Re: Strange wined3d crash

[Francois Gouget]

On Wed, 29 Apr 2015, Henri Verbeet wrote:
> On 29 April 2015 at 12:40, Francois Gouget  wrote:
> > I am getting a strange wined3d crash during the d3d8:device test.
>
> Is that a regression? We did touch the shader generation code recently.

It is actually. This is the patch that introduced the crash:

commit 28343db2083de7b4afe4079da090520f44b5544d
Author: Matteo Bruni 
Date:   Mon Apr 27 17:39:12 2015 +0200

wined3d: Don't use the builtin FFP uniforms for the lights.

While at it, use D3D-like uniforms for the light parameters and
    (consequently) implement range and proper D3D spotlights support.

I can test patches but I'm not sure I'll have time to debug this much 
further for a while :-( Note that this is with an Intel HD Graphics 4600 
GPU.

Strange wined3d crash

[Matteo Bruni]

2015-04-29 18:03 GMT+02:00 Francois Gouget :
> On Wed, 29 Apr 2015, Henri Verbeet wrote:
> > On 29 April 2015 at 12:40, Francois Gouget  wrote:
> > > I am getting a strange wined3d crash during the d3d8:device test.
> >
> > Is that a regression? We did touch the shader generation code recently.
>
> It is actually. This is the patch that introduced the crash:
>
> commit 28343db2083de7b4afe4079da090520f44b5544d
> Author: Matteo Bruni 
> Date:   Mon Apr 27 17:39:12 2015 +0200
> wined3d: Don't use the builtin FFP uniforms for the lights.
>
> While at it, use D3D-like uniforms for the light parameters and
> (consequently) implement range and proper D3D spotlights support.
>
>
> I can test patches but I'm not sure I'll have time to debug this much
> further for a while :-( Note that this is with an Intel HD Graphics 4600
> GPU.

Thanks Francois, I can reproduce.

Essentially it crashes in shader_vaddline() as soon as we have to
resize the buffer because vsnprintf() tries to use "args" for a second
time.

I attached a fix. Henri, is something like that okay with you?

Strange wined3d crash

[Henri Verbeet]

On 29 April 2015 at 19:02, Matteo Bruni  wrote:
> Essentially it crashes in shader_vaddline() as soon as we have to
> resize the buffer because vsnprintf() tries to use "args" for a second
> time.

Oops. I guess this really got broken with
cf12f51b1520546dc855e8e8e43135a06e7a0cbf already, but the larger
initial buffer size made it work most of the time.
> I attached a fix. Henri, is something like that okay with you?

In principle, but I think va_copy() is C99; I'm not sure we can rely
on it being present. Since shader_vaddline() only has a couple of
callers, it may be better to just move the resizing code into a helper
function and use that from the callers.