Oops, this caused two link errors in a testdir of all modules:
gcc -ftrapv -Wno-error -g -O2 -L/media/develdata/devel/inst-x86_64-64/lib
-rdynamic -o test-execute-child test-execute-child.o -lm -lm -lm -lm -lm -lm
-lm -lm -lm -lm -lm
/usr/bin/ld: test-execute-child.o: in function `main':
Hi Bruno,
I setup an Alpine VM and looked into the getusershell () failure you
reported earlier.
On alpine the default /etc/shells:
# valid login shells
/bin/sh
/bin/ash
/bin/bash
Fails because of the comment as I explained earlier. When adding a few
blank lines before the comment and an
In a testdir created through
./gnulib-tool --create-testdir --dir=../testdir1 --single-configure
--with-c++-tests stdbit
on MSVC 14 I see a compilation error:
/home/bruno/msvc/compile cl -nologo -DHAVE_CONFIG_H -I. -I../../gltests -I..
-DGNULIB_STRICT_CHECKING=1 -DIN_GNULIB_TESTS=1 -I.
On 5/17/24 3:49 PM, Paul Eggert wrote:
> POSIX does not require uint64_t, and the C standard
> does not require uint16_t or uint32_t either, so port
> to platforms that lack these types. The POSIX limitation
> is the only significant one in practice. I ran into this
> issue when updating Emacs,
In a testdir on MSVC 14, I see these compilation errors:
/home/bruno/msvc/compile cl -nologo -DHAVE_CONFIG_H -I. -I../../gltests -I..
-DGNULIB_STRICT_CHECKING=1 -DIN_GNULIB_TESTS=1 -I. -I../../gltests -I..
-I../../gltests/.. -I../gllib -I../../gltests/../gllib
-D_WIN32_WINNT=_WIN32_WINNT_WIN7
On 5/17/24 4:51 PM, Bruno Haible wrote:
> I think this produces wrong code with gcc 4.4.7 and older. See:
>
> -- foo.c --
> unsigned long long x = 0xff00;
> ---
>
> $ gcc -Wall -S
Paul Eggert wrote:
> -#else
> - return x) & 0xff00ull) >> 56)
> - | (((x) & 0x00ffull) >> 40)
> - | (((x) & 0xff00ull) >> 24)
> - | (((x) & 0x00ffull) >> 8)
> - | (((x) & 0xff00ull) << 8)
> -
Collin Funk wrote:
> Perhaps it is worth mentioning somewhere that XFAIL
> may refer to an actual bug that needs investigation?
Well, you know it now: All XFAIL are potential bugs, and need investigation
if the platform is considered important. :)
Bruno
POSIX does not require uint64_t, and the C standard
does not require uint16_t or uint32_t either, so port
to platforms that lack these types. The POSIX limitation
is the only significant one in practice. I ran into this
issue when updating Emacs, which still ports to platforms
lacking 64-bit
On 5/17/24 2:10 PM, Bruno Haible wrote:
> Since I don't have time to investigate them and, OTOH, I don't want that
> the NetBSD test failures mark the entire continuous integration run as
> "failed", here are patches to mark them as expected failures.
Seems reasonable. Perhaps it is worth
On MSVC 14 in 64-bit mode, in a testdir of the modules
stdbit count-leading-zeros count-one-bits count-trailing-zeros ffs ffsl ffsll
I see these 7 test failures:
FAIL: test-stdc_bit_ceil.exe
FAIL: test-stdc_bit_floor.exe
FAIL: test-stdc_bit_width.exe
PASS: test-stdc_count_ones.exe
PASS:
On MSVC, I see this link error:
/home/bruno/msvc/compile cl -nologo -MD -L/usr/local/msvc32/lib -o
test-getusershell.exe test-getusershell.obj libtests.a ../gllib/libgnu.a
libtests.a ../gllib/libgnu.a libtests.a
cl : Command line warning D9024 : unrecognized source file type 'libtests.a',
On NetBSD 9.0 and 10.0, I always see the same tests fail:
FAIL: test-dup3
===
../../gltests/test-dup3.c:113: assertion 'dup3 (fd, fd, o_flags) == -1' failed
FAIL test-dup3 (exit status: 134)
FAIL: test-fdutimensat
==
../../gltests/test-utimens.h:75: assertion
When a unit test fails, we call abort(), because
- this allows finding the failure spot a posteriori with a core dump,
- it works fine in the debugger, even if the debugger cannot put a
breakpoint on, say, the function exit().
However, when running a unit test in a continuous integration
Paul Eggert wrote:
> In my weaker moments I sometimes think we should declare every test-case
> variable volatile, along the lines of the "volatile-by-default" approach
> for Java. See:
>
> Liu L, Millstein T, Musuvathi M. Safe-by-default concurrency for modern
> programming languages. ACM
On 2024-05-16 23:25, Collin Funk wrote:
Hi Paul,
On 5/16/24 10:16 PM, Paul Eggert wrote:
+#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))\
+ || (defined __has_builtin && __has_builtin (__builtin_bswap32))
Unfortunately this usage is not portable, not for
On 2024-05-17 02:48, Bruno Haible wrote:
-flto was buggy for a long time. Is whole-program optimization working
fine for you? If so, which compiler version, binutils version, and
compiler options can you recommend?
