commit f4871fb981c09a6148217dc2ec396a6393893585
author Rubens Farias <rubensf@google.com> Mon Nov 30 16:24:40 2020 -0500
committer GitHub <noreply@github.com> Mon Nov 30 16:24:40 2020 -0500
tree 09ee2a39c40d2fc3d099b2b64fc61aed70dc7b4c
parent 02e109933bbd578b2d4738185b3e7fcd8a33ff06

Further write compression patch. (#253)

The existing code had a couple issues:
1) An EOF from the underlying uncompressed data would trigger finalizing
a read of the compressed data - even though the compressed data itself
wasn't ready or finalized reading.
2) *After* the `Close` call on the encoder, which triggers the
compressed data to be available for reading, is when you have to
actually read the final chunks and pieces of data.

The test code was also faulty as it checked the bytes up to the length
of the code compressed data (instead of the test compressed data), which
was 0.

This issue never arove on the fakes servers as cas_test doesn't seem to
have File tests - this compression issue wouldn't show up on in memory
blob tests. Just in case, I also added further checks on fake CAS to
ensure compressed calls were actually being made.