Re: [PATCH v2 13/30] ldbl-128ibm-compat: Add ISO C99 versions of scanf functions

["Gabriel F. T. Gomes" <gabriel at inconstante dot net dot br>]

Hi, Paul,

I'm sorry I took so long to reply to this message.  It sent me down a
rabbit hole.

On Fri, 08 Nov 2019, Paul E Murphy wrote:
>
>On 10/25/19 10:33 AM, Gabriel F. T. Gomes wrote:
>> From: "Gabriel F. T. Gomes" <gabrielftg@linux.ibm.com>
>> 
>> --- /dev/null
>> +++ b/sysdeps/ieee754/ldbl-128ibm-compat/test-isoc99-scanf-ieee128.c
>> @@ -0,0 +1,13 @@
>> +#undef _GNU_SOURCE
>> +/* The following macro definitions are a hack.  They word around disabling
>> +   the GNU extension while still using a few internal headers.  */  
>
>Indeed a hack. How frail is this?

A bit too much! :B

I spent some time trying to remember why I thought that was necessary and
I have a better understanding, now.

This hack is only really necessary for the wide-character versions of the
tests, because in wcsmbs/wchar.h, the redirections for vfwscanf, vwscanf,
and, more specifically, vswscanf to their ISO C99 versions is protected by
#if !defined __USE_GNU, whereas the redirections for fwscanf, wscanf, and
swscanf, as well as for all of the non-wide equivalents is protected by
#if !__GLIBC_USE (DEPRECATED_SCANF).

The use of DEPRECATED_SCANF is somewhat recent (added by commit ID
03992356e6fe) and it is my understanding that the change is missing for
vswscanf (and friends) simply due to oversight.

I have a patch to implement the missing change and I just finished running
it with build-many-glibcs.py.  I'll send it for review.

>> +#define u_char unsigned char
>> +#define u_short unsigned short
>> +#define u_int unsigned int
>> +#define u_long unsigned long
>> +#include <libio/stdio.h>
>> +#include <stdlib/stdlib.h>
>> +#include <string/string.h>
>> +#include <wcsmbs/wchar.h>
>>
>> [...]
>>
>> --- /dev/null
>> +++ b/sysdeps/ieee754/ldbl-128ibm-compat/test-isoc99-wscanf-ieee128.c
>> @@ -0,0 +1,13 @@
>> +#undef _GNU_SOURCE
>> +/* The following macro definitions are a hack.  They word around disabling
>> +   the GNU extension while still using a few internal headers.  */  
>
>The same hack shows up four times, should it be isolated in one place?

With the fix, I'll try and make it all go away.  Thanks for pointing that out.