invalid output from deflate_medium

[jsnell]

The deflate_medium strategy occasionally produces outputs that are corrupt in various ways. The archive http://jsnell.iki.fi/tmp/deflate-medium-testcase.tar.gz has a minimal test case. The result of running it against this zlib at level 5 can't be decompressed, but can be if run at level 9. Likewise decompressing the output works with baseline zlib at all levels.

This test case has been heavily pruned from a much larger one. I'm hoping there's just one bug here, but it's possible there are multiple, since the behavior is occasionally different with the larger test cases. (I can't really direct the automated pruning process in any way, so if there are multiple issues, creating a minimal test case would require that this issue be fixed).

[jtkukunas]

Sorry for the delay in responding ...

I can reproduce the issue with your testcase. Interestingly, it doesn't occur when decompressing with minigzip or gzip, as opposed to openssl. I'll investigate and get back to you.

Thanks.

[joshtriplett]

Any progress tracking down the issue?

[jtkukunas]

Yeah, I tracked the issue down to an edge case, triggered by the sync flush in the testcase, where we would update the hash with invalid data.

jsnell,

I've pushed a branch, issue2, that should resolve the issue. Can you confirm whether this fixes the bug for you? Specifically, commit ef27907.

Thanks.

[jsnell]

Thanks, that fixes my problem.