Vicent Martí

@Alhadis I'm honestly not sure of how exactly does PCRE implement this -- you should be able to test it out by simply downloading libpcre and trying to compile in...

@Alhadis :bow::bow::bow:

The issue is that it's loading `Syntaxes/Oz.tmLanguage` _and_ `Originals/Oz.tmLanguage`. The grammar in `Originals` is not even valid XML, but has the same extension, so we attempt to load both. This...

I haven't looked in depth at the actual new implementation. I thought it would be much more interesting to look at the benchmarks before reviewing it... But the benchmarks were...

@harshit-gangal: the original pool also uses `resourceWrapper` for all connections, but it never allocates it on the heap because it's very careful to pass it always by value. You don't...

It seems like that should be a slice of `[]resourceWrapper` -- without a pointer. That's where the allocation is coming from.

Note that right now all the remaining benchmarks that allocate are significantly slower than their old `static` counterparts. You need to fix those too! Once they don't allocate they will...

We've just noticed @harshit-gangal's https://github.com/vitessio/vitess/pull/10793, which found this bug already and added a check to the `evalengine` to error out when the bug was triggered. This PR follows up on...

@harshit-gangal: the end2end test no longer errors, but it doesn't appear to be giving the right results either: ``` === RUN TestFilterOnLeftOuterJoin mysql.go:188: Query (select team.id from team_fact join team...

Yes, we're pausing this until the rest of the tooling ecosystem catches up with 1.18.