When I run the testsuite, I have:
Running .../gdb/testsuite/gdb.base/shlib-call.exp ...
DUPLICATE: gdb.base/shlib-call.exp: print g
DUPLICATE: gdb.base/shlib-call.exp: set print sevenbit-strings
DUPLICATE: gdb.base/shlib-call.exp: set print address off
DUPLICATE: gdb.base/shlib-call.exp: set width 0
DUPLICATE: gdb.base/shlib-call.exp: continue until exit
Fix by adjusting the test names when required, and by removing
un-necessary commands.
While at it, do some cleanup:
- Replace an explicit GDB restart sequence with a call to clean_restart.
- Remove trailing whitespaces.
- Use $gdb_test_name in gdb_test_multiple.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/set-cwd.exp ...
DUPLICATE: gdb.base/set-cwd.exp: test_cwd_reset: continue to breakpoint: break-here
Fix by moving the tests after the 'runto_main' within the same
with_test_prefix scope.
While at it, I fix some indentation issues.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/exprs.exp ...
DUPLICATE: gdb.base/exprs.exp: \$[0-9]* = red (setup)
Fix by using with_test_prefix where appropriate.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/readline.exp ...
DUPLICATE: gdb.base/readline.exp: Simple operate-and-get-next - final prompt
Fix by adjusting the prefix given to the second 'simple' call to
operate_and_get_next.
Tested on x86_64-linux.
When I run the testsuite, I have:
Running .../gdb/testsuite/gdb.base/pretty-array.exp ...
DUPLICATE: gdb.base/pretty-array.exp: print nums
DUPLICATE: gdb.base/pretty-array.exp: print nums
Fix by giving a name to the test cases.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/ui-redirect.exp ...
DUPLICATE: gdb.base/ui-redirect.exp: redirect while already logging: set logging redirect off
Fix by moving the first 'set logging redirect off' to the end of the
previous [with_test_prefix] test block. The statement's purpose is to
clean the on flag set in this previous block, so moving it there makes
sense and does not change the sequence of commands in the test file.
Tested on x86_64-linux.
There is a expect support library in the source tree designed to help
developers test the auto-completion capabilities of GDB.
One of the functions is test_gdb_complete_unique_re. It is used
(usually indirectly via test_gdb_complete_unique) to test that a given
input line is completed as a given output line. The test checks for two
ways to do the completion: using tab-completion, or using the
'complete' command. To do this, calls to two dedicated functions are
performed. If we omit few details, we can consider that a call to
test_gdb_complete_unique $input $expected
is equivalent to the two following calls:
test_gdb_complete_tab_unique $input $expected
test_gdb_complete_cmd_unique $input $expected
When using the tab-completion, everything works as expected, but some
care must be taken when using the 'complete' command if the given input
has leading whitespaces. In such situation, the output of the
'complete' command will drop the leading whitespaces.
The current approach is that in such situation, the input and expected
outputs are right trimmed (i.e. all leading whitespaces are removed)
when performing the command completion check.
This means that the following call:
test_gdb_complete_unique " $input" " $expected"
is almost equivalent to (again, omitting few details and arguments):
test_gdb_complete_tab_unique " $input" " $expected"
test_gdb_complete_cmd_unique "$input" "$expected"
This approach comes with a problem that we encounter when running the
tests in complete-empty.exp. When doing so, we have:
Running .../gdb/testsuite/gdb.base/complete-empty.exp ...
DUPLICATE: gdb.base/complete-empty.exp: empty-input-line: cmd complete ""
This is because the test file does something like:
test_gdb_complete_unique "" "!" " " 1
test_gdb_complete_unique " " " !" " " 1¬
which, if we do the substitution introduced above is equivalent to:
test_gdb_complete_tab_unique "" "!"
test_gdb_complete_cmd_unique "" "!"
test_gdb_complete_tab_unique " " " !"
test_gdb_complete_cmd_unique "" "!"
We see that the lines 2 and 4 are now the same, and for this reason the
testing framework complains about DUPLICATE test names.
To fix that, this commit proposes that instead of left trimming both
input and expected outputs, only the expected output is trimmed.
Care must be taken in the case the completion gives more possibilities
than allowed by the max-completions setting. In this case, the input
will be repeated in the output in its left trimmed version. This commit
also ensures that this is taken care of.
