[Mesa-dev] [RFC PATCH 3/4] ff_fragment_shader: mark impossible switch values with unreachable

[Giuseppe Bilotta]

On Mon, Apr 24, 2017 at 5:06 PM, Brian Paul <brianp at vmware.com> wrote:
> On 04/22/2017 03:46 PM, Giuseppe Bilotta wrote:
>>
>> Makes sense. As Gustaw suggests, it _shouldn't_ happen, but there's
>> nothing to lose in being cautious about it 8-). I'll respin this patch
>> with the assert+return here.
>
> I just want to second that.

Unless I did something wrong, a v2 of the patch, with the
assert+return should be floating around in the patchwork now 8-) But
then again, it replaces a lot of the other assert+return with an
unreachable, so you might not like that either 8-P

>  It's really bad when our driver crashes in the
> hands of a customer.  I'd rather have a rendering glitch than a crash.  I'm
> kind of worried about the proliferation of unreachable() in Mesa.

Well, the obvious question would be why such unreachable code actually
gets reached in the first place, but I guess it's not always easy to
get the conditions under which these supposedly invalid codepaths get
reached.

Would it be possible to preserve the assert+return semantics with
something similar to a Linux kernel WARN_ON? (Ideally even with debug
backtrace support, but wrapping it all in a macro would allow this to
be added at a later stage if deemed appropriate/possible).

-- 
Giuseppe "Oblomov" Bilotta