_wide_c_string): Special case
for NULL.
gdb/gdbserver/ChangeLog
2018-03-14 Tom Tromey <tom@tromey.com>
PR cli/14977:
* ax.c (ax_printf): Special case for NULL.
gdb/testsuite/ChangeLog
2018-03-14 Tom Tromey <tom@tromey.com>
PR cli/14977:
* gdb.base/printcmds.exp (test_printf): Add printf test of %s with
a null pointer.
* gdb.base/wchar.exp: Likewise.
PR cli/19918 points out that a printf format like "%-5p" will cause a
gdb crash. The bug is problem is that printf_pointer doesn't take the
"-" flag into account.
gdb/ChangeLog
2018-03-14 Tom Tromey <tom@tromey.com>
PR cli/19918:
* printcmd.c (printf_pointer): Allow "-" in format.
gdb/testsuite/ChangeLog
2018-03-14 Tom Tromey <tom@tromey.com>
PR cli/19918:
* gdb.base/printcmds.exp (test_printf): Add printf test using '-'
flag.
As described here
https://sourceware.org/bugzilla/show_bug.cgi?id=22841
there seems to be situations where the remote-stdio-gdbserver board
fails to delete the uploaded binary file. Passing "target" fixes the
issue for Christian who reported the bug.
I did not experience this problem, but passing "target" to remote_exec
still works for me, so I'm fine with changing it.
Any objection?
gdb/testsuite/ChangeLog:
PR gdb/22841
* boards/remote-stdio-gdbserver.exp (${board}_file): Pass
"target" to remote_exec.
Before patch
Make native gdbserver boards no longer be "remote" (in DejaGnu terms)
739b3f1d8f
the local gdbserver boards (except native-extended-gdbserver...) were
considered as remote by DejaGNU. To avoid DejaGNU trying to use ssh/scp
to download the files to the target (which is actually local), the
gdbserver-base.exp file defined some _download, _upload and _file board
operations to override the default behavior, and instead just use local
operations.
The same patch also changed remote-stdio-gdbserver.exp to make it
inherit from gdbserver-base.exp. Since then, this board (which is
actually remote) uses the overrides with local file operations. As a
result, files are never actually copied to the target.
I think we can simply remove the overrides from gdbserver-base.exp.
Because all boards should be properly considered local or remote by
DejaGNU, it should by default use the right method for transferring
files.
gdb/testsuite/ChangeLog:
PR gdb/22841
* boards/gdbserver-base.exp (${board}_file, ${board}_download,
${board}_upload): Remove.
Some of the watchpoint logic depends on the fact that the head of the
value chain represents the user-specified value to watch. Thus no
additional values should be added to the value chain after that. However,
if a watchpoint is defined for a C++ structure/class object, then run-time
type information (RTTI) may be present. Thus, while constructing the
value chain for the watchpoint, the dynamic type is fetched by
gnuv3_rrti_type, which invokes value_addr, which then adds a new value to
the head of the value chain. This new value represents the pointer to the
structure instead of the structure itself.
With such a "polluted" value chain the watchpoint logic does not recognize
when the user intended to watch a struct, and can_use_hardware_watchpoint
returns zero. Instead of a hardware watchpoint, a software watchpoint
will then be set for no apparent reason.
This is fixed by adding an early exit to gnuv3_rtti_type when the input
value is not a dynamic class object.
gdb/testsuite/ChangeLog:
* gdb.cp/watch-cp.cc: New test.
* gdb.cp/watch-cp.exp: New file.
gdb/ChangeLog:
* gnu-v3-abi.c (gnuv3_rtti_type): Add early exit if the given
value is not a dynamic class object.
This commit introduces basic support for baremetal RiscV as a GDB
target. This target is currently only tested against the RiscV software
simulator, which is not included as part of this commit. The target has
been tested against the following RiscV variants: rv32im, rv32imc,
rv32imf, rv32imfc, rv64im, rv64imc, rv64imfd, rv64imfdc.
Across these variants we pass on average 34858 tests, and fail 272
tests, which is ~0.8%.
The RiscV has a feature of its ABI where structures with a single
floating point field, a single complex float field, or one float and
one integer field are treated differently for argument passing. The
new test gdb.base/infcall-nested-structs.exp is added to cover this
feature. As passing these structures should work on all targets then
I've made the test as a generic one, even though, for most targets,
there's probably nothing special about any of these cases.
gdb/ChangeLog:
* Makefile.in (ALL_TARGET_OBS): Add riscv-tdep.o
(HFILES_NO_SRCDIR): Add riscv-tdep.h.
(ALLDEPFILES): Add riscv-tdep.c
* configure.tgt: Add riscv support.
* riscv-tdep.c: New file.
* riscv-tdep.h: New file.
* NEWS: Mention new target.
* MAINTAINERS: Add entry for riscv.
gdb/testsuite/ChangeLog:
* gdb.base/infcall-nested-structs.exp: New file.
* gdb.base/infcall-nested-structs.c: New file.
* gdb.base/float.exp: Add riscv support.
While using @progbits in .pushsection work on some targets, it does not
work on arm target where this introduces a comment. This patch replaces
its use in gdb.dlang/watch-loc.c and gdb.mi/dw2-ref-missing-frame-func.c
by %progbits which should work on all targets since it is used in
target-independent elf/section7.s GAS test.
2018-03-02 Thomas Preud'homme <thomas.preudhomme@arm.com>
gdb/testsuite/
* gdb.dlang/watch-loc.c: Use %progbits instead of @progbits.
* gdb.mi/dw2-ref-missing-frame-func.c: Likewise.
Simon mentioned on IRC that, after the startup-with-shell feature has
been implemented on gdbserver, it is not possible to specify a
filename-only binary, like:
$ gdbserver :1234 a.out
/bin/bash: line 0: exec: a.out: not found
During startup program exited with code 127.
Exiting
This happens on systems where the current directory "." is not listed
in the PATH environment variable. Although including "." in the PATH
variable is a possible workaround, this can be considered a regression
because before startup-with-shell it was possible to use only the
filename (due to reason that gdbserver used "exec*" directly).
