IRC channel logs

<ekaitz>ieure: did you try the example from the docs? that should work

<ieure>ekaitz, Yes, all it does is print "No samples recorded."

<ekaitz>hmmm

<ekaitz>mine does show something

<ekaitz>(use-modules (statprof))

<ekaitz>(statprof (lambda ()

<ekaitz> (map 1+ (iota 1000000))

<ekaitz> #f))

<ekaitz>that i mean

<ieure>Oh, I see. That does print something.

<ieure>But my code does not.

<ekaitz>hm!

<ekaitz>maybe your code is optimized away :S

<ieure>The code I'm running is:

<ieure>,use (gnu)

<ieure>(operating-system (inherit (operating-system (inherit (operating-system (bootloader #f) (host-name "test") (file-systems #f))))))

<ekaitz>hmmmm

<ieure>This does almost nothing, but it takes ~25s to evaluate on a pretty fast machine.

<ekaitz>civodul should probably know better how to do this, because it involves guile

<ekaitz>guix*

<ekaitz>and probably most of that is not calling things but expanding and running things on the daemon

<ieure>The daemon isn't involved at all.

<ieure>It's creating records and copying fields between them.

<ekaitz>hmmm

<ieure>Even if it was calling the daemon, I'd see that reflected in the (current-time) calls before/after the form that gets evaluated.

<ieure>And as I mentioned -- the values are identical to the second, even though it takes 25s to eval the code.

<ieure>I used the stopwatch function of my phone to time it since I can't make Guile measure what I'm seeing.

<ekaitz>that's weird and surely has a good explanation

<ekaitz>but I don't have an answer for it

<ieure>Thought maybe it was the printer, changed it to (begin (operating-system ...) #f), but it's still very very very slow.

<ekaitz>hmmm

<ieure>It should be trivial to replicate if you have Guix.

<ekaitz>the expression takes forever to run

<ieure>Yep.

<ekaitz>now the question is, what is it running actually?

<ekaitz>macro expanding it creates a HUGE piece of code

<ekaitz>and now running with the statprof

<ekaitz>no samples

<ekaitz>it's probably something guix does

<ieure>Well, there are two issues, right? One is the poor performance of the Guix code, the other is that being invisible ro Guile's profiling tools.

<ieure>And I'm really only interested in the latter at this time.

<ieure>That seems like a Guile issue, right?

<graywolf>Maybe https://issues.guix.gnu.org/74748 would be of use? I do not know if the cause is the same, but symptoms (slow, but no samples) are

<graywolf>civodul suggested ,o optimization-level 1 as workaround

<ieure>graywolf, Similar but different issue. Constructing the record isn't slow, but inheriting is extremely slow and gets worse-than-quadratically slower with every level of inheritance added.

<ieure>What I'm trying to do is quantify how bad it is, so I can make a useful bug report.

<ieure>This is also not just REPL slowness.

<ieure>If I invoke from the shell with `time guile -c ...', it takes 1.5s for three levels of inheritance, 34s for four, and I started timing five levels, that's been running for 6m40s so far, using 100% CPU, and 10gb RAM.

<ieure>Which seems just completely nuts.

<graywolf>Oh, macro expansion shows a lot of code, the inherited packages is expanded over and over and over

<graywolf>(operating-system (inherit (begin (pk 'x) (operating-system (bootloader #f) (host-name "test") (file-systems #f))))) is illuminating

<graywolf>s/inherited packages/inherited system/

<ieure>Oooooohhhhh, interesting.

<ieure>(begin (define %os (operating-system (bootloader #f) (host-name "test") (file-systems #f))) (operating-system (inherit (begin (pk 'x) %os))))

<ieure>That is much, much faster.

<graywolf>And much, much less code. :)

<graywolf>(in the expansion)

<graywolf>Well, off to watch anime, have fun.

<ieure>graywolf, Thank you very much for the help.

<graywolf>Is scm_make_foreign_object_1 subject to non-local exits? I am thinking how to safely wrap external pointer in a foreign object. Based on the manual it seems that db = ...; scm_make_foreign_object_1(conn_type, db); could leak memory given the right conditions.

<graywolf>Do I need to wrap it in scm_dynwind_begin?

<ekaitz>rlb: reproduced the fibers segfault

<rlb>nice (and not nice)

<ekaitz>i reproduced on a remote machine, which is a little bit of a pain in the balls to deal with but i'll manage

<ekaitz>most of them fail

<ekaitz>9 out of 13 i believe

IRC channel logs

2026-03-22.log