Better testing, both for the Satoshi code and across different implementations, is right at the top of my priority list. So I wrote some code to make it easier to exercise edge cases in Script evaluation: https://github.com/bitcoin/bitcoin/pull/1121/files
The unit tests are data-driven by two files:
script_valid.json contains pairs of (scriptSig,scriptPubKey) that, when used in a transaction, should evaluate to true.
script_invalid.json contains pairs that are invalid for some reason.
I'm looking for help to expand those to cover all the quirky edge cases that might trip up alternative "fully validating" implementations (excluding OP_CHECKSIG/OP_CHECKMULTISIG, I'm still thinking about how best to write test cases for those). If you can help, please jump in and write some more. Assuming you've got the bitcoin source tree cloned the process for adding new tests is easy:
1. Edit src/test/data/script*.json
2. From the src/ directory run:
make -f makefile.unix test_bitcoin
3. Run the unit tests with: ./test_bitcoin
Or if you've already got Script test cases in some other human-readable form please point me to them.
I'm planning on resetting the testnet with a new genesis block and a chain that contains all of the valid tests as spent transactions (and a checkpoint so all those nice test cases aren't overwritten).
It would be really nifty to have a 'bad chain' node running; -connect to it and it would send you otherwise-valid blocks that contained one or more of the invalid transactions from script_invalid.json. Does that sound like a project any of you would like to tackle?