Open JustinAzoff opened 3 months ago
Also, are all weirds tested? do we have a pcap that generates each one?
A very naive survey of grepping the source for "Weird(", filtering the results a bit, and then searching through testing/btests/Baselines for matches shows we're not anywhere close to 100% test coverage on the weirds. This matches with the lack of test coverage on a lot of the error cases across the source in general though (see also https://github.com/zeek/zeek/issues/2283).
This ended up on my radar recently due to popular LLMs completely hallucinating what
DNS_Conn_count_too_large
means.
Considering we don't have test coverage for any of the DNS wierds other than the EDNS ones, this isn't surprising.
Weirds are generated as strings and are not documented like other
enum
s.This ended up on my radar recently due to popular LLMs completely hallucinating what
DNS_Conn_count_too_large
means. This is because that weird and many others like it are not really documented anywhere.For each weird we should probably have:
Also, are all weirds tested? do we have a pcap that generates each one?