The idea of the patch is to verify if the program path provided by the
user (or by the remote protocol) contains a directory separator
character. If it doesn't, it means we're dealing with a filename-only
binary, so we call "gdb_abspath" to properly expand it and transform
it into a full path. Otherwise, we leave the program path untouched.
This mimicks the behaviour seen on GDB (look at "openp" and
"attach_inferior", for example).
I am also submitting a testcase which exercises the scenario described
above. This test requires gdbserver to be executed in a different CWD
than the original, so I also created a helper function, "with_cwd" (on
testsuite/lib/gdb.exp), which takes care of cd'ing into and out of the
specified dir.
Built and regtested on BuildBot, without regressions.
gdb/ChangeLog:
2018-02-28 Sergio Durigan Junior <sergiodj@redhat.com>
Simon Marchi <simon.marchi@polymtl.ca>
* common/common-utils.c: Include "sys/stat.h".
(is_regular_file): Move here from "source.c"; change return
type to "bool".
* common/common-utils.h (is_regular_file): New prototype.
* common/pathstuff.c (contains_dir_separator): New function.
* common/pathstuff.h (contains_dir_separator): New prototype.
* source.c: Don't include "sys/stat.h".
(is_regular_file): Move to "common/common-utils.c".
gdb/gdbserver/ChangeLog:
2018-02-28 Sergio Durigan Junior <sergiodj@redhat.com>
* server.c: Include "filenames.h" and "pathstuff.h".
(program_name): Delete variable.
(program_path): New anonymous class.
(get_exec_wrapper): Use "program_path" instead of
"program_name".
(handle_v_run): Likewise.
(captured_main): Likewise.
(process_serial_event): Likewise.
gdb/testsuite/ChangeLog:
2018-02-28 Sergio Durigan Junior <sergiodj@redhat.com>
* gdb.server/abspath.exp: New file.
* lib/gdb.exp (with_cwd): New procedure.
In patch
Add test for load command
3275ef4774
I removed gdb_is_target_remote_prompt, but did not realize it was used
in mi_is_target_remote. This makes the gdb.mi/mi-nonstop.exp crash, for
example:
ERROR: (DejaGnu) proc "gdb_is_target_remote_prompt {[(]gdb[)]
}" does not exist.
The error code is TCL LOOKUP COMMAND gdb_is_target_remote_prompt
The info on the error is:
invalid command name "gdb_is_target_remote_prompt"
while executing
"::tcl_unknown gdb_is_target_remote_prompt {[(]gdb[)]
}"
("uplevel" body line 1)
invoked from within
"uplevel 1 ::tcl_unknown $args"
This patch restores it.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (gdb_is_target_1): Add prompt_regexp parameter and
use it.
(gdb_is_target_remote_prompt): New proc.
(gdb_is_target_remote): Use gdb_is_target_remote_prompt.
(gdb_is_target_native): Pass prompt parameter to
gdb_is_target_1.
There doesn't seem to by any test for the load command. I suggest to
add this test, so that we can have a minimum of confidence we don't
break it completely while refactoring the code that implements it.
gdb/testsuite/ChangeLog:
* gdb.base/load-command.c: New file.
* gdb.base/load-command.exp: New file.
* lib/gdb.exp (gdb_is_target_remote_prompt): Rename to...
(gdb_is_target_1): ...this, and generalize for other targets
than just remote.
(gdb_is_target_remote): Use gdb_is_target_1.
(gdb_is_target_native): use gdb_is_target_1.
PR python/16497 notes that using "bt" with a positive argument prints
the wrong number of frames when a frame filter is in use. Also, in this
case, the non-frame-filter path will print a message about "More stack
frames" when there are more; but this is not done in the frame-filter
case.
The first problem is that backtrace_command_1 passes the wrong value
to apply_ext_lang_frame_filter -- that function takes the final
frame's number as an argument, but backtrace_command_1 passes the
count, which is off by one.
The solution to the second problem is to have the C stack-printing
code stop at the correct number of frames and then print the message.
Tested using the buildbot.
ChangeLog
2018-02-26 Tom Tromey <tom@tromey.com>
PR python/16497:
* stack.c (backtrace_command_1): Set PRINT_MORE_FRAMES flag. Fix
off-by-one in py_end computation.
* python/py-framefilter.c (gdbpy_apply_frame_filter): Handle
PRINT_MORE_FRAMES.
* extension.h (enum frame_filter_flags) <PRINT_MORE_FRAMES>: New
constant.
2018-02-26 Tom Tromey <tom@tromey.com>
PR python/16497:
* gdb.python/py-framefilter.exp: Update test.
This changes dwarf2read to understand DW_TAG_variant_part and
DW_TAG_variant.
Note that DW_AT_discr_list is not handled. I did not need this for
Rust. I imagine this should not be too hard to add later, should
someone need it. Meanwhile I have gdb emit a complaint if it is seen.
There is a lurking issue concerning the placement of the discriminant
in the DWARF. For Rust, I ended up following the letter of the
standard and having the discriminant be a child of the
DW_TAG_variant_part. However, GCC's Ada support does not do this.
Pierre-Marie filed this with the DWARF committee:
http://dwarfstd.org/ShowIssue.php?issue=180123.1
However as that is read-only, if you have comments you might consider
adding them to the GCC bug:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83935
Finally, there is a DWARF extension lurking in here. In Rust, a
univariant enum will not have a discriminant. However, in order to
unify the representation of all data-carrying enums, I've made LLVM
(and my forthcoming rustc patch) emit a univariant enum using a
DW_TAG_variant with a single variant part and without DW_AT_discr.
The lack of this DW_AT_discr is the extension. I will submit an issue
on dwarfstd.org about this.
2018-02-26 Tom Tromey <tom@tromey.com>
* dwarf2read.c (struct variant_field): New.
(struct nextfield) <variant>: New field.
(dwarf2_add_field): Handle DW_TAG_variant_part.
(dwarf2_attach_fields_to_type): Attach a discriminant_info to a
discriminated union.
(read_structure_type): Handle DW_TAG_variant_part.
(handle_struct_member_die): New function, extracted from
process_structure_scope. Handle DW_TAG_variant.
