AhoCorasick::stream_find_iter may fail if the needle is split between buffer. about aho-corasick HOT 3 CLOSED

I've tried to investigate the problem. It appear that for the specific pattern in the testcase, the prefilter used is RareBytesOne { byte1: 106, offset: RareByteOffset { max: 4 } })
What happens is that calling earliest_find_at from ahocorasick::StreamChunkIter::next will not find that byte in the buffer and will therefore skip everything. Then buf.roll will be called to rotate the buffer and read from the Reader. But the next call to earliest_find_at will only search from the newly read data (despite the match is still in the buffer)
RareBytesOne::next_candidate will find the rare byte, but will then discard the match because its start position is earlier than the at argument.

I've tried replacing this line

pos.saturating_sub(self.offset.max as usize)

so that RareBytesOne::next_candidate can return a match smaller than at, also adjusting the call to update_skipped_bytes to handle negative values. And this solve the testcase.
However it causes some other test to run into infinite loops, so that's not quite right.

from aho-corasick.

@ogoffart Thanks so much for that investigation! That was enough to tell me where the problem was and helped me craft a fix. See #67 for a bit more detail, but the short story is that I've disabled the prefilter in this case unfortunately. Fixing stream searching to work with the rare byte prefilter is delicate and I don't trust my test coverage enough to attempt that.

from aho-corasick.

A fix for this is on crates.io in aho-corasick 0.7.14.

from aho-corasick.