Does this mean that secondary indexes which had a .match could be in a broken, nonsensical state?

I hadn't thought of that. Indexes that call match with an empty string might be broken If env_t gets reused to compute indexes. A quick look at the code makes me think that they aren't being reused.

Secondary indexes are supposed to use their own env_t, yeah.

Okay, I get it now -- empty regexes are going to get clobbered and point at the wrong iterator. So secondary index functions that use an empty regex, followed by a non-empty regex, are going to run into trouble.

As far as I can tell, the patch in #6241 broke secondary index compatibility around this in 2.3.6, for those that were using empty regular expressions.

So now we can't make 2.4 offer a backwards-compatible implementation layer in either direction.

See #6562.