Corrupted data when using bitshuffle? about python-blosc HOT 4 CLOSED

I try this example with typesize=1, and it seems to work:

>>> blosc.decompress(blosc.compress(xx, typesize=1, clevel=5, cname='zstd', shuffle=blosc.BITSHUFFLE))[-3:]
b'\x01\x01\x01'

Am I required to always set the typesize so len(buffer) % typesize == 0? Does it seem somewhat dangerous when you can end up with corrupted data if you forget to set the typesize?

from python-blosc.

Yep. The bitshuffle filter is a bit finicky on the buffer sizes. That was a bug in C-Blosc, so I fixed it in the copy here (but still needs to be propagated upstream).

from python-blosc.

master on travis-ci is still broken, should we re-open?

from python-blosc.

Yeah, what is failing is the new test introduced in a1d1f6f. The problem is that I cannot not reproduce it on neither my Linux or my Mac box :/ . Any help trying to reproduce this would be welcome!

from python-blosc.