No, I merely found that problem by program inspection. Like you, I've
had
musl libc does not have the backtrace* functions. It's worth mentioning.
(I don't know why https://github.com/ronchaine/libexecinfo has been archived.)
Ideally we should document musl libc portability, like we do for all other
platforms. This would be about 2-3 days of work; not in my plans for
On 5/17/24 6:45 AM, Bruno Haible wrote:
>> It also looks like musl will return empty lines (getline () with
>> length 1), which I believe is incorrect from the FreeBSD man page
>> description.
>
> So, once you have a musl libc VM yourself, you are ready for reporting
> your first musl libc bug.
Hi Collin,
> I remember you saying a few days ago saying that "every function is
> worth testing", so here is more evidence. :)
:-)
> It appears that musl doesn't check for comments which explains that
> failure [2].
>
> It also looks like musl will return empty lines (getline () with
> length
Hi Bruno,
On 5/17/24 5:59 AM, Bruno Haible wrote:
> The test fails on Alpine Linux.
>
>
> FAIL: test-getusershell
> ===
>
> ../../gltests/test-getusershell.c:53: assertion 'ptr[0] != '#'' failed
> Aborted (core dumped)
> FAIL test-getusershell (exit status: 134)
Hahaha, I
A test module should depend on 'test-framework-sh' if and only if the
test uses tests/init.sh.
2024-05-17 Bruno Haible
tests: Fix dependencies to test-framework-sh.
* modules/argp-tests (Depends-on): Add test-framework-sh.
* modules/argp-version-etc-tests
Hi Collin,
> + getusershell: Add tests.
> + * modules/getusershell-tests: New file.
> + * tests/test-getusershell.c: New file.
The test fails on Alpine Linux.
FAIL: test-getusershell
===
../../gltests/test-getusershell.c:53: assertion 'ptr[0] != '#'' failed
On MSVC, I see this compilation error:
/home/bruno/msvc/compile cl -nologo -DHAVE_CONFIG_H -I. -I../../gllib -I..
-DGNULIB_STRICT_CHECKING=1 -D_WIN32_WINNT=_WIN32_WINNT_WIN7
-I/usr/local/msvc64/include -MD -c -o stdbit.obj `cygpath -w
'../../gllib/stdbit.c'`
stdbit.c
.\stdbit.h(252): error
On 5/17/24 2:38 AM, Bruno Haible wrote:
> Now we can use the usual doc structure.
>
>
> 2024-05-17 Bruno Haible
>
> putenv-gnu: Update documentation.
> * doc/posix-functions/putenv.texi: Refer also to the glibc
> documentation. Use the usual doc structure.
Looks good.
On macOS 13 and 14 I'm seeing this compilation error:
../gllib/sys/select.h:822:19: error: declaration of 'select' has a different
language linkage
_GL_CXXALIASWARN (select);
^
This patch should fix it.
2024-05-17 Bruno Haible
sys_select: Fix compilation error in
Hi Paul,
> Now volatile, to foil whole-program optimization.
-flto was buggy for a long time. Is whole-program optimization working
fine for you? If so, which compiler version, binutils version, and
compiler options can you recommend?
Bruno
Collin Funk asked:
> I'm assuming that _GL_[module-name]_(rest-of-macro) can be assumed
> safe?
Yes, such a macro name is safe.
Bruno
Collin Funk wrote:
> Ah, interesting. I think in English it would mostly depend on the
> context. "Interruption" means you are stopping something from
> progressing, but I don't think the word itself describes if it is
> temporary or not. "Stop" more strongly indicates a permanent halting
> of the
Collin Funk wrote:
> Done in the two attached patches.
Now we can use the usual doc structure.
2024-05-17 Bruno Haible
putenv-gnu: Update documentation.
* doc/posix-functions/putenv.texi: Refer also to the glibc
documentation. Use the usual doc structure.
diff --git
In the NEWS file:
2009-04-01 visibility Renamed to lib-symbol-visibility.
The commit that did this missed a reference to 'visibility' in the
documentation. This patch fixes it.
CollinFrom fe8453457c44194354bea89347f713562f67aae2 Mon Sep 17 00:00:00 2001
From: Collin Funk
Date: Fri,
Hi Paul,
On 5/16/24 10:16 PM, Paul Eggert wrote:
>> +#if (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))\
>> + || (defined __has_builtin && __has_builtin (__builtin_bswap32))
>
> Unfortunately this usage is not portable, not for __has_builtin or for any of
> the other
32 matches
Mail list logo