(process_structure_scope): Handle discriminated unions. Call
handle_struct_member_die.
2018-02-26 Tom Tromey <tom@tromey.com>
* gdb.dwarf2/variant.c: New file.
* gdb.dwarf2/variant.exp: New file.
A Rust enum is, essentially, a discriminated union. Currently the
Rust language support handles Rust enums locally, in rust-lang.c.
However, because I am changing the Rust compiler to use
DW_TAG_variant* to represent enums, it seemed better to have a single
internal representation for Rust enums in gdb.
This patch implements this idea by moving the current Rust enum
handling code to dwarf2read. This allows the simplification of some
parts of rust-lang.c as well.
2018-02-26 Tom Tromey <tom@tromey.com>
* rust-lang.h (rust_last_path_segment): Declare.
* rust-lang.c (rust_last_path_segment): Now public. Change
contract.
(struct disr_info): Remove.
(RUST_ENUM_PREFIX, RUST_ENCODED_ENUM_REAL)
(RUST_ENCODED_ENUM_HIDDEN, rust_union_is_untagged)
(rust_get_disr_info, rust_tuple_variant_type_p): Remove.
(rust_enum_p, rust_enum_variant): New function.
(rust_underscore_fields): Remove "offset" parameter.
(rust_print_enum): New function.
(rust_val_print) <TYPE_CODE_UNION>: Remove enum code.
<TYPE_CODE_STRUCT>: Call rust_print_enum when appropriate.
(rust_print_struct_def): Add "for_rust_enum" parameter. Handle
enums.
(rust_internal_print_type): New function, from rust_print_type.
Remove enum code.
(rust_print_type): Call rust_internal_print_type.
(rust_evaluate_subexp) <STRUCTOP_ANONYMOUS, STRUCTOP_STRUCT>:
Update enum handling.
* dwarf2read.c (struct dwarf2_cu) <rust_unions>: New field.
(rust_fully_qualify, alloc_discriminant_info, quirk_rust_enum)
(rust_union_quirks): New functions.
(process_full_comp_unit, process_full_type_unit): Call
rust_union_quirks.
(process_structure_scope): Update rust_unions if necessary.
2018-02-26 Tom Tromey <tom@tromey.com>
* gdb.rust/simple.exp: Accept more possible results in enum test.
I noticed some failures of some buildbot slaves, e.g.:
FAIL: gdb.cp/nested-types.exp: ptype S10 (limit = 1) // wrong nested type enum definition: enum S10::E10 {S10::A10, S10::B10, S10::C10};
The issue is that they have an older gcc (not c++11 by default?) that
doesn't emit the enum underlying type information. When the
enum type is printed by ptype, it looks like this:
enum S10::E10 {S10::A10, S10::B10, S10::C10};
instead of this on older gccs:
enum S10::E10 : unsigned int {S10::A10, S10::B10, S10::C10};
The regex that matches this is in cp_test_ptype_class, and is
enum $nested_name (: (unsigned )?int)? \{
If the "unsigned int" portion is not present, then it requires the
string to have two spaces between the enum name and opening bracket.
The fix is simply to move the trailing space inside the ? group.
gdb/testsuite/ChangeLog:
* lib/cp-support.exp (cp_test_ptype_class): Move space inside
parentheses.
Fix a commit 883fd55ab1 ("Record nested types") issue:
ERROR: tcl error sourcing .../gdb/testsuite/gdb.cp/nested-types.exp.
ERROR: can't read "actual_linejj": no such variable
while executing
"append txt " definition: $actual_linejj""
(procedure "cp_test_ptype_class" line 324)
invoked from within
"cp_test_ptype_class $name "ptype $name (limit = $limit)" $key $name $children" (procedure "test_nested_limit" line 28)
invoked from within
"test_nested_limit -1 false"
(file ".../gdb/testsuite/gdb.cp/nested-types.exp" line 310)
invoked from within
"source .../gdb/testsuite/gdb.cp/nested-types.exp"
("uplevel" body line 1)
invoked from within
"uplevel #0 source .../gdb/testsuite/gdb.cp/nested-types.exp"
invoked from within
"catch "uplevel #0 source $test_file_name""
testcase .../gdb/testsuite/gdb.cp/nested-types.exp completed in 9 seconds
caused by $actual_line having been accidentally referred to as
$actual_linejj in one place.
gdb/testsuite/
* lib/cp-support.exp (cp_test_ptype_class): Fix a typo in the
name of a variable: $actual_linejj -> $actual_line.
In gdb.btrace/buffer-size.exp we explicitly ask for the BTS recording format.
This may lead to spurious fails on systems where PT is being used by some other
process at the same time.
Set both PT and BTS buffer sizes to 1 and check that whatever recording format
is used will use a 4KB buffer.
testsuite/
* gdb.btrace/buffer-size.exp: Do not force BTS.
I ran into a GDB crash in gdb.base/bp-cmds-continue-ctrl-c.exp in my
multi-target branch, which turns out exposed a bug that exists in
master too.
That testcase has a breakpoint with a "continue" command associated.
Then the breakpoint is constantly being hit. At the same time, the
testcase is continualy interrupting the program with Ctrl-C, and
re-resuming it, in a loop.
