Fix warning caused by unused-result in bug-atexit3-lib.cc
Commit Message
The test case dlfcn/bug-atexit3-lib.cc calls write and doesn't check the
result. When building with GCC 6.2 from IBM's branch, this generates a
warning in 'make check', which is treated as an error. This patch adds a
return variable to get rid of the warning and of the error.
Tested for powerpc64le.
2016-10-28 Gabriel F. T. Gomes <gftg@linux.vnet.ibm.com>
* dlfcn/bug-atexit3-lib.cc (statclass): Assign return of call to
write (defined with __wur) to unused variable.
---
dlfcn/bug-atexit3-lib.cc | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
Comments
On 10/28/2016 03:00 PM, Gabriel F. T. Gomes wrote:
> The test case dlfcn/bug-atexit3-lib.cc calls write and doesn't check the
> result. When building with GCC 6.2 from IBM's branch, this generates a
> warning in 'make check', which is treated as an error. This patch adds a
> return variable to get rid of the warning and of the error.
You could use write_message from test-skeleton.c instead. it has a
proper unused variable guard.
Thanks,
Florian
On Fri, 28 Oct 2016 15:04:13 +0200
Florian Weimer <fweimer@redhat.com> wrote:
> On 10/28/2016 03:00 PM, Gabriel F. T. Gomes wrote:
> > The test case dlfcn/bug-atexit3-lib.cc calls write and doesn't check the
> > result. When building with GCC 6.2 from IBM's branch, this generates a
> > warning in 'make check', which is treated as an error. This patch adds a
> > return variable to get rid of the warning and of the error.
>
> You could use write_message from test-skeleton.c instead. it has a
> proper unused variable guard.
Is it ok to use write_message in c++ code being built as object?
Thanks,
Gabriel
On 10/28/2016 03:32 PM, Gabriel F. T. Gomes wrote:
> On Fri, 28 Oct 2016 15:04:13 +0200
> Florian Weimer <fweimer@redhat.com> wrote:
>
>> On 10/28/2016 03:00 PM, Gabriel F. T. Gomes wrote:
>>> The test case dlfcn/bug-atexit3-lib.cc calls write and doesn't check the
>>> result. When building with GCC 6.2 from IBM's branch, this generates a
>>> warning in 'make check', which is treated as an error. This patch adds a
>>> return variable to get rid of the warning and of the error.
>>
>> You could use write_message from test-skeleton.c instead. it has a
>> proper unused variable guard.
>
> Is it ok to use write_message in c++ code being built as object?
Oh, right, this probably will not work. Bummer.
You could copy the function into the .cc file, though. The unused
attribute may be needed in the long term.
Florian
@@ -4,11 +4,15 @@ struct statclass
{
statclass()
{
- write (1, "statclass\n", 10);
+ size_t unused_ret;
+ (void) unused_ret;
+ unused_ret = write (1, "statclass\n", 10);
}
~statclass()
{
- write (1, "~statclass\n", 11);
+ size_t unused_ret;
+ (void) unused_ret;
+ unused_ret = write (1, "~statclass\n", 11);
}
};