[4/5] msvcrt: Move write flag setting code for reuse.

[Grazvydas Ignotas]

On Tue, Jun 17, 2014 at 1:16 PM, Piotr Caban <piotr.caban at gmail.com> wrote:
> On 06/14/14 01:23, Grazvydas Ignotas wrote:
>>
>> ---
>>   dlls/msvcrt/file.c |   27 +++++++++++++++------------
>>   1 file changed, 15 insertions(+), 12 deletions(-)
>>
>> diff --git a/dlls/msvcrt/file.c b/dlls/msvcrt/file.c
>> index 7523bbb..d6fb9ef 100644
>> --- a/dlls/msvcrt/file.c
>> +++ b/dlls/msvcrt/file.c
>> @@ -3641,6 +3641,17 @@ MSVCRT_wchar_t * CDECL MSVCRT_fgetws(MSVCRT_wchar_t
>> *s, int size, MSVCRT_FILE* f
>>     return buf_start;
>>   }
>>
>> +static inline int msvcrt_set_write_direction(MSVCRT_FILE* file)
>> +{
>> +    if(!(file->_flag & MSVCRT__IOWRT)) {
>> +        if(file->_flag & MSVCRT__IORW)
>> +            file->_flag |= MSVCRT__IOWRT;
>> +        else
>> +            return MSVCRT_EOF;
>> +    }
>> +    return 0;
>> +}
>
> I don't see a good reason for making this code reusable. The code is used to
> inform how the buffer is used (if it's used for writing or reading). This
> means that it doesn't make sense to set it if buffer is not used. The only
> valid use of it I can see in wine sources is _flsbuf.
>
> I've wrote a test that shows that _IOWRT flag is not set if _bufsiz==0 or in
> the case you're implementing. You can see it's code and result on windows
> here:
> https://testbot.winehq.org/JobDetails.pl?Key=7552

I see, thanks for the test.

Not setting the flag in that case however uncovers another problem
where after rewind/fflush _cnt is not 0, so _flsbuf doesn't get called
and _IOWRT doesn't get set, as a result next fflush doesn't flush the
buffer at all. I've added a few patches that try to deal with that.

-- 
Gražvydas