Running that testcase manually under Valgrind, after a few sequences
of 'Ctrl-C' + 'continue', I got:
Breakpoint 1, Quit
(gdb) ==21270== Invalid read of size 8
==21270== at 0x4D8185: pyuw_this_id(frame_info*, void**, frame_id*) (py-unwind.c:461)
==21270== by 0x6D426A: compute_frame_id(frame_info*) (frame.c:505)
==21270== by 0x6D43B7: get_frame_id(frame_info*) (frame.c:537)
==21270== by 0x84F3B8: scoped_restore_current_thread::scoped_restore_current_thread() (thread.c:1678)
==21270== by 0x718E3D: fetch_inferior_event(void*) (infrun.c:4076)
==21270== by 0x7067C9: inferior_event_handler(inferior_event_type, void*) (inf-loop.c:43)
==21270== by 0x45BEF9: handle_target_event(int, void*) (linux-nat.c:4419)
==21270== by 0x6C4255: handle_file_event(file_handler*, int) (event-loop.c:733)
==21270== by 0x6C47F8: gdb_wait_for_event(int) (event-loop.c:859)
==21270== by 0x6C3666: gdb_do_one_event() (event-loop.c:322)
==21270== by 0x6C3712: start_event_loop() (event-loop.c:371)
==21270== by 0x746801: captured_command_loop() (main.c:329)
==21270== Address 0x0 is not stack'd, malloc'd or (recently) free'd
==21270==
==21270==
==21270== Process terminating with default action of signal 11 (SIGSEGV): dumping core
==21270== Access not within mapped region at address 0x0
==21270== at 0x4D8185: pyuw_this_id(frame_info*, void**, frame_id*) (py-unwind.c:461)
==21270== by 0x6D426A: compute_frame_id(frame_info*) (frame.c:505)
==21270== by 0x6D43B7: get_frame_id(frame_info*) (frame.c:537)
==21270== by 0x84F3B8: scoped_restore_current_thread::scoped_restore_current_thread() (thread.c:1678)
==21270== by 0x718E3D: fetch_inferior_event(void*) (infrun.c:4076)
==21270== by 0x7067C9: inferior_event_handler(inferior_event_type, void*) (inf-loop.c:43)
==21270== by 0x45BEF9: handle_target_event(int, void*) (linux-nat.c:4419)
==21270== by 0x6C4255: handle_file_event(file_handler*, int) (event-loop.c:733)
==21270== by 0x6C47F8: gdb_wait_for_event(int) (event-loop.c:859)
==21270== by 0x6C3666: gdb_do_one_event() (event-loop.c:322)
==21270== by 0x6C3712: start_event_loop() (event-loop.c:371)
==21270== by 0x746801: captured_command_loop() (main.c:329)
==21270== If you believe this happened as a result of a stack
==21270== overflow in your program's main thread (unlikely but
==21270== possible), you can try to increase the size of the
==21270== main thread stack using the --main-stacksize= flag.
==21270== The main thread stack size used in this run was 8388608.
==21270==
Above, when we get to compute_frame_id, fi->unwind is non-NULL,
meaning, we found an unwinder, in this case the Python unwinder, but
somehow, fi->prologue_cache is left NULL. pyuw_this_id then crashes
because it assumes fi->prologue_cache is non-NULL:
static void
pyuw_this_id (struct frame_info *this_frame, void **cache_ptr,
struct frame_id *this_id)
{
*this_id = ((cached_frame_info *) *cache_ptr)->frame_id;
^^^^^^^^^^
'*cache_ptr' here is 'fi->prologue_cache'.
There's a quit() call in pyuw_sniffer that I believe is the one that
sometimes triggers the crash above. The crash can be reproduced
easily with this hack to force a quit out of the python unwinder:
--- a/gdb/python/py-unwind.c
+++ b/gdb/python/py-unwind.c
@@ -497,6 +497,8 @@ pyuw_sniffer (const struct frame_unwind *self, struct frame_info *this_frame,
struct gdbarch *gdbarch = (struct gdbarch *) (self->unwind_data);
cached_frame_info *cached_frame;
+ quit ();
+
gdbpy_enter enter_py (gdbarch, current_language);
TRACE_PY_UNWIND (3, "%s (SP=%s, PC=%s)\n", __FUNCTION__,
After that quit is thrown, any subsequent operation that involves
unwinding results in GDB crashing with SIGSEGV like above.
The problem is that this commit:
commit 30a9c02fef
CommitDate: Sun Oct 8 23:16:42 2017 -0600
Subject: Remove cleanup from frame_prepare_for_sniffer
missed that we need to call frame_cleanup_after_sniffer before
rethrowing the exception too.
Without the fix, the "bt" added to
gdb.base/bp-cmds-continue-ctrl-c.exp in this commit makes GDB crash:
Running src/gdb/testsuite/gdb.base/bp-cmds-continue-ctrl-c.exp ...
ERROR: Process no longer exists
gdb/ChangeLog:
2018-02-14 Pedro Alves <palves@redhat.com>
* frame-unwind.c (frame_unwind_try_unwinder): Always call
frame_cleanup_after_sniffer on exception.
gdb/testsuite/ChangeLog:
2018-02-14 Pedro Alves <palves@redhat.com>
* gdb.base/bp-cmds-continue-ctrl-c.exp (do_test): Test "bt" after
getting a "Quit".
As reported here:
https://sourceware.org/ml/gdb/2018-02/msg00019.html
the type of values representing static members that are optimized out is
wrong. It currently assigns the type of the containing class rather
than the type of the field. This patch fixes that.
I found a place in m-static.exp already dealing with optimized out
static members, so I just added some gdb_test there.
gdb/ChangeLog:
* value.c (value_static_field): Assign field type instead of
containing type when returning an optimized out value.
gdb/testsuite/ChangeLog:
* gdb.cp/m-static.exp: Check type of optimized out static
member.
The 'cleanup' proc has been removed from dejagnu (Feb 15 2016). The
proc has not done anything useful since at least 2001 so removing
these calls should be harmless.
gdb/testsuite/ChangeLog:
* config/sid.exp (gdb_target_sid): Remove use of cleanup.
* config/sim.exp (gdb_target_sim): Remove use of cleanup.
Make the MI variable object expression evaluation, with the
-var-evaluate-expression command, recursively call pretty printers, to
match the output of normal expression printing.