With this commit, the gdb.base/complete-empty.exp still passes all its
tests but does not report the DUPLICATE anymore.
Tested on x86_64-linux.
When I run the testsuite, I have:
Running .../gdb/testsuite/gdb.base/subst.ex ...
DUPLICATE: gdb.base/subst.exp: unset substitute-path from, no rule entered yet
Fix by adjusting the problematic test name.
Tested on x86_64-linux.
When I run the testsuite, I have:
Running ../gdb/testsuite/gdb.base/dfp-exprs.exp ...
DUPLICATE: gdb.base/dfp-exprs.exp: p 1.2dl < 1.3df
Replace hand-written tests checking various comparison operators between
various decimal floating point types with a loop to programmatically
generate all the combinations. This removes the need to eyeball for all
suffixes, which lead to the original duplication.
Also add a lot more combinations, testing all comparison operators
comprehensively. The result is 262 unique tests vs 104 before this
patch.
Tested on x86_86-linux.
Change-Id: Id215a3d610aa8e032bf06ee160b5e3aed4a92d1e
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/ptype.exp ...
DUPLICATE: gdb.base/ptype.exp: ptype the_highest
DUPLICATE: gdb.base/ptype.exp: list intfoo
DUPLICATE: gdb.base/ptype.exp: list charfoo
Fix by adjusting the offending test names.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/dfp-test.exp ...
DUPLICATE: gdb.base/dfp-test.exp: 1.23E is an invalid number
DUPLICATE: gdb.base/dfp-test.exp: 1.23E45A is an invalid number
DUPLICATE: gdb.base/dfp-test.exp: 1.23E is an invalid number
DUPLICATE: gdb.base/dfp-test.exp: 1.23E45A is an invalid number
Fix by using proc_with_prefix where appropriate.
Tested on x86_64-linux.
Co-Authored-By: Andrew Burgess <aburgess@redhat.com>
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/del.exp ...
DUPLICATE: gdb.base/del.exp: info break after removing break on main
Refactor slightly this test to run the various configurations under
foreach_with_prefix so each variant is automatically prefixed, ensuring
that the forgotten custom test name cannot happen.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/solib-display.exp ...
DUPLICATE: gdb.base/solib-display.exp: NO: break 25
DUPLICATE: gdb.base/solib-display.exp: NO: continue
DUPLICATE: gdb.base/solib-display.exp: IN: break 25
DUPLICATE: gdb.base/solib-display.exp: IN: continue
DUPLICATE: gdb.base/solib-display.exp: SEP: break 25
DUPLICATE: gdb.base/solib-display.exp: SEP: continue
The 'break 25' appears because the test inserts two breakpoints at the
same location. Fix this by only inserting the breakpoint once.
Fix the 'continue' DUPLICATE by giving a phony name to the second
continue: 'continue two'.
While at it, this commit also removes a trailing space.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/decl-before-def.exp ...
DUPLICATE: gdb.base/decl-before-def.exp: p a
Fix by giving explicit names to the two tests that use the same command.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/pending.exp ...
DUPLICATE: gdb.base/pending.exp: disable other breakpoints
Fix by adjusting the test names.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/checkpoint.exp ...
DUPLICATE: gdb.base/checkpoint.exp: verify lines 5 two
DUPLICATE: gdb.base/checkpoint.exp: restart 0 one
This patch fixes the various erroneous incorrect test names.
While at it, this patch also remove some trailing white spaces across
the file.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/pie-fork.exp ...
DUPLICATE: gdb.base/pie-fork.exp: test_no_detach_on_fork: continue
Fix by giving explicit names to the 'continue' commands that cause the
duplicate message.
Tested on x86_64-linux.
When running the testsuite, I have:
Running .../gdb/testsuite/gdb.base/realname-expand.exp ...
DUPLICATE: gdb.base/realname-expand.exp: set basenames-may-differ on
This is due to the fact that the test restarts GDB twice and each time
sets the basenames-may-differ setting. This patch proposes to fix this
by not restarting GDB so the setting is maintained. It just clears the
breakpoints between the two tests and updates the breakpoints number as
required.
This patch also perform some minor refactorings to improve visibility.
Tested on x86_64-linux.
When running the testsuite I have:
Running .../gdb/testsuite/gdb.base/interp.exp ...
