It is not necessary to pass an empty string to mi_gdb_start, passing
the empty string is equivalent to passing no arguments, which is what
we do everywhere else (that we don't need to specify an actual
argument).
The only place we use 'mi_gdb_start ""' is in
gdb.mi/mi-break-qualified.exp, so in this commit I just replace that
with a call to 'mi_gdb_start' - just for consistency.
There should be no change in what is tested after this commit.
After this previous commit I was thinking about the API of
mi_gdb_start. I felt that the idea of passing flags as separate
arguments and using 'args' to gather these into a list, though clever,
was not an intuitive API.
In this commit I modify mi_gdb_start so that it expects a single
argument, which should be a list of flags. Thus, where we previously
would have said:
mi_gdb_start separate-mi-tty separate-inferior-tty
We would now say:
mi_gdb_start { separate-mi-tty separate-inferior-tty }
However, it turns out we never actually call mi_gdb_start passing two
arguments in this way at all. We do in some places do this:
mi_gdb_start separate-inferior-tty
But that's fine, a single string like this works equally well as a
single item list, so this will not need updating.
There is also one place where we do this:
eval mi_gdb_start $start_ops
where $start_ops is a list that might contains 0, 1, or 2 items. The
eval here is used to expand the $start_ops list so mi_gdb_start sees
the list contents as separate arguments. In this case we just need to
drop the use of eval.
I think that the new API is more intuitive, but others might
disagree, in which case I can drop this change.
There should be no change in what is tested after this commit.
When running with the native-extended-gdbserver board, I currently see
one failure in gdb.mi/mi-exec-run.exp:
FAIL: gdb.mi/mi-exec-run.exp: inferior-tty=separate: mi=separate: force-fail=0: breakpoint hit reported on console (timeout)
In this test the MI interface should be started in a separate tty,
which means we should have a CLI tty and an MI tty, however, this is
not happening. Instead GDB is just started in MI mode and there is no
CLI tty.
The test script tries to switch between the CLI an MI terminals and
look for some expected output on each, however, as there is no CLI
terminal the expected output never arrives, and the test times out.
It turns out that this is not a GDB problem, rather, this is an issue
with argument passing within the test script.
The proc default_mi_gdb_start expects to take a set of flags (strings)
as arguments, each of flag is expected to be a separate argument. The
default_mi_gdb_start proc collects all its arguments into a list using
the special 'args' parameter name, and then iterates over this list to
see which flags were passed.
In mi_gdb_start, which forwards to default_mi_gdb_start, the arguments
are also gathered into the 'args' parameter list, but are then
expanded back to be separate arguments using the eval trick, i.e.:
proc mi_gdb_start { args } {
return [eval default_mi_gdb_start $args]
}
This ensures that when we arrive in default_mi_gdb_start each flag is
a separate argument, rather than appearing as a single list containing
all arguments.
When using the native-extended-gdbserver board however, the file
boards/native-extended-gdbserver.exp is loaded, and this file replaces
the default mi_gdb_start with its own version.
This new mi_gdb_start also gathers the arguments into an 'args' list,
but forgets to expand the arguments out using the eval trick.
As a result, when using the native-extended-gdbserver board, by the
time we get to default_mi_gdb_start, we end up with the args list
containing a single item, which is a list containing all the arguments
the user passed.
What this means is that if the user passes two arguments, then, in
default_mi_gdb_start, instead of seeing two separate arguments, we see
a single argument made by concatenating the two arguments together.
The only place this is a problem is in the test mi-exec-run.exp,
which (as far as I can see) is the only test where we might try to
pass both arguments at the same time. Currently we think we passed
both arguments to mi_gdb_start, but mi_gdb_start behaves as if no
arguments were passed.
This commit fixes the problem by making use of the eval trick within
the native-extended-gdbserver version of mi_gdb_start. After this,
the FAIL listed at the top of this message is resolved.
When running the gdb.mi/mi-exec-run.exp test using the
native-extended-gdbserver I see failures like this:
FAIL: gdb.mi/mi-exec-run.exp: inferior-tty=main: mi=main: force-fail=1: run failure detected
FAIL: gdb.mi/mi-exec-run.exp: inferior-tty=main: mi=separate: force-fail=1: run failure detected
FAIL: gdb.mi/mi-exec-run.exp: inferior-tty=separate: mi=separate: force-fail=1: run failure detected
There's a race condition here, so you might see a slightly different
set of failures, but I always see some from the 'run failure detected'
test.
NOTE: I also see an additional test failure:
FAIL: gdb.mi/mi-exec-run.exp: inferior-tty=separate: mi=separate: force-fail=0: breakpoint hit reported on console (timeout)
but that is a completely different issue, and is not being addressed
in this commit.
