freenode/#clasp - IRC Chatlog
Search
5:05:54
drmeister
Some of the timing results were off a bit because of generic function discriminator compilation. I added a warm up stage to the micro-benchmarks that evaluates the form a few times before running the timing loop.
13:00:49
Colleen
Bike: drmeister said 5 hours, 48 minutes ago: Is regular inlining of things like primop:two-arg-- still broken?
13:31:46
attila_lendvai1
hi! did the build get slower? or is dev broken? or did i break it? seems to be stuck at "(DEFSTRUCT (COMPILER-STYLE-WARNING (:TYPE VECTOR) ...))"
13:34:06
Nephromancer
@attilla_lendvai1 are you talking about the docker image? one of the other undergrads in our lab was having some issues with that this morning
13:35:17
attila_lendvai1
Nephromancer: no, i'm just building dev + a bitvector cleanup of mine locally
13:46:36
attila_lendvai1
BTW, the cando docker file has apt-get install references to llvm-5.0 stuff. i guess that's a mistake
14:04:17
Bike
defstruct has been slow to compile for a while. dunno about increasing memory, though.
15:15:54
drmeister
I'm getting an error when loading jupyter lab that chem:define-tests is not defined - is this something that you are adding? It has to do with smarts.
15:18:17
drmeister
https://github.com/drmeister/cando/blob/102ad64eb4bc9459d4c582819991da25400d4438/src/chem/chemInfo.cc#L314
16:30:34
Bike
i did somethin, and now C-c C-c resultsin different source offset numbers from a full compile file. but M-. goes to the right place either way.
16:32:41
Bike
oh. i see. emacs's idea of source position is different from clasp's, but they're both right up to next-sexp, so it works.
16:39:09
Bike
drmeister: you put in something to hedge if compiled-function-file is not passed a function?
16:39:33
drmeister
Yes - because when there are no debugging frames any error threw it into an infinite loop
16:40:32
drmeister
What we need is better way of getting backtraces what we have is hacked together bullshit.
16:41:14
drmeister
https://github.com/clasp-developers/clasp/blob/dev/src/lisp/kernel/lsp/top.lsp#L1010
16:42:55
Bike
with C-c C-c i got offset 161, with compile-file it was 38... but the only thing between 38 and 161 is whitespace and comments, which emacs ignores anyway
16:48:24
Bike
with inlining i hit another problem instead. also finicky. inlining makes the compile enormously slower anyway so i don't want to bother right now
16:49:25
karlosz
after inlining my system funcall instruction it complains that one of the slots are unbound
16:49:29
Bike
but i don't think that's what it's doing. i've messed up and given it incorrect offsets before, and it's caused errors
16:54:13
attile_lendvai
then it must be my changes... :/ we'll see, i rebased them to some older commit and started a build
16:55:30
karlosz
i was thinking about using stealth mixins and specializing methods on stuff like delte-instruction