vfprintf: Add test case for user-defined types and format specifiers
Commit Message
On 06/29/2017 12:51 PM, H.J. Lu wrote:
> On Thu, Jun 29, 2017 at 12:31 AM, Florian Weimer <fweimer@redhat.com> wrote:
>> On 06/19/2017 03:48 PM, Florian Weimer wrote:
>>> 2017-06-19 Florian Weimer <fweimer@redhat.com>
>>>
>>> * stdio-common/tst-vfprintf-user-type.c: New file.
>>> * stdio-common/Makefile (tests): Add tst-vfprintf-user-type.
>>
>> I am going to commit this soon.
>>
>
> It fails on i686:
>
> tst-vfprintf-user-type.c: In function \u2018my_printf_function\u2019:
> tst-vfprintf-user-type.c:58:55: error: format \u2018%c\u2019 expects
> argument of type \u2018int\u2019, but argument 7 has type
> \u2018wchar_t {aka const long int}\u2019 [-Werror=format=]
> tst-vfprintf-user-type.c: In function \u2018my_arginfo_function\u2019:
> tst-vfprintf-user-type.c:112:55: error: format \u2018%c\u2019 expects
> argument of type \u2018int\u2019, but argument 7 has type
> \u2018wchar_t {aka const long int}\u2019 [-Werror=format=]
Please try the attached patch.
The need for the wint_t cast looks like a GCC bug.
Thanks,
Florian
Comments
On Thu, Jun 29, 2017 at 4:34 AM, Florian Weimer <fweimer@redhat.com> wrote:
> On 06/29/2017 12:51 PM, H.J. Lu wrote:
>> On Thu, Jun 29, 2017 at 12:31 AM, Florian Weimer <fweimer@redhat.com> wrote:
>>> On 06/19/2017 03:48 PM, Florian Weimer wrote:
>>>> 2017-06-19 Florian Weimer <fweimer@redhat.com>
>>>>
>>>> * stdio-common/tst-vfprintf-user-type.c: New file.
>>>> * stdio-common/Makefile (tests): Add tst-vfprintf-user-type.
>>>
>>> I am going to commit this soon.
>>>
>>
>> It fails on i686:
>>
>> tst-vfprintf-user-type.c: In function \u2018my_printf_function\u2019:
>> tst-vfprintf-user-type.c:58:55: error: format \u2018%c\u2019 expects
>> argument of type \u2018int\u2019, but argument 7 has type
>> \u2018wchar_t {aka const long int}\u2019 [-Werror=format=]
>> tst-vfprintf-user-type.c: In function \u2018my_arginfo_function\u2019:
>> tst-vfprintf-user-type.c:112:55: error: format \u2018%c\u2019 expects
>> argument of type \u2018int\u2019, but argument 7 has type
>> \u2018wchar_t {aka const long int}\u2019 [-Werror=format=]
>
> Please try the attached patch.
>
> The need for the wint_t cast looks like a GCC bug.
>
Works for me.
Thanks.
On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
> The need for the wint_t cast looks like a GCC bug.
%lc wants wint_t.
Andreas.
On 06/29/2017 02:39 PM, Andreas Schwab wrote:
> On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
>
>> The need for the wint_t cast looks like a GCC bug.
>
> %lc wants wint_t.
Sure, but why warn about this? I doubt there are any architectures
which define wchar_t and wint_t in such a way that the variable
arguments promotion diverges.
Thanks,
Florian
On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
> On 06/29/2017 02:39 PM, Andreas Schwab wrote:
>> On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
>>
>>> The need for the wint_t cast looks like a GCC bug.
>>
>> %lc wants wint_t.
>
> Sure, but why warn about this?
Because the standard says so.
> I doubt there are any architectures which define wchar_t and wint_t in
> such a way that the variable arguments promotion diverges.
The standard does not make that distinction.
Andreas.
