Hi Bruno,
On 23.05.20 22:47, Bruno Haible wrote:
> Tim Rühsen wrote:
>> - clang-10 with unset CFLAGS (*.clang-10)
>>
>> And the same set with -fsanitize... (*.sanitize).
>
> The other differences between the configure results without and with
> sanitizers
> can be explained by the fact that
Tim Rühsen wrote:
> - clang-10 with unset CFLAGS (*.clang-10)
>
> And the same set with -fsanitize... (*.sanitize).
The other differences between the configure results without and with sanitizers
can be explained by the fact that these -fsanitize options have the effect that
the resulting
Hi Tim,
> Well, all made on the _same_ system (my desktop) with just ~35 minutes
> in between.
>
> Since system updates are made manually, it is _very_ unlikely that I
> made an update between the tests.
:-)
> But yes, I made one test without $CC set (so likely with gcc) and the
> other one
Hi Bruno,
On 21.05.20 21:31, Bruno Haible wrote:
> Hi Tim,
>
> These comparisons are not so useful, because they not only come from different
> compilers (gcc vs. clang) but also from systems with different libcs: there
> are differences regarding libm, calloc, thrd_create, pthread_sigmask, and
Hi Tim,
These comparisons are not so useful, because they not only come from different
compilers (gcc vs. clang) but also from systems with different libcs: there
are differences regarding libm, calloc, thrd_create, pthread_sigmask, and other
functions. It would be more useful to compare, on the
Paul Eggert wrote:
> > - if (memmem (haystack, 3, NULL, 0) != haystack)
> > + if (memmem (haystack, 3, (const char *) 1, 0) != haystack)
>
> This has undefined behavior in general, no?
No. memmem is not supposed to access more than NEEDLELEN bytes at NEEDLE.
> How about using '""'
On 5/21/20 11:40 AM, Bruno Haible wrote:
- if (memmem (haystack, 3, NULL, 0) != haystack)
+ if (memmem (haystack, 3, (const char *) 1, 0) != haystack)
This has undefined behavior in general, no? So some future sanitizer may
complain about it, for good reason.
How about using '""'
Tim Rühsen wrote:
> configure:36833: checking whether memmem works
> SUMMARY: UndefinedBehaviorSanitizer: invalid-null-argument
This patch should fix it.
2020-05-21 Bruno Haible
memmem: Avoid wrong configure results with "clang -fsanitize=undefined".
Reported by Tim Rühsen
> The attached patch fixes it. (Please feel free to amend.)
Thanks, I've pushed it in your name.
Bruno
Hi Tim,
> With my getdelim patch applied, this is the summary of sanitizer findings:
>
> $ egrep 'SUMM|checking' config.log|grep -B1 SUMM
> configure:14038: checking for working C stack overflow detection
> SUMMARY: UndefinedBehaviorSanitizer: implicit-integer-sign-change
> conftest.c:379:30 in
Hi Tim,
> > You may try to pass '--avoid=gettext' to overcome this.
>
> ./configure --config-cache ends (stops ?) with
> The BISON_I18N macro is used without being preceded by AM_GNU_GETTEXT.
Oh well. Then you apparently need '--avoid=gettext --avoid=bison-i18n'.
> And I see this output here
Hi Bruno,
On 18.05.20 21:44, Bruno Haible wrote:
> Hi Tim,
>
>>> The way to determine the answer is:
>>> 1. Create a test dir of all gnulib modules.
>>> 2. Configure it with --config-cache.
>>> 3. Configure it with --config-cache and your sanitizer options.
>>> 4. Compare the generated
Hi Bruno,
On 20.05.20 00:46, Bruno Haible wrote:
> Hi Tim,
>
>>> rm -rf ../testdir-all; ./gnulib-tool --create-testdir
>>> --dir=../testdir-all --single-configure
>>
>> This results in:
>>
>> executing autopoint --force
>> autopoint: *** The AM_GNU_GETTEXT_VERSION declaration in your
Hi Tim,
> > rm -rf ../testdir-all; ./gnulib-tool --create-testdir
> > --dir=../testdir-all --single-configure
>
> This results in:
>
> executing autopoint --force
> autopoint: *** The AM_GNU_GETTEXT_VERSION declaration in your configure.ac
>file requires the infrastructure
On 18.05.20 21:44, Bruno Haible wrote:
> Hi Tim,
>
>>> The way to determine the answer is:
>>> 1. Create a test dir of all gnulib modules.
>>> 2. Configure it with --config-cache.
>>> 3. Configure it with --config-cache and your sanitizer options.
>>> 4. Compare the generated config.cache and
Hi Tim,
> > The way to determine the answer is:
> > 1. Create a test dir of all gnulib modules.
> > 2. Configure it with --config-cache.
> > 3. Configure it with --config-cache and your sanitizer options.
> > 4. Compare the generated config.cache and config.status files.
>
> I am short on time
Hi Bruno,
On 18.05.20 13:50, Bruno Haible wrote:
> Hi Tim,
>
>> With leak sanitizer on, the test for getdelim fails due to a memory leak.
>
> Is this the only configure test that fails this way?
The only one that fails when running GNU Poke ./configure.
> I mean, it's 3 years since we last
Hi Tim,
> With leak sanitizer on, the test for getdelim fails due to a memory leak.
Is this the only configure test that fails this way?
I mean, it's 3 years since we last looked for such configure test failures. [1]
The way to determine the answer is:
1. Create a test dir of all gnulib
18 matches
Mail list logo