DUPLICATE: gdb.base/interp.exp: interpreter-exec mi "-var-update *"
This is due to the fact that multiple successive instances of
gdb_test_multiple use 'pass $cmd', but one of them forgets to reset $cmd
to a new test name.
Fix by using 'pass $gdb_test_name', given that the gdb_test_name is set
by gdb_test_multiple.
While fixing this, this patch refactors all occurrences of the following
pattern:
set cmd foo
gdb_test_multiple $cmd $cmd {
-re ... {
pass $cmd
}
}
into
gdb_test_multiple foo "" {
-re ... {
pass $gdb_test_name
}
}
This makes this test file coherent in its use of $gdb_test_name.
Tested on x86_64-linux.
When running the testsuite I see:
Running .../gdb/testsuite/gdb.base/miscexprs.exp ...
DUPLICATE: gdb.base/miscexprs.exp: print value of !ibig.i[100]
DUPLICATE: gdb.base/miscexprs.exp: print value of !ibig.i[100]
This is due to an explicit test name repeated across multiple tests.
The actual test explicit names do not add much over the command from
wich default test names are derived.
Fix by removing the explicit test names across the file where they do
not add value. While at doing some cleaning, also use $gdb_test_name in
the various uses of gdb_test_multiple.
Tested on x86_64-linux.
When running the testsuite I have:
Running .../gdb/testsuite/gdb.base/stack-checking.exp ...
DUPLICATE: gdb.base/stack-checking.exp: bt
DUPLICATE: gdb.base/stack-checking.exp: bt
Fix by using with_test_prefix.
Tested on x86_64-linux.
Building on the previous commit, this makes use of a trailing @ to
split long @deffn lines in the guile.texi source file. This splitting
doesn't change how the document is laid out by texinfo.
I have also wrapped keyword and argument name pairs in @w{...} to
prevent line breaks appearing between the two. I've currently only
done this for the longer @deffn lines, where a line break is
possible. This makes the @deffn lines much nicer to read in the
generated pdf.
Most guile procedures in the guile.texi file are defined like:
@deffn {Scheme Procedure} name arg1 arg2 arg3
But there are two places where we do this:
@deffn {Scheme Procedure} (name arg1 arg2 arg3)
Notice the added (...). Though this does represent how a procedure
call is written in scheme, it's not the normal style throughout the
manual. I also checked the 'info guile' info page to see how they
wrote there declarations, and they use the first style too.
The second style also has the drawback that index entries are added as
'(name', and so they are grouped in the '(' section of the index,
which is not very user friendly.
In this commit I've changed the definitions of make-command and
make-parameter to use the first style.
The procedure declaration lines can get pretty long with all of the
arguments, and this was true for both of the procedures I am changing
in this commit. I have made use of a trailing '@' to split the deffn
lines, and keep them under 80 characters in the texi source. This
makes no difference to how the final document looks.
Finally, our current style for keyword arguments, appears to be:
[#:keyword-name argument-name]
I don't really understand the reason for this, 'info guile' just seems
to use:
[#:keyword-name]
which seems just as good to me. But I don't propose to change
that just now. What I do notice though, is that sometimes, texinfo
will place a line break between the keyword-name and the
argument-name, for example, the pdf of make-command is:
make-command name [#:invoke invoke] [#:command-class
command-class] [#:completer-class completer] [#:prefix? prefix] [#:doc
doc-string]
Notice the line break after '#:command-class' and after '#:doc',
neither of which are ideal. And so, for the two commands I am
changing in this commit, I have made use of @w{...} to prevent line
breaks between the keyword-name and the argument-name. Now the pdf
looks like this:
make-command name [#:invoke invoke]
[#:command-class command-class] [#:completer-class completer]
[#:prefix? prefix] [#:doc doc-string]
Which seems much better. I'll probably update the other deffn lines
at some point.
target_announce_detach was added in commit 0f48b757 ("Factor out
"Detaching from program" message printing"). There, Pedro wrote:
(For now, I left the couple targets that print this a bit differently
alone. Maybe this could be further pulled out into infcmd.c. If we
did that, and those targets want to continue printing differently,
this new function could be converted to a target method.)