Consider the following code:
struct Foo { int val; };
struct Wrapper { Foo foo; };
int main() {
Wrapper w;
w.foo.val = 23;
}
and this pretty printer file:
import gdb.printing
class FooPrinter:
def __init__(self, val):
self.val = val
def to_string(self):
return "Foo" + str(self.val["val"])
class WrapperPrinter:
def __init__(self, val):
self.val = val
def to_string(self):
return self.val["foo"]
test_printer = gdb.printing.RegexpCollectionPrettyPrinter("test")
test_printer.add_printer('Foo', '^Foo$', FooPrinter)
test_printer.add_printer('Wrapper', '^Wrapper$', WrapperPrinter)
gdb.printing.register_pretty_printer(None, test_printer)
Setting a breakpoint at the end of the function, we call the following commands:
-enable-pretty-printing
^done
-var-create var_w @ w
^done,name="var_w",numchild="0",value="{val = 23}",type="Wrapper",dynamic="1",has_more="0"
-var-create var_w_foo @ w.foo
^done,name="var_w_foo",numchild="0",value="Foo23",type="Foo",dynamic="1",has_more="0"
-var-evaluate-expression var_w
^done,value="{val = 23}"
-var-evaluate-expression var_w_foo
^done,value="Foo23"
-data-evaluate-expression w
^done,value="Foo23"
-data-evaluate-expression w.foo
^done,value="Foo23"
So, in the -var-evaluate-expression var_w case, we print the "raw" value
of w.foo, while in the -data-evaluate-expression w case, we print the
pretty printed w.foo value. After this patch, all of the above print
"Foo23".
gdb/ChangeLog:
* varobj.c (varobj_formatted_print_options): Allow recursive
pretty printing if pretty printing is enabled.
gdb/testsuite/ChangeLog:
* gdb.python/py-prettyprint.c
(struct to_string_returns_value_inner,
struct to_string_returns_value_wrapper): New.
(main): Add tsrvw variable.
* gdb.python/py-prettyprint.py (ToStringReturnsValueInner,
ToStringReturnsValueWrapper): New classes.
(register_pretty_printers): Register new pretty-printers.
* gdb.python/py-prettyprint.exp (run_lang_tests): Test printing
recursive pretty printer.
* gdb.python/py-mi.exp: Likewise.
There is existing logic in C/C++ expression parsing to avoid classifying
names as a filename when they are a field on the this object. This
change extends this logic to also avoid classifying names after a
struct-op (-> or .) as a filename, which otherwise causes a syntax
error.
Thus, it is now possible in the file
#include <map>
struct D {
void map();
}
D d;
to call
(gdb) print d.map()
where previously this would have been a syntax error.
Tested on gdb.cp/*.exp
gdb/ChangeLog:
* c-exp.y (lex_one_token, classify_name, yylex): Don't classify
names after a structop as a filename
gdb/testsuite/ChangeLog:
* gdb.cp/filename.cc, gdb.cp/filename.exp: Test that member
functions with the same name as an include file are parsed
correctly.
I see some test fails in gdb.base/attach.exp when gdb is configured
--with-sysroot=/.
FAIL: gdb.base/attach.exp: attach2, with no file
FAIL: gdb.base/attach.exp: load file manually, after attach2 (re-read) (got interactive prompt)
FAIL: gdb.base/attach.exp: attach when process' a.out not in cwd
If gdb is configured this way, sysroot is "/" in default, and if binfile
is a absolute path, the regexp pattern $sysroot$escapedbinfile is
incorrect.
There are different ways to fix it, but I don't want to complicate the
test, so I choose this naive way.
gdb/testsuite:
2018-02-01 Yao Qi <yao.qi@linaro.org>
* gdb.base/attach.exp (do_attach_tests): Set sysroot to
"\[^\r\n\]*".
One of conditions in skip_prologue() was never visited if there was mflr
instruction that moves the link register to a register different than r0.
This condition expects non shifted value of `lr_reg`. Previously offset
of link register was never saved for registers different than r0.
gdb/ChangeLog:
2018-01-31 Nikola Prica <nikola.prica@rt-rk.com>
* rs6000-tdep.c (skip_prologue): Remove shifting for lr_reg and
assign shifted lr_reg to fdata->lr_register when lr_reg is set.
gdb/testsuite/ChangeLog:
2018-01-31 Nikola Prica <nikola.prica@rt-rk.com>
* gdb.arch/powerpc-prologue-frame.s: New file.
* gdb.arch/powerpc-prologue-frame.c: Likewise.
* gdb.arch/powerpc-prologue-frame.exp: Likewise.
Following my previous commit which add support for stopping at start of
exception handler, this commit adds required gdb-mi support for this
feature.
gdb/ChangeLog:
* mi/mi-cmd-catch.c (mi_cmd_catch_handlers): New function.
* mi/mi-cmds.c (mi_cmds): Add catch-handlers command.
* mi/mi-cmds.h (mi_cmd_catch_handlers): Add external declaration.
* NEWS: Document "-catch-handlers" command.
gdb/doc/ChangeLog:
* gdb.texinfo (Ada Exception gdb/mi Catchpoints): Add
documentation for new "-catch-handlers" command.
gdb/testsuite/ChangeLog:
* gdb.ada/mi_catch_ex_hand.exp: New testcase.
* gdb.ada/mi_catch_ex_hand/foo.adb: New file.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
* gdb.ada/mi_catch_assert.exp: New testcase.
* gdb.ada/mi_catch_assert/bla.adb: New file.
* gdb.ada/mi_catch_assert/pck.ads: New file.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
* gdb.ada/catch_assert_if.exp: New testcase.
* gdb.ada/catch_assert_if/bla.adb: New file.
* gdb.ada/catch_assert_if/pck.ads: New file.
Tested on x86_64-linux.
Using the following Ada declarations (the same as in
gdb.ada/dyn_stride.exp)...
subtype Small_Type is Integer range L .. U;
type Record_Type (I : Small_Type := L) is record
S : String (1 .. I);
end record;
type Array_Type is array (Integer range <>) of Record_Type;
A1 : Array_Type :=
(1 => (I => U, S => (others => ASCII.NUL)),
2 => (I => 1, S => "A"),
3 => (I => 2, S => "AB"));
... where "L" and "U" are variables, trying to apply the repeat
operator to "A1(1)" yields to an internal error:
| (gdb) print a1(1)@3
| $5 = /[...]/gdbtypes.c:4883: internal-error: type* copy_type(const type*):
| Assertion `TYPE_OBJFILE_OWNED (type)' failed.
What happens first is that the ada-lang module evaluated the "A1(1)"
sub-expression returning a structure where "I" (one of the fields
in that structure) has a type which is dynamic, because it is
a range type whose bounds are not statically known.
