[PATCH] Add glretrace option '--snapshot-format=MD5' that allow write image MD5 to a text format file.

[Nigel Stewart]

> your choice of hash algorithm is questionable

The choice of hash algorithm may well be arguable, but I don't find it
so questionable.
The speed of the hash algorithm isn't so critical, in comparison to
writing out .png files.
It tends to be the glReadPixels that is the bottleneck in this kind of
testing scenario.
MD5 is a widely known and available algorithm easily verified by means
of command-line cross-checking.
As a single .c implementation it's a modest addition to what is
already in thirdparty.
I have test coverage running day-in day-out based on MD5 hashing the
color buffer, I'd be likely to make use of a similar scheme for
apitrace testing.

- Nigel

On Tue, Oct 29, 2013 at 8:39 AM, Alexander Monakov <amonakov at ispras.ru> wrote:
> Hi,
>
> It would be nice if you explained the motivation for this change.  Presumably
> this is for improving speed of verification that rendering results are 100%
> reproducible.
>
> If that is true, your choice of hash algorithm is questionable.  Consider that
> bandwidth of RGBA 1920x1080 at 60 fps stream is about 0.5 GB/s.  I'm not certain
> that md5 implementation you're importing can hash at that speed.  Even if it
> can, using MD5 and importing a library definitely seems like an overkill.
>
> Surely you can do with a much simpler and faster (but weaker) hash that would
> not require importing yet another library into thirdparty/?
>
> Thanks.
> Alexander