It seems to me that the differences aren't very big, and in some cases
other targets handled the output a bit more nicely. In particular,
some targets will print a different message when exec_file==NULL,
rather than printing the same output with an empty string as
exec_file.
This patch incorporates the nicer output into target_announce_detach,
then changes the remaining ports to use this function.
This introduces target_announce_attach, by analog with
target_announce_detach. Then it converts existing targets to use
this, rather than emitting their own output by hand.
In gnu-nat.c we currently implement some set/show prefix commands
"manually", that is, we call add_prefix_cmd, and assign a set and show
function to each prefix command.
These set/show functions print an error indicating that the user
didn't type a complete command.
If we instead switch to using add_setshow_prefix_cmd then we can
delete the set/show functions, GDB provides some default functions,
which give a nice help style summary that lists all of the available
sub-commands, along with a one line summary of what each does.
Though this clearly changes the existing behaviour, I think this
change is acceptable as the new behaviour is more inline with other
set/show prefix commands, and the new behaviour is more informative.
This change will conflict with Tom's change here:
https://sourceware.org/pipermail/gdb-patches/2022-January/184724.html
Where Tom changes the set/show functions that I delete. My suggestion
is that the set/show functions still be deleted even after Tom's
patch (or instead of Tom's patch).
For testing I've build GDB on GNU/Hurd, and manually tested these
functions. I did a grep over the testsuite, and don't believe the
existing error messages are being checked for in any tests.
windows-tdep.c checks the result of xmalloc, which isn't necessary. I
initially removed this dead check, but then went a bit further and
modified the code so that some "goto"s and explicit memory management
could be removed. Then, I added a couple of missing bounds checks.
I believe this also fixes a possible bug with a missing 0-termination
of a string. I am not certain, but that is why I think the existing
code allocates a buffer that is 1 byte too long -- but then it fails
to set this byte to 0.
language_info calls:
show_language_command (NULL, 1, NULL, NULL);
... "knowing" that show_language_command does not use its ui_file
parameter. However, this was changed in commit 7514a661
("Consistently Use ui_file parameter to show callbacks").
This patch changes language_info to pass a ui_file.
It took a while to write the test -- this function is only called when
'verbose' is on and when switching the "expected" language in auto
mode.
langs.exp currently has some fails for me because the stack trace
includes full paths to the source files.
FAIL: gdb.base/langs.exp: up to foo in langs.exp
FAIL: gdb.base/langs.exp: up to cppsub_ in langs.exp
FAIL: gdb.base/langs.exp: up to fsub in langs.exp
This fixes the failures by making the filename regexps a bit more lax.
I just realized that I forgot to update the year before pushing the
patch that created this file. Since it landed after the global
copyright year update have been done, this file’s copyright year is
updated.
This patch fixes that.
Change-Id: I280f7d86e02d38425f7afdcf19a1c3500d51c23f
Internally, AdaCore has a test that is equivalent to (really a direct
translation of) gdb.base/gnu_vector.exp. On 32-bit RISC-V, the
"return" part of this test fails.
Joel tracked this down to riscv_return_value returning
RETURN_VALUE_ABI_RETURNS_ADDRESS. Using
RETURN_VALUE_ABI_PRESERVES_ADDRESS is more correct here, and fixes the
bug.
I tested this for both 32- and 64-bit RISC-V using the AdaCore
internal test suite, and Andrew Burgess tested it using
gnu_vector.exp.
Most of the expression-dumping code uses filtered output, but a few
functions did not. This patch cleans up these instance.
Note that this won't cause any behavior change, because the only calls
to dump_prefix_expression pass in gdb_stdlog. However, in the long
run it's easier to audit the code if the number of uses of _unfiltered
is reduced.
This changes one terminal_info implementation, and
default_terminal_info, to use filtered output. Other implementations
of this method already use filtered output.
I can't compile go32-nat.c, so this is a 'best effort' patch.
gnu-nat.c has a number of ordinary commands that should use filtered
output. In a few cases, I changed the output to use gdb_stderr as
well. I can't compile this file, so this patch is split out as a
"best effort".
Various targets introduce their own commands, which then use
unfiltered output. It's better to use filtered output by default, so
this patch fixes the instances I found.
