09:02 <cfbolz> mattip: I've written him a message

09:03 <cfbolz> the main known speed regression is the dict stuff

09:09 <mattip> thanks

09:53 <fijal> cfbolz: do we have something in __pypy__ to check "is this a buffer?"

10:06 <cfbolz> fijal: I don't think so

11:08 <fijal> bort: push failed:

11:08 <fijal> 'remote repository changed while pushing - please try again'

11:08 <fijal> is this normal?

11:09 <cfbolz> 🤔

11:09 <cfbolz> never had that

11:13 <fijal> no, I can't push

11:13 <fijal> I can pull, but I can't push

11:16 <mattip> https://foss.heptapod.net/heptapod/heptapod/-/issues/259

11:16 <mattip> maybe remove evolve?

11:20 <fijal> I don't have one?

11:21 <fijal> marmoute: ^^^

11:30 <fijal> any clue how to work around that?

11:30 <fijal> updating mercurial did not help

13:06 <marmoute> This is probably not the same error

13:06 <marmoute> what's your version of everything ?

13:06 <marmoute> Can I get you to jump on the suppose channel at : https://mattermost.heptapod.net/

14:18 <mattip> python pep 653: Immortal objects https://www.python.org/dev/peps/pep-0683/

14:18 <mattip> the first step to removing the GIL

18:56 <Corbin> Is there a hash-cons or union-find helper in rlib? I don't mind writing it, but I was hoping that somebody had already figured out the details around symbols and interning.

19:00 <cfbolz> Corbin: there isn't

19:03 <Corbin> cfbolz: Oh well. Is there a better way to intern than to have a dict as a field on a global PBC?

19:03 <cfbolz> Corbin: intern what? strings?

19:04 <Corbin> cfbolz: I'm parsing a Lisp-like S-expression, with holes. I'm then going to substitute stuff into the holes, as well as pattern-match parts of the tree. So they'll be strings or trees.

19:06 <cfbolz> right

19:06 <cfbolz> for prolog I just coded it

19:06 <cfbolz> (it's a bit different in prolog because of backtracking, so the union find needs to be integrated with the trail)

19:12 <Corbin> Sure. I'm just looking to reuse code. I don't like that I'm not able to convert this sort of opportunity into an rlib improvement.

19:12 <Corbin> Thanks for the advice; I appreciate it.

19:17 <cfbolz> Corbin: it's our experience that some of these things are really quite integrated in every language

19:17 <cfbolz> so in theory they are similar, but in practice it's extremely hard to write something that can cover different use cases

19:18 <Corbin> That's true. And RPython JITs aren't perfect Futamura evaluators either, so it's not just about API but about performance guarantees; sometimes we have to write special code just for the JIT.

19:21 <cfbolz> yes