Add known-to-fail counterexamples to Bech32 string corruption tests. #328
Conversation
The space of valid Bech32 strings is very large. By increasing the number of test runs, we can be more confident that there are no counterexamples to be found, for a very small increase in test time.
…sion test. In the case where the tail of a valid Bech32 string is composed of one or more consecutive 'q' characters followed by a single 'p' character, omitting any or all of the 'q' characters will still result in a valid Bech32 string. We amend the character omission test to allow for this case, so that when it occurs, it won't cause the test to fail.
…rtion test. In the case where the last character of a valid Bech32 string is the character 'p', inserting any number of consecutive 'q' characters immediately before the 'p' will still result in a valid Bech32 string. We amend the character insertion test to allow for this case, so that when it occurs, it won't cause the test to fail.
jonathanknowles
changed the title
| (T.length suffix > 0 | ||
| && T.last suffix == 'p' | ||
| && char == 'q' | ||
| && T.all (== 'q') (T.dropEnd 1 suffix)) |
There was a problem hiding this comment.
Good question. I'm currently updating the issue with a longer explanation. Will let you know when I've finished writing. :)
There was a problem hiding this comment.
In our current test case, we assume that deleting a character from a valid Bech32 string will create an invalid Bech32 string.
I think this is a fair assumption, as Bech32 was designed with the specific goal of collision resistance. When entering one Bech32 string, it's supposed to be very hard to collide with another Bech32 string just by making a small typo.
Experimental testing shows that there is at least one known counterexample to this law: when a Bech32 string ends in qp, it's possible to delete a single q character and still produce a valid Bech32 string. This works for any number of q characters.
The fact that it's possible to produce a collision just by deleting a single character comes as a surprise.
The above change adjusts the test case to account for this counterexample.
Issue Number
Issue #324
Overview