There are several commands that may optionally send their output to a
file -- they take an optional filename argument and open a file. This
patch changes these commands to use filtered output. The rationale
here is that, when printing to gdb_stdout, filtering is appropriate --
it is, and should be, the default for all commands. And, when writing
to a file, paging will not happen anyway (it only happens when the
stream==gdb_stdout), so using the _filtered form will not change
anything.
This changes the kill command to use filtered output. I split this
one into its own patch because, out of an abundance of caution, I
changed the function to call bfd_cache_close_all a bit earlier, in
case pagination caused an exception.
Many otherwise ordinary commands choose to use unfiltered output
rather than filtered. I don't think there's any reason for this, so
this changes many such commands to use filtered output instead.
Note that complete_command is not touched due to a comment there
explaining why unfiltered output is believed to be used.
Change language_info to use filtered output. This is ok because the
sole caller uses filtered output elsewhere, and because this function
calls show_language_command, which also uses filtered output.
This changes the implementations of the target files_info method to
use filtered output. This makes sense because the sole caller of this
method is an ordinary command (info_program_command). This patch
changes this command to use filtered output as well.
target-descriptions.c uses unfiltered output. However, if you happen
to invoke this command interactively, it's probably better for it to
use filtering. For non-interactive use, this doesn't matter.
This changes gdbarch dumping to use filtered output. This seems a bit
better to me, both on the principle that this is an ordinary command,
and because the output can be voluminous, so it may be nice to stop in
the middle.
In my tour of the ui_file subsystem, I found that fputstr and fputstrn
can be simplified. The _filtered forms are never used (and IMO
unlikely to ever be used) and so can be removed. And, the interface
can be simplified by removing a callback function and moving the
implementation directly to ui_file.
A new self-test is included. Previously, I think nothing was testing
this code.
Regression tested on x86-64 Fedora 34.
A change to BFD caused a gdb regression when using the Ada "catch
exception" feature. The bug is visible when a shared library throws
an exception that is caught in the main executable.
This was discussed here:
https://sourceware.org/pipermail/binutils/2021-July/117538.html
This patch implements Alan's proposed fix, namely to use VERSYM_HIDDEN
rather than the name when deciding to install a version-less symbol.
The internal test case is identical to the catch_ex_std.exp that is
in-tree, so I haven't added a new test. I could not make that one
fail on x86-64 Linux, though. It's possible that maybe I'd have to
update the system linker first, but I didn't want to try that.
Regression tested on x86-64 Fedora 32.
Commit 72ee03ff58 fixed a use-after-move bug in add_thread_object, but
it changed the inferior_thread attribute to contain the inferior instead
of the actual thread.
This now uses the thread_obj in its new location instead.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28429
execute_control_commands_to_string can be rewritten in terms of
execute_fn_to_string, which consolidates some knowledge about which
streams to redirect.
Regression tested on x86-64 Fedora 34.
Right now, gdb's self-backtrace feature will still print something
when a backtrace is unavailable:
sig_write (_("----- Backtrace -----\n"));
[...]
sig_write (_("Backtrace unavailable\n"));
sig_write ("---------------------\n");
However, if GDB_PRINT_INTERNAL_BACKTRACE is undefined, it seems better
to me to print nothing at all.
This patch implements this change. It also makes a couple of other
small changes in this same module: it adds a header guard to
bt-utils.h, and it protects the definitions of
gdb_internal_backtrace_1 with a check of GDB_PRINT_INTERNAL_BACKTRACE.
The patch to fix paging with redirection caused a regression in the
internal AdaCore test suite. The problem occurs when running an MI
command from the CLI using interpreter-exec, when paging is enabled.
This scenario isn't covered by the current test suite, so this patch
includes a new test.
The problem is that, in this situation, MI does:
fputs_unfiltered (strcmp (context->command, "target-select") == 0
? "^connected" : "^done", mi->raw_stdout);
Here raw_stdout is a stdio_file wrapping stdout, so the pager thinks
that it is ok to buffer the output. However, in this setup, it isn't
ok, and flushing the wrap buffer doesn't really work properly. Also,
MI next does:
mi_out_put (uiout, mi->raw_stdout);
... but this uses ui_file::write, which also doesn't flush the wrap
buffer.
I think all this will be fixed by the pager rewrite series I'm working
on. However, in the meantime, adding the old gdb_stdout check back to
the pager fixes this problem.
Regression tested on x86-64 Fedora 34.