Next, we apply the repeat ('@') operator, which is done via
allocate_repeat_value, which creates an array type with the correct
bounds to associate to our value, by calling lookup_array_range_type:
| struct type *
| lookup_array_range_type (struct type *element_type,
| LONGEST low_bound, LONGEST high_bound)
| {
| struct gdbarch *gdbarch = get_type_arch (element_type);
| struct type *index_type = builtin_type (gdbarch)->builtin_int;
| struct type *range_type
| = create_static_range_type (NULL, index_type, low_bound, high_bound);
|
| return create_array_type (NULL, element_type, range_type);
| }
As we can see, this creates an array type whose index type is
always owned by the gdbarch. This is where the problem lies.
Next, we use that type to construct a struct value. That value
then gets passed to the valprint module, which then checks
whether our object is dynamic or not. And because field "I" above
had a dynamic range type, we end up determining by association
that the artificial repeat array itself is also dynamic. So
we attempt to resolve the type, which leads to trying to copying
that type. And because the artifical array created by
lookup_array_range_type has an index which is not objfile-owned,
we trip the assertion.
This patch fixes the issue by enhancing lookup_array_range_type
to create an index type which has the same owner as the element
type.
gdb/ChangeLog:
* gdbtypes.c (lookup_array_range_type): Make sure the array's
index type is objfile-owned if the element type is as well.
gdb/testsuite/ChangeLog:
* testsuite/gdb.ada/dyn_stride.exp: Add "print a1(1)@3" test.
In my multi-target branch I ran into problems with GDB's terminal
handling that exist in master as well, with multi-inferior debugging.
This patch adds a testcase for said problems
(gdb.multi/multi-term-settings.exp), fixes the problems, fixes PR
gdb/13211 as well (and adds a testcase for that too,
gdb.base/interrupt-daemon.exp).
The basis of the problem I ran into is the following. Consider a
scenario where you have:
- inferior 1 - started with "attach", process is running on some
other terminal.
- inferior 2 - started with "run", process is sharing gdb's terminal.
In this scenario, when you stop/resume both inferiors, you want GDB to
save/restore the terminal settings of inferior 2, the one that is
sharing GDB's terminal. I.e., you want inferior 2 to "own" the
terminal (in target_terminal::is_ours/target_terminal::is_inferior
sense).
Unfortunately, that's not what you get currently. Because GDB doesn't
know whether an attached inferior is actually sharing GDB's terminal,
it tries to save/restore its settings anyway, ignoring errors. In
this case, this is pointless, because inferior 1 is running on a
different terminal, but GDB doesn't know better.
And then, because it is only possible to have the terminal settings of
a single inferior be in effect at a time, or make one inferior/pgrp be
the terminal's foreground pgrp (aka, only one inferior can "own" the
terminal, ignoring fork children here), if GDB happens to try to
restore the terminal settings of inferior 1 first, then GDB never
restores the terminal settings of inferior 2.
This patch fixes that and a few things more along the way:
- Moves enum target_terminal::terminal_state out of the
target_terminal class (it's currently private) and makes it a
scoped enum so that it can be easily used elsewhere.
- Replaces the inflow.c:terminal_is_ours boolean with a
target_terminal_state variable. This allows distinguishing is_ours
and is_ours_for_output states. This allows finally making
child_terminal_ours_1 do something with its "output_only"
parameter.
- Makes each inferior have its own copy of the
is_ours/is_ours_for_output/is_inferior state.
- Adds a way for GDB to tell whether the inferior is sharing GDB's
terminal. Works best on Linux and Solaris; the fallback works just
as well as currently.
- With that, we can remove the inf->attach_flag tests from
child_terminal_inferior/child_terminal_ours.
- Currently target_ops.to_ours is responsible for both saving the
current inferior's terminal state, and restoring gdb's state.
Because each inferior has its own terminal state (possibly handled
by different targets in a multi-target world, even), we need to
split the inferior-saving part from the gdb-restoring part. The
patch adds a new target_ops.to_save_inferior target method for
that.
- Adds a new target_terminal::save_inferior() function, so that
sequences like:
scoped_restore_terminal_state save_state;
target_terminal::ours_for_output ();
... restore back inferiors that were
target_terminal_state::is_inferior before back to is_inferior, and
leaves inferiors that were is_ours alone.
- Along the way, this adds a default implementation of
target_pass_ctrlc to inflow.c (for inf-child.c), that handles
passing the Ctrl-C to a process running on GDB's terminal or to
some other process otherwise.
- Similarly, adds a new target default implementation of
target_interrupt, for the "interrupt" command. The current
implementation of this hook in inf-ptrace.c kills the whole process
group, but that's incorrect/undesirable because we may not be
attached to all processes in the process group. And also, it's
incorrect because inferior_process_group() doesn't really return
the inferior's real process group id if the inferior is not a
process group leader... This is the cause of PR gdb/13211 [1],
which this patch fixes. While at it, that target method's "ptid"
parameter is eliminated, because it's not really used.
- A new test is included that exercises and fixes PR gdb/13211, and
also fixes a GDB issue reported on stackoverflow that I ran into
while working on this [2]. The problem is similar to PR gdb/13211,
except that it also triggers with Ctrl-C. When debugging a daemon
(i.e., a process that disconnects from the controlling terminal and
is not a process group leader, then Ctrl-C doesn't work, you just
can't interrupt the inferior at all, resulting in a hung debug
session. The problem is that since the inferior is no longer
associated with gdb's session / controlling terminal, then trying
to put the inferior in the foreground fails. And so Ctrl-C never
reaches the inferior directly. pass_signal is only used when the
inferior is attached, but that is not the case here. This is fixed
by the new child_pass_ctrlc. Without the fix, the new
interrupt-daemon.exp testcase fails with timeout waiting for a
SIGINT that never arrives.