On Thu, 29 Jun 2017, Florian Weimer wrote:
> On 06/29/2017 02:39 PM, Andreas Schwab wrote:
> > On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
> >
> >> The need for the wint_t cast looks like a GCC bug.
> >
> > %lc wants wint_t.
>
> Sure, but why warn about this? I doubt there are any architectures
> which define wchar_t and wint_t in such a way that the variable
> arguments promotion diverges.
The different signedness can break the ABI, if you pass a negative wchar_t
(so sign-extended to 64-bit) but the called function expects wint_t
(zero-extended to 64-bit). (E.g., the powerpc64 ABI requires such
zero-extension of an unsigned argument.)
On 06/29/2017 04:37 PM, Joseph Myers wrote:
> On Thu, 29 Jun 2017, Florian Weimer wrote:
>
>> On 06/29/2017 02:39 PM, Andreas Schwab wrote:
>>> On Jun 29 2017, Florian Weimer <fweimer@redhat.com> wrote:
>>>
>>>> The need for the wint_t cast looks like a GCC bug.
>>>
>>> %lc wants wint_t.
>>
>> Sure, but why warn about this? I doubt there are any architectures
>> which define wchar_t and wint_t in such a way that the variable
>> arguments promotion diverges.
>
> The different signedness can break the ABI, if you pass a negative wchar_t
> (so sign-extended to 64-bit) but the called function expects wint_t
> (zero-extended to 64-bit). (E.g., the powerpc64 ABI requires such
> zero-extension of an unsigned argument.)
I see. Based on what is said in the standard, the cast is indeed
required. That's rather unfortunate from a programmer convenience point
of view.
Florian
vfprintf: Fix tst-vfprintf-mbs-prec and tst-vfprintf-user-type
2017-06-29 Florian Weimer <fweimer@redhat.com>
* stdio-common/tst-vfprintf-user-type.c (my_printf_function):
Adjust format string.
* stdio-common/Makefile (tst-vfprintf-mbs-prec.out): Add locale
dependency.
@@ -87,6 +87,7 @@ $(objpfx)tst-grouping.out: $(gen-locales)
$(objpfx)tst-sprintf.out: $(gen-locales)
$(objpfx)tst-sscanf.out: $(gen-locales)
$(objpfx)tst-swprintf.out: $(gen-locales)
+$(objpfx)tst-vfprintf-mbs-prec.out: $(gen-locales)
endif
tst-printf-bz18872-ENV = MALLOC_TRACE=$(objpfx)tst-printf-bz18872.mtrace
@@ -29,6 +29,7 @@
#include <support/check.h>
#include <support/support.h>
#include <support/test-driver.h>
+#include <wchar.h>
/* Initialized by do_test using register_printf_type. */
static int user_type;
@@ -55,8 +56,9 @@ my_printf_function (FILE *fp, const struct printf_info *info,
const void *const *args)
{
if (test_verbose > 0)
- printf ("info: %s (%p, %p, {%p}@%p) called for %%%c (prec %d)\n",
- __func__, fp, info, args[0], args, info->spec, info->prec);
+ printf ("info: %s (%p, %p, {%p}@%p) called for %%%lc (prec %d)\n",
+ __func__, fp, info, args[0], args, (wint_t) info->spec,
+ info->prec);
TEST_VERIFY (info->spec == 'P');
size_t nargs;
@@ -109,8 +111,9 @@ my_arginfo_function (const struct printf_info *info,
if (info->spec != 'P')
return -1;
if (test_verbose > 0)
- printf ("info: %s (%p, %zu, %p, %p) called for %%%c (prec %d)\n",
- __func__, info, n, argtypes, size, info->spec, info->prec);
+ printf ("info: %s (%p, %zu, %p, %p) called for %%%lc (prec %d)\n",
+ __func__, info, n, argtypes, size, (wint_t) info->spec,
+ info->prec);
TEST_VERIFY_EXIT (n >= 1);
size_t nargs;