https://skia-review.googlesource.com/c/189866 "SkWuffsCodec: Initialize
memory when incomplete" made some buffer zero'ing conditional on a
frame's dirty rectangle, which required moving the this->decodeFrame
call earlier in onIncrementalDecode, since we can't know the dirty
rectangle until after decodeFrame is called.
However, moving that earlier meant that it was now earlier than the
"sk_bzero(pixels.ptr + etc, etc)" call that zero-initialized the
destination buffer that decodeFrame writes to. The actual pixel indexes
that decodeFrame decodes are now overwritten by zeroes.
That prior commit fixed the fuzzer-discovered bug, in that it no longer
swizzles from uninitialized memory. Unfortunately, it's now often
swizzling from all-zeroed memory, so that the decoding is incorrect.
This commit moves the zero-ing to onStartIncrementalDecode, even
earlier. The "dm --match Gif" tests are happy again.
Bug: skia:8235
Change-Id: Iae993b7d9f45e1b375ed99f6cc86536091a92eba
Reviewed-on: https://skia-review.googlesource.com/c/190941
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
b7a1b5189d