[1] PR gdb/13211 - Async / Process group and interrupt not working
https://sourceware.org/bugzilla/show_bug.cgi?id=13211
[2] GDB not reacting Ctrl-C when after fork() and setsid()
https://stackoverflow.com/questions/46101292/gdb-not-reacting-ctrl-c-when-after-fork-and-setsid
Note this patch does _not_ fix:
- PR gdb/14559 - The 'interrupt' command does not work if sigwait is in use
https://sourceware.org/bugzilla/show_bug.cgi?id=14559
- PR gdb/9425 - When using "sigwait" GDB doesn't trap SIGINT. Ctrl+C terminates program when should break gdb.
https://sourceware.org/bugzilla/show_bug.cgi?id=9425
The only way to fix that that I know of (without changing the kernel)
is to make GDB put inferiors in a separate session (create a
pseudo-tty master/slave pair, make the inferior run with the slave as
its terminal, and have gdb pump output/input on the master end).
gdb/ChangeLog:
2018-01-30 Pedro Alves <palves@redhat.com>
PR gdb/13211
* config.in, configure: Regenerate.
* configure.ac: Check for getpgid.
* go32-nat.c (go32_pass_ctrlc): New.
(go32_target): Install it.
* inf-child.c (inf_child_target): Install
child_terminal_save_inferior, child_pass_ctrlc and
child_interrupt.
* inf-ptrace.c (inf_ptrace_interrupt): Delete.
(inf_ptrace_target): No longer install it.
* infcmd.c (interrupt_target_1): Adjust.
* inferior.h (child_terminal_save_inferior, child_pass_ctrlc)
(child_interrupt): Declare.
(inferior::terminal_state): New.
* inflow.c (struct terminal_info): Update comments.
(inferior_process_group): Delete.
(terminal_is_ours): Delete.
(gdb_tty_state): New.
(child_terminal_init): Adjust.
(is_gdb_terminal, sharing_input_terminal_1)
(sharing_input_terminal): New functions.
(child_terminal_inferior): Adjust. Use sharing_input_terminal.
Set the process's actual process group in the foreground if
possible. Handle is_ours_for_output/is_ours distinction. Don't
mark terminal as the inferior's if not sharing GDB's terminal.
Don't check attach_flag.
(child_terminal_ours_for_output, child_terminal_ours): Adjust to
pass down a target_terminal_state.
(child_terminal_save_inferior): New, factored out from ...
(child_terminal_ours_1): ... this. Handle
target_terminal_state::is_ours_for_output.
(child_interrupt, child_pass_ctrlc): New.
(inflow_inferior_exit): Clear the inferior's terminal_state.
(copy_terminal_info): Copy the inferior's terminal state.
(_initialize_inflow): Remove reference to terminal_is_ours.
* inflow.h (inferior_process_group): Delete.
* nto-procfs.c (nto_handle_sigint, procfs_interrupt): Adjust.
* procfs.c (procfs_target): Don't install procfs_interrupt.
(procfs_interrupt): Delete.
* remote.c (remote_serial_quit_handler): Adjust.
(remote_interrupt): Remove ptid parameter. Adjust.
* target-delegates.c: Regenerate.
* target.c: Include "terminal.h".
(target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this.
(target_terminal::init): Adjust.
(target_terminal::inferior): Adjust to per-inferior
terminal_state.
(target_terminal::restore_inferior, target_terminal_is_ours_kind): New.
(target_terminal::ours, target_terminal::ours_for_output): Use
target_terminal_is_ours_kind.
(target_interrupt): Remove ptid parameter. Adjust.
(default_target_pass_ctrlc): Adjust.
* target.h (target_ops::to_terminal_save_inferior): New field.
(target_ops::to_interrupt): Remove ptid_t parameter.
(target_interrupt): Remove ptid_t parameter. Update comment.
(target_pass_ctrlc): Update comment.
* target/target.h (target_terminal_state): New scoped enum,
factored out of ...
(target_terminal::terminal_state): ... here.
(target_terminal::inferior): Update comments.
(target_terminal::restore_inferior): New.
(target_terminal::is_inferior, target_terminal::is_ours)
(target_terminal::is_ours_for_output): Adjust.
(target_terminal::scoped_restore_terminal_state): Adjust to
rename, and call restore_inferior() instead of inferior().
(target_terminal::scoped_restore_terminal_state::m_state): Change
type.
(target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this and change type.
gdb/gdbserver/ChangeLog:
2018-01-30 Pedro Alves <palves@redhat.com>
PR gdb/13211
* target.c (target_terminal::terminal_state): Rename to ...
(target_terminal::m_terminal_state): ... this.
gdb/testsuite/ChangeLog:
2018-01-30 Pedro Alves <palves@redhat.com>
PR gdb/13211
* gdb.base/interrupt-daemon.c: New.
* gdb.base/interrupt-daemon.exp: New.
* gdb.multi/multi-term-settings.c: New.
* gdb.multi/multi-term-settings.exp: New.
The last test of this testcase fails when run on Ubuntu 16.04 using
the system compiler (16.04):
FAIL: gdb.base/break.exp: verify that they were cleared
This is because the testcase expected that a breakpoint on line 47 of break.c...
printf ("%d\n", factorial (atoi ("6"))); /* set breakpoint 1 here */
... would actually be inserted on an instruction belonging to
that line. However, what actually happens is that system GCC on
that version of Ubuntu ends up inlining everything, including
the call to printf, thus reporting every instruction of generated
for this line of code as belonging to a different function. As
a result, GDB ends up insering the breakpoint on the next line
of code, which is line 49:
(gdb) break break.c:$l
Breakpoint 3 at 0x4005c1: file /[...]/gdb.base/break.c, line 49.
This causes a spurious failure in the "info break" test later on,
as it assumed that the breakpoint above is inserted on line 47:
gdb_test "info break" "$srcfile:$line" "verify that they were cleared"
This patch fixes the issue by saving the actual source location where
the breakpoint was inserted.
gdb/testsuite/ChangeLog:
* gdb.base/break.exp: Save the location where the breakpoint
on break.c:47 was actually inserted when debugging the version
compiled at -O2 and use it in the expected output of the "info
break" test performed soon after.
