libera/#clasp - IRC Chatlog

23:31:51 drmeister bike: What can I do about this...

23:31:53 drmeister https://usercontent.irccloud-cdn.com/file/Y0S4NsRK/image.png

23:32:01 drmeister sbit being very slow.

0:19:59 bike drmeister: i think i need to put in a fix for that one, gimme a bit. and probably just use aref directly with the array type declared

0:20:52 drmeister Ok. Thank you.

0:22:06 drmeister Is it done by adding these?

0:22:08 drmeister https://www.irccloud.com/pastebin/EjX0IhpH/

0:22:18 drmeister I haven't tried it yet. but that's what claude came up with.

0:22:28 drmeister I won't push anything to clasp/main

0:25:45 bike those would probably be good too, but the underlying operations aren't hooked up. we have optimizers for a few element types but not bit because i hadn't got around to it and they can't be addressed like the bigger types can

0:26:14 bike we have cc_simpleBitVectorAref and Aset already, they just need to work with the primops

0:26:21 drmeister Yeah - I figured it wasn't that easy.

0:41:59 bike aref done, aset hit an llvm assert failure so i'll fix that, then i'll double check it doesn't break the build and i'll push

0:42:13 drmeister Thank you very much

0:46:35 bike i'll put in the sbit transforms too, so you shouldn't need to change the code

0:47:25 drmeister I won't. I'm going to try again to be a better person and use claude to teach me how to use git better and submit main changes using PR's

0:47:38 drmeister I'm putting one together for two small changes.

0:49:53 drmeister A "core:fsync" function for speeding up NFS writes and adding a missing `ext:float-nan-string`

0:58:03 bike PRs are nice. the CI running tests is a godsend

1:03:24 drmeister bike: In the comments for the trampolines PR that I submitted. You asked here:

1:03:25 drmeister https://usercontent.irccloud-cdn.com/file/qiGsokSB/image.png

1:03:59 drmeister You want the comment added to the code - right? If I do that I just push the change and then state in my comment that I did that?

1:04:21 drmeister And it will just get added to the PR - correct?

1:05:42 bike if you add commits to the branch and push to the branch, the PR is updated

1:08:33 drmeister Ok, but that was how you wanted it dealt with, a comment added to the code. I changed it to look like this...

1:09:07 drmeister https://www.irccloud.com/pastebin/9qRNbb4c/

1:09:26 drmeister I see - but I push this and it will show up in the PR and in the conversation presumably.

1:09:29 drmeister I'll try it

1:11:13 bike Right

1:13:57 drmeister I pushed the change but the conversation still shows the old code.

1:13:58 drmeister https://usercontent.irccloud-cdn.com/file/JN4tcRqo/image.png

1:14:00 bike bit vector changes build and the tests run for me - i'm just going to push directly to main since this really shouldn't break anything on mac (knock on wood)

1:14:51 bike well i might have to go in and look at the new changes, but that's no problem

1:15:36 drmeister I did this in the conversation

1:15:37 drmeister https://usercontent.irccloud-cdn.com/file/EzSHsNln/image.png

1:15:47 drmeister Would you like me to behave like this?

1:17:47 bike you don't need to be very verbose - honestly just saying "done" is probably enough if it's something simple. i get notifications if there are new commits, anyway

1:18:27 bike Pushed the bit vector changes, let me know how it goes

1:18:41 bike I also merged frgo's FASL load time improvement earlier, so hopefully that helps too

1:18:56 drmeister Ok, "done" is good. I'll make changes in the code.

1:20:33 bike the conversations are mostly good if there's something complicated. e.g. here's frgo responding to my comment to explain that no, he shouldn't make the change i implied, since linux does this undocumented thing https://github.com/clasp-developers/clasp/pull/1770#discussion_r3342246198

1:42:38 drmeister I responded to everything, changed code and pushed changes.

1:42:46 drmeister How does that look. Did I do good?

1:44:19 drmeister You said it was failing tests. Where do you see that?

2:34:12 bike if the branch is mergeable the CI will run tests automatically, but right now there's a merge conflict (over clasp_gc.sif, so no big deal) as you can see above the comment box, which means github won't test

2:34:25 bike the test failures before your changes are back in https://github.com/clasp-developers/clasp/actions/runs/26539508904

2:35:24 bike i think the relevant ones are the DESCRIBE-FUNCTION-DOCUMENTATION-A et al tests failing, but that wasn't due to the trampoline changes, it was due to the XepSimpleFunction changes and that problem is now fixed in main

3:13:59 bike your comments look good, i'll go over it in more detail in the morning

4:00:05 drmeister the SBIT change looks good.

4:00:07 drmeister https://usercontent.irccloud-cdn.com/file/kaT39ldW/image.png

4:00:15 drmeister That's the new flame chart. No sbit to be seen.

4:04:05 bike very good... those typeps probably shouldn't be happening though

4:05:08 bike oh, wait, bct is bytecode trampoline?

4:05:23 bike i guess that makes a bit more sense then, the-single doesn't get optimized out in bytecode

4:10:57 bike i could probably make them a little less dumb if need by by caching type checkers in the-single.