I've done some testing with pybitcointools and one of the BitcoinCore test vectors with a SIGHASH of zero.

For SIGHASH =0, the Tx which is signed

1. does not have the 4 nullbytes appended, ie for SIGHASH 1 there'd be 01000000 appended
2. the Tx structure is hashed, and the hash is reversed

See this test vector

I've done some testing with pybitcointools and one of the BitcoinCore test vectors with a SIGHASH of zero.

For SIGHASH =0, the Tx which is signed

It's just a matter of appending 4 nullbytes, instead of 01000000, to the signing transaction.

The Core software checks for this non-standard hashtype now, however we can see there's no problem in validating these rare transactions since the code only checks for SIGHASH ACP, SINGLE & NONE. See Python-Bitcoinlib's code for a Python interpretation

See this test vector, where the hash being validated is 11743b220e9e24e89abd4ff124a2740531fe7d7f9b4e26de14710a532fd543e2.

FWIW, Pybitcointools needs to have the txhash function edited so that instead of if hashcode:, it reads if hashcode is not None:. The else clause in that function was confusing me, but it appears to be for hashing signed transactions (ie to return the TxID)