tested on x86_64-linux, with two configurations:
- Ubuntu 16.04 with the system compiler (breakpoint lands on line 49)
- Ubuntu 16.04 with GCC 7.3.1 (breakpoint lands on line 47)
Compiling GDB with a recent GCC exposes a problem:
../../gdb/typeprint.c: In function 'void whatis_exp(const char*, int)':
../../gdb/typeprint.c:515:12: warning: 'val' may be used uninitialized in this function [-Wmaybe-uninitialized]
real_type = value_rtti_type (val, &full, &top, &using_enc);
~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The warning is correct. There are indeed code paths that use
uninitialized 'val', leading to crashes. Inside the
value_rtti_indirect_type/value_rtti_type calls here in whatis_exp:
if (opts.objectprint)
{
if (((TYPE_CODE (type) == TYPE_CODE_PTR) || TYPE_IS_REFERENCE (type))
&& (TYPE_CODE (TYPE_TARGET_TYPE (type)) == TYPE_CODE_STRUCT))
real_type = value_rtti_indirect_type (val, &full, &top, &using_enc);
else if (TYPE_CODE (type) == TYPE_CODE_STRUCT)
real_type = value_rtti_type (val, &full, &top, &using_enc);
}
We reach those calls above with "set print object on", and then with
any of:
(gdb) whatis struct some_structure_type
(gdb) whatis struct some_structure_type *
(gdb) whatis struct some_structure_type &
because "whatis" with a type argument enters this branch:
/* The behavior of "whatis" depends on whether the user
expression names a type directly, or a language expression
(including variable names). If the former, then "whatis"
strips one level of typedefs, only. If an expression,
"whatis" prints the type of the expression without stripping
any typedef level. "ptype" always strips all levels of
typedefs. */
if (show == -1 && expr->elts[0].opcode == OP_TYPE)
{
which does not initialize VAL. Trying the above triggers crashes like
this:
(gdb) set print object on
(gdb) whatis some_structure_type
Thread 1 "gdb" received signal SIGSEGV, Segmentation fault.
0x00000000005dda90 in check_typedef (type=0x6120736573756170) at src/gdb/gdbtypes.c:2388
2388 int instance_flags = TYPE_INSTANCE_FLAGS (type);
...
This is a regression caused by a recent-ish refactoring of the code on
'whatis_exp', introduced by:
commit c973d0aa4a
Date: Mon Aug 21 11:34:32 2017 +0100
Fix type casts losing typedefs and reimplement "whatis" typedef stripping
Fix this by setting VAL to NULL in the "whatis TYPE" case, and
skipping fetching the dynamic type if there's no value to fetch it
from.
New tests included.
gdb/ChangeLog:
2018-01-22 Pedro Alves <palves@redhat.com>
Sergio Durigan Junior <sergiodj@redhat.com>
* typeprint.c (whatis_exp): Initialize "val" in the "whatis type"
case.
gdb/testsuite/ChangeLog:
2018-01-22 Pedro Alves <palves@redhat.com>
Sergio Durigan Junior <sergiodj@redhat.com>
* gdb.base/whatis.exp: Add tests for 'set print object on' +
'whatis <struct>' 'whatis <struct> *' and 'whatis <struct> &'.
Consider the following situation, where we have one file containing...
$ cat -n body.inc
1 i = i + 1;
... we include that file from some code, like so:
$ cat -n cat -n small.c
[...]
17 int
18 next (int i)
19 {
20 #include "body.inc"
21 return i;
22 }
When trying to insert a breakpoint on line 18, for instance:
(gdb) b small.c:18
Breakpoint 1 at 0x40049f: file body.inc, line 18.
^^
||
Here, the issue is that GDB reports the breakpoint to be in file
body.inc, which is true, but with the line number that corresponding
to the user-requested location, which is not correct.
Although the simple reproducer may look slightly artificial,
the above is simply one way to reproduce the same issue observed
when trying to insert a breakpoint on a function provided in
a .h files and then subsequently inlined in a C file.
What happens is the following:
1. We resolve the small.c:18 linespec into a symtab_and_line which
has "small.c" and 18 as the symtab and line number.
2. Next, we call skip_prologue_sal, which calculates the PC
past the prologue, and updates the symtab_and_line: PC,
but also symtab (now body.inc) and the new line (now 1).
3. However, right after that, we do:
/* Make sure the line matches the request, not what was
found. */
intermediate_results.sals[i].line = val.line;
We should either restore both symtab and line, or leave the actual
line to match the actual symtab. This patch chose the latter.
This introduces a few changes in a few tests, which required some
updates, but looking at those change, I believe them to be expected.
gdb/ChangeLog:
* linespec.c (create_sals_line_offset): Remove code that preserved
the symtab_and_line's line number.
gdb/testsuite/ChangeLog:
* gdb.base/break-include.c, gdb.base/break-include.inc,
gdb.base/break-include.exp: New files.
* gdb.base/ending-run.exp: Minor adaptations due to the breakpoint's
line number now being the actual line number where the breakpoint
was inserted.
* gdb.mi/mi-break.exp: Likewise.
* gdb.mi/mi-reverse.exp: Likewise.
* gdb.mi/mi-simplerun.exp: Ditto.
Tested on x86_64-linux.
When creating a varobj with -var-create a user can create either fixed
varobj, or floating varobj.
A fixed varobj will always be evaluated within the thread/frame/block in
which the varobj was created, if that thread/frame/block is no longer
available then the varobj is considered out of scope.
A floating varobj will always be evaluated within the current
thread/frame/block.
Despite never using them GDB was storing the thread/frame/block into a
floating varobj, and the thread-id would then be displayed when GDB
reported on the state of the varobj, this could confuse a user into
thinking that the thread-id was relevant.
This commit prevents GDB storing the thread/frame/block onto floating
varobj, and updates the few tests where this impacts the results.
gdb/ChangeLog:
* varobj.c (varobj_create): Don't set valid_block when creating a
floating varobj.
gdb/testsuite/ChangeLog:
* gdb.python/py-mi.exp: Don't expect a thread-id for floating
varobj.
* gdb.mi/mi-var-create-rtti.exp: Likewise.
Joel Brobecker
Weimin Pan
Tom Tromey
Keith Seitz
Andrew Burgess
Andreas Arnez
Stephen Roberts
Simon Marchi
Simon Marchi
Thomas Preud'homme
Sergio Durigan Junior
Maciej W. Rozycki
John Baldwin
Markus Metzger
Pedro Alves
Simon Marchi
Leszek Swirski via gdb-patches
Leszek Swirski
Yao Qi
Nikola Prica
Xavier Roirand