Testing build failure with latest GCC

Testing build failure with latest GCC

[Steve Ellcey]

I am building the top-of-tree glibc with top-of-tree gcc and getting
these messages when running 'make check'.  Should we explicitly ignore
the return value of fgets in these tests or check it and set 'failed'
if it is not the expected value?  Or do we think GCC is at fault?  It
seems like a legimate warning and we do compile with -Wall -Werror.

Steve Ellcey
sellcey@cavium.com

test-assert.c: In function ‘main’:
test-assert.c:75:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
   ^
test-assert.c:79:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
   ^
test-assert.c:83:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
test-assert-perr.c: In function ‘main’:
test-assert-perr.c:73:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
   ^
test-assert-perr.c:77:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
   ^
test-assert-perr.c:81:3: error: ignoring return value of ‘fgets’, declared with attribute warn_unused_result [-Werror=unused-result]
   fgets (buf, 160, stderr);
   ^
cc1: all warnings being treated as errors

Re: Testing build failure with latest GCC

[Florian Weimer]

On Thu, May 25, 2017 at 6:50 PM, Steve Ellcey <sellcey@cavium.com> wrote:
> I am building the top-of-tree glibc with top-of-tree gcc and getting
> these messages when running 'make check'.  Should we explicitly ignore
> the return value of fgets in these tests or check it and set 'failed'
> if it is not the expected value?  Or do we think GCC is at fault?  It
> seems like a legimate warning and we do compile with -Wall -Werror.

Any idea why it didn't warn before?

The f* stream functions are special. You can check for errors using
the return value, or using ferror. Therefore, the attribute might not
be correct.

Considering how difficult is to bypass warn-used attribute (a cast to
void is not enough), I wonder if we should drop the attribute.

Thanks,
Florian

Re: Testing build failure with latest GCC

[Steve Ellcey]

On Thu, 2017-05-25 at 18:56 +0200, Florian Weimer wrote:
> On Thu, May 25, 2017 at 6:50 PM, Steve Ellcey <sellcey@cavium.com>
> wrote:
> > 
> > I am building the top-of-tree glibc with top-of-tree gcc and getting
> > these messages when running 'make check'.  Should we explicitly ignore
> > the return value of fgets in these tests or check it and set 'failed'
> > if it is not the expected value?  Or do we think GCC is at fault?  It
> > seems like a legimate warning and we do compile with -Wall -Werror.
> Any idea why it didn't warn before?

No I don't know why it didn't warn before.  I assume it is a recent GCC
change but I don't know what change.  I'll see if I can track down a
specific change that caused this to start.

> The f* stream functions are special. You can check for errors using
> the return value, or using ferror. Therefore, the attribute might not
> be correct.
> 
> Considering how difficult is to bypass warn-used attribute (a cast to
> void is not enough), I wonder if we should drop the attribute.

That is a good point, I thought a '(void)' cast could fix this but I
see that it does not.

Steve Ellcey

Re: Testing build failure with latest GCC

[Steve Ellcey]

On Thu, 2017-05-25 at 18:56 +0200, Florian Weimer wrote:
> 
> The f* stream functions are special. You can check for errors using
> the return value, or using ferror. Therefore, the attribute might not
> be correct.
> 
> Considering how difficult is to bypass warn-used attribute (a cast to
> void is not enough), I wonder if we should drop the attribute.
> 
> Thanks,
> Florian

I think this is a bug in my scripts.  I thought I was compiling with
the top-of-tree GCC but it looks like I was actually compiling with GCC
5.4 when building the tests (bad path).  The actual library however was
compiled with ToT GCC and I think that mixing of building the library
with ToT GCC and the tests with GCC 5.4 is causing the problem.

I think we can just call this a user error.

Steve Ellcey
sellcey@cavium.com

Re: Testing build failure with latest GCC

[Cyril Hrubis]

Hi!
> > The f* stream functions are special. You can check for errors using
> > the return value, or using ferror. Therefore, the attribute might not
> > be correct.
> > 
> > Considering how difficult is to bypass warn-used attribute (a cast to
> > void is not enough), I wonder if we should drop the attribute.
> 
> That is a good point, I thought a '(void)' cast could fix this but I
> see that it does not.

FYI that's bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66425

-- 
Cyril Hrubis
chrubis@suse.cz