fix: return EOF when buffer is empty - #472
Conversation
The bra reader could return invalid results when a read ended exactly so no data was left: In this case, the rc.rc read returned (0, EOF), rc.n was 0. Therefore the rc.buf.Read() was called with an empty slice, which caused it to return (0, nil), which was then returned to the caller, and left the reader in the same state as before. In combination with e.g. io.ReadAll, this could cause an endless loop.
|
Is there an archive available that demonstrates the problem so it can be added as a test case? |
|
I'll upload a test. Please note that I don't own (or trust) that test data file; and I've noticed that reads on some other files in it fail, so it might even be partially invalid... |
|
@bodgit Have you had the time to look at the uploaded file yet? |
Trying to extract the whole archive I get: Either the archive is corrupt or it's triggering a bug in the LZMA library that I use. That means I can't add that file to the existing test that just tries to extract the whole archive, but your new test case that just extracts the one file seems to be enough to trigger the bug and it's reproducible on my side; it just gets stuck in a loop. Your fix seems to work and doesn't impact any of the other existing tests so I will merge this shortly. |
|
I cherry-picked your commits and tweaked the test in #476 so it is wrapped in a timeout and quickly times out rather than wait for the default This will be in the next release. Thanks for the contribution! |
2 participants
The bra reader could return invalid results when a read ended exactly so no data was left:
In this case, the rc.rc read returned (0, EOF), rc.n was 0.
Therefore the rc.buf.Read() was called with an empty slice, which caused it to return (0, nil).
This was then returned to the caller, and left the reader in the same state as before.
In combination with e.g. io.ReadAll, this could cause an endless loop.