Fixed infinite recursion on recursive hash() on CBORTag

[agronholm]

Relates to #198.

[coveralls]

coverage: 93.164% (+0.1%) from 93.069%
when pulling 03dfb8e on cbortag-hash-fix
into c0007fc on master.

[mschwager]

I fuzzed this branch for quite some time (multiple hours) and didn't produce any more crashes. The original segfaults were found in a matter of minutes, so I suspect we've found the low hanging fruit.

[agronholm]

This was no low hanging fruit, I tell you :) It was the hardest fix to date, for me at least.

[agronholm]

What input is producing segfaults now? I only remember one being mentioned, but I fixed that one already.

[mschwager]

This was no low hanging fruit, I tell you :) It was the hardest fix to date, for me at least.

Sorry, I meant low hanging fruit for the fuzzer to find crashes, not for the fix. By low hanging fruit I just meant the fuzzer was able to find the crashes in minutes or hours rather than days or weeks of runtime 👍

What input is producing segfaults now? I only remember one being mentioned, but I fixed that one already.

As far as segfault crashes, there were two:

• unexpected exceptions raised while parsing untrusted inputs using cbor2.loads #198 (comment)
• unexpected exceptions raised while parsing untrusted inputs using cbor2.loads #198 (comment)

I re-ran the fuzzer against this branch and neither of those crashes, or any others, appeared after a few hours of runtime.

[agronholm]

Ok, we're on the same page then. Perhaps the MemoryError still needs to be investigated, but the rest of the errors just need to be wrapped in CBORDecodeError.