The problem for the 'run failure detected' test is that we end up
in gdb_expect looking for output from two spawn-ids, one from
gdbserver, and one from gdb. We're looking for one output pattern
from each spawn-id, and for the test to pass we need to see both
patterns.
Now, if gdb exits then this is a test failure (this would indicate gdb
crashing, which is bad), so we have an eof pattern associated with
the gdb spawn-id.
However, in this particular test we expect gdbserver to fail to
execute the binary (the test binary is set non-executable), and so we
get an error message from gdbserver (which matches the pattern), and
then gdbserver exits, this is expected.
The problem is that after spotting the pattern from gdbserver, we
often see the eof from gdbserver before we see the pattern from gdb.
If this happens then we drop out of the gdb_expect without ever seeing
the pattern from gdb, and fail the test.
In this commit, I place the spawn-id of gdbserver into a global
variable, and then use this global variable as the -i option within
the gdb_expect.
Now, once we have seen the expected pattern on the gdbserver spawn-id,
the global variable is cleared. After this the gdb_expect no longer
checks the gdbserver spawn-id for additional output, and so never sees
the eof event. This leaves the gdb_expect running, which allows the
pattern from gdb to be seen, and for the test to pass.
I now see no failures relating to 'run failure detected'.
Starting with gcc 12.1 / gcc 11.3, for test-case gdb.cp/align.exp we run into:
...
align.cc:29:23: error: invalid application of 'alignof' to a void type^M
29 | unsigned a_void = alignof (void);^M
| ^~~~~~~~~~~~~~^M
...
Fix this by using __alignof__ instead.
Tested on x86_64-linux, with gcc 7.5.0, gcc 12.1 and clang 12.0.1.
Currently debuginfod is disabled when the string of server URLs
is unset or set to be the empty string (via the $DEBUGINFOD_URLS
environment variable or the 'set debuginfod urls' gdb command).
Extend this functionality so that a whitespace-only URL also disables
debuginfod.
Modify a testcase to verify that a whitespace-only URL disables
debuginfod.
The type_wanted value, passed down to the create_sals_from_location
callback, is never used. Remove it.
Change-Id: Ic363ee13f6af593a3e875ff7fe46de130cdc190c
Update the gnulib import to fixes these issues:
- GDB build with clang + glibc < 2.33.
https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=d6a07b4dc21b3118727743142c678858df442853https://lists.gnu.org/archive/html/bug-gnulib/2022-04/msg00072.html
With glibc < 2.33, gnulib (since relatively recently) enables a
replacement for free (see gnulib/import/m4/free.m4). In that path,
clang shows this error:
make[2]: Entering directory '/home/smarchi/build/binutils-gdb-clang/gdbsupport'
CXX agent.o
In file included from /home/smarchi/src/binutils-gdb/gdbsupport/agent.cc:20:
In file included from /home/smarchi/src/binutils-gdb/gdbsupport/common-defs.h:95:
../gnulib/import/string.h:636:19: error: exception specification in declaration does not match previous declaration
_GL_EXTERN_C void free (void *) throw ();
^
../gnulib/import/stdlib.h:737:17: note: expanded from macro 'free'
# define free rpl_free
^
../gnulib/import/stdlib.h:739:1: note: previous declaration is here
_GL_FUNCDECL_RPL (free, void, (void *ptr));
^
../gnulib/import/sys/select.h:251:23: note: expanded from macro '_GL_FUNCDECL_RPL'
_GL_FUNCDECL_RPL_1 (rpl_##func, rettype, parameters_and_attributes)
^
<scratch space>:139:1: note: expanded from here
rpl_free
^
The gnulib commit mentioned fixes the exception specification of `free`.
- GDB build on RHEL 7:
CC libgnu_a-openat-proc.o
In file included from /usr/include/string.h:633,
from ./string.h:41,
from ../../../binutils-gdb/gnulib/import/openat-proc.c:30:
./string.h:1105:1: error: expected identifier or '(' before '__extension__'
1105 | _GL_FUNCDECL_SYS (strndup, char *,
| ^~~~~~~~~~~~~~~~
https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=84863a1c4dc8cca8fb0f6f670f67779cdd2d543bhttps://lists.gnu.org/archive/html/bug-gnulib/2022-04/msg00075.html
Change-Id: Ibd51302feece6f385d0c53e0d08921b5d95e2776
The breakpoint C++-ification series introduced a regression for Ada
catchpoints. Specifically, commit 2b5ab5b8 ("Convert base breakpoints
to vtable ops") caused these to start failing. I didn't notice this
because testing Ada using a Linux distro compiler requires installing
the GNAT debuginfo, which I hadn't done.
This patch fixes the problem. I'm checking it in.
