Possible WideCharToMultiByte CP_UNIXCP bug

[Gabriel Ivăncescu]

Hi,

In the course of investigating why patch 169366 fails on test VMs with 
locales, I found out a possible bug or quirk in Wine's implementation of 
CP_UNIXCP.

Since this code page is not found on Windows, I cannot know what is 
"correct" behavior here.

So, WideCharToMultiByte(CP_UNIXCP, ...) with a locale doesn't seem to 
use CP_UTF8, but instead uses unix_cptable.

The problem seems to be in the fact that the 'used' parameter is *never* 
set in this case. It passes 'used ? &used_tmp : NULL' and then breaks 
out of the switch -- and the 'if (used) *used = used_tmp;' statement 
only applies in the default case of the switch. So I'm basically using 
an uninitialized variable.

Is this intended, or a bug?

What should I do then for patch 169365? Should I simply use CP_UTF8 or 
ignore the 'used' parameter?

Or perhaps fix this bug and set the 'used' parameter properly for 
CP_UNIXCP in the implementation of WideCharToMultiByte? Is there a 
reason it is even using used_tmp in the first place?

Thanks,
Gabriel

[Alexandre Julliard]

Gabriel Ivăncescu <gabrielopcode(a)gmail.com> writes:
> The problem seems to be in the fact that the 'used' parameter is
> *never* set in this case. It passes 'used ? &used_tmp : NULL' and then
> breaks out of the switch -- and the 'if (used) *used = used_tmp;'
> statement only applies in the default case of the switch. So I'm
> basically using an uninitialized variable.
>
> Is this intended, or a bug?

It's a bug, thanks for catching it.

-- 
Alexandre Julliard
julliard(a)winehq.org

[Gabriel Ivăncescu]

On 9/4/19 5:32 PM, Alexandre Julliard wrote:
> Gabriel Ivăncescu <gabrielopcode(a)gmail.com> writes:
> > The problem seems to be in the fact that the 'used' parameter is
> > *never* set in this case. It passes 'used ? &used_tmp : NULL' and then
> > breaks out of the switch -- and the 'if (used) *used = used_tmp;'
> > statement only applies in the default case of the switch. So I'm
> > basically using an uninitialized variable.
> >
> > Is this intended, or a bug?
>
> It's a bug, thanks for catching it.

I've sent some patches to fix this and simplify CP_UNIXCP for callers.