IRC channel logs

<flatwhatson>old: re: parameter vs thread-local fluid, did you mean parameter object? shouldn't that be equivalent to a fluid?

<old>flatwhatson: it is. but there's actual thread-fluid

<old>which are not captured by dynamic-state

<old>which is problematic in guile-parallel because I'm trying to use statprof on scheduler threads

<old>however, the job are executed in a captured dynamic state, which confuse the state of statprof

<old>setting the statprof to a thread-fluid makes it resistant to change in dynamic-state, which is great for context-switching!

<flatwhatson>ah interesting. did you have to hack statprof for that?

<old>basically copy/paste statprof.scm in the projet and change the parameter for a thread fluid

<flatwhatson>my attempt at a hygienic high-level ffi: https://notabug.org/flatwhatson/guile-openai/commit/8ca1f32a51419bb454419673a7990cc2f46476ca

<old>flatwhatson: here's mine for guile-ev: https://git.sr.ht/~old/guile-ev/tree/master/item/ev/foreign.scm

<old>and its usage here: https://git.sr.ht/~old/guile-ev/tree/master/item/ev/ffi.scm.in

<old>It's kind of slow though. Compiling ffi.scm is very long :-/

<civodul>old: i just mapped your nickname to your real name! :-)

<old>civodul: ^^

<civodul>you're doing HPC stuff when you don't write Guile code, right?

<old>Not really. I do instrumentation related stuff

<old>right now I do work on a HPC exascale project, on the instrumentation part.

<dthompson>can anyone explain the mysterious 3rd argument to call-with-values when looking at a backtrace?

<dthompson>run this: (call-with-values (lambda () (error 'oh-no)) list)

<dthompson>then ,bt in the debugger

<dthompson>and you'll see a frame like: 2 (_ #<procedure c32b18 at <unknown port>:34:18 ()> #<procedure list _> 1)

<dthompson>what the heck is the 1 doing at the end of the call?

<dthompson>and how about something even stranger: (append '(a b c) (call-with-values (lambda () (error 'oh-no)) list))

<dthompson>backtrace frame: 2 (_ #<procedure c346e0 at <unknown port>:37:35 ()> #<procedure list _> . #<unknown-type (0xb . 0x304) @ 0x7fe944c8d280>)

<dthompson>even worse, want to make guile segfault?

<dthompson>eval: (symbol? (call-with-values (lambda () (error 'oh-no)) list))

<dthompson>then run ,bt and say bye-bye repl

<Arsen>heh, nice

<dthompson>seems like a VM bug

<dthompson>or something... I dunno. going to file a bug report

<old>are we playing segfault golf?

<dthompson>yes :)

<dthompson>this took me quite some hours, actually

<dthompson>over at Spritely, we were scratching our heads over a very strange segfault issue

<old>try me. I'm a good hunter for that

<dthompson>after some fiddling with gdb, I traced it to frame-local-ref.

<dthompson>and after some more time I traced it to call-with-values stack frames.

<dthompson>the hunt is over :)

<dthompson>I'm about to file a bug report

<old>ah okay it was that ^^

<dthompson>bug report filed. zero thanks to gmail for mangling my original text with its auto word wrap -_-

<dthompson>this is what I deserve for still using gmail...

<old>*laught in emacs-notmuch*

<dthompson>;)

<dthompson>used to use that with offlineimap but got sick of dealing with it

<old>dthompson: I kind of have mix feeling toward it

<old>mu4e was great last time I used it. Perhaps I should migrate back to this workflow.

IRC channel logs

2023-05-04.log