When running test-case gdb.multi/attach-no-multi-process.exp with
check-readmore, I get:
...
(gdb) attach 13411^M
Attaching to Remote target^M
No unwaited-for children left.^M
(gdb) Reading symbols from attach-no-multi-process...^M
Reading symbols from /lib64/libm.so.6...^M
(No debugging symbols found in /lib64/libm.so.6)^M
Reading symbols from /lib64/libc.so.6...^M
(No debugging symbols found in /lib64/libc.so.6)^M
Reading symbols from /lib64/ld-linux-x86-64.so.2...^M
(No debugging symbols found in /lib64/ld-linux-x86-64.so.2)^M
0x00007f5df1fffc8a in clock_nanosleep@GLIBC_2.2.5 () from /lib64/libc.so.6^M
FAIL: gdb.multi/attach-no-multi-process.exp: target_non_stop=off: \
attach to the program via remote (timeout)
...
The problem is that the attach output is matched using gdb_test, which uses
the '$gdb_prompt $' regexp, and this does not handle the case that '(gdb) ' is
not the last available output.
Fix this by using a gdb_test_multiple instead with a '$gdb_prompt ' regexp, so
without the '$' anchor.
Tested on x86_64-linux with native, check-read1 and check-readmore.
Currently, the get_disassembler() implementations for riscv, csky, and
rl78--and mep_print_insn() for mep--access ELF variants of union fields
without first checking that the bfd actually represents an ELF. This
causes undefined behavior and crashes when disassembling non-ELF files
(the "binary" BFD, for example). Fix that.
At this point, all implementations of breakpoints use the vtable. So,
we can now remove most function pointers from breakpoint_ops and
switch to using methods directly in the callers. Only the two "static
virtual" methods remain in breakpoint_ops.
Right now, probe tracepoints are handled by a separate ops object.
However, they differ only in a small way from ordinary tracepoints,
and furthermore can be distinguished by their event location.
This patch merges the two cases, just as was done for breakpoints.
Right now, probe breakpoints are handled by a separate ops object.
However, they differ only in a small way from ordinary breakpoints,
and furthermore can be distinguished by their "probe" object.
This patch merges the two cases. This avoids having to introduce a
new bp_ constant (which can be quite subtle to do correctly) and a new
subclass.
Because the actual construction of a breakpoint is buried deep in
create_breakpoint, at present it's necessary to have a new bp_
enumerator constant any time a new subclass is needed. Static marker
tracepoints are one such case, so this patch introduces
bp_static_marker_tracepoint and updates various spots to recognize it.
This converts ranged breakpoints to use vtable_breakpoint_ops. This
requires introducing a new ranged_breakpoint type, but this is
relatively simple because ranged breakpoints can only be created by
break_range_command.
This converts "ordinary" breakpoint to use vtable_breakpoint_ops.
Recall that an ordinary breakpoint is both the kind normally created
by users, and also a base class used by other classes.
The dprintf breakpoint ops is mostly a copy of bpkt_breakpoint_ops,
except it's written out explicitly -- and, importantly, there's
nothing that bpkt_breakpoint_ops overrides that dprintf does not.
This changes dprintf to simply inherit directly, and updates struct
dprintf_breakpoint to reflect the change as well.
This adds a few new subclasses of breakpoint. The inheritance
hierarchy is chosen to reflect what's already present in
initialize_breakpoint_ops -- it mirrors the way that the _ops
structures are filled in.
This patch also changes new_breakpoint_from_type to create the correct
sublcass based on bptype. This is important due to the somewhat
inverted way in which create_breakpoint works; and in particular later
patches will change some of these entries.
This converts watchpoints and masked watchpoints. to use
vtable_breakpoint_ops. For masked watchpoints, a new subclass must be
introduced, and watch_command_1 is changed to create one.
This adds methods to struct breakpoint. Each method has a similar
signature to a corresponding function in breakpoint_ops, with the
exceptions of create_sals_from_location and create_breakpoints_sal,
which can't be virtual methods on breakpoint -- they are only used
during the construction of breakpoints.
Then, this adds a new vtable_breakpoint_ops structure and populates it
with functions that simply forward a call from breakpoint_ops to the
corresponding virtual method. These are all done with lambdas,
because they are just a stepping stone -- by the end of the series,
this structure will be deleted.
This changes breakpoint_ops::print_one to return bool, and updates all
the implementations and the caller. The caller is changed so that a
NULL check is no longer needed -- something that will be impossible
with a real method.
This adds an assertion to clone_momentary_breakpoint. This will
eventually be removed, but in the meantime is is useful for helping
convince oneself that momentary breakpoints will always use
momentary_breakpoint_ops. This understanding will help when cleaning
up the code later.