[PDF] If matrix inversion fails, use the identity matrix.

[PDF] If matrix inversion fails, use the identity matrix.

BUG=chrome:123078

Committed: https://code.google.com/p/skia/source/detail?r=3676

[thakis, 2012-04-13 20:43:09]

lgtm stamp based on reed's patch (
http://code.google.com/p/skia/source/detail?r=3657 ) looking identical.

If many callers of inverse() do this, maybe inverse() should do this itself when
it fails and not be warn_unused_result.

[Steve VanDeBogart, 2012-04-13 20:45:27]

On 2012/04/13 20:43:09, thakis wrote:
> lgtm stamp based on reed's patch (
> http://code.google.com/p/skia/source/detail?r=3657 ) looking identical.
> 
> If many callers of inverse() do this, maybe inverse() should do this itself
when
> it fails and not be warn_unused_result.

I agree, I'll bring it up with Mike.

[bungeman, 2012-04-13 21:38:30]

This set of changes leaves much to be desired, and is probably worse than
leaving the warnings. In all the cases I see here, if the matrix were singular
you need to not draw or cause future draws not to happen. If the inverse fails
it does not mean "oops" it means that the matrix you were trying to invert draws
nothing (as it has no area).

Mike made some changes earlier but is planning on going back to fix them. They
should not be seen as models of how to handle the situation. inverse() most
certainly should *not* be returning identity in this case, if anything it should
be returning the matrix which maps everything to 0 (though this would less
efficient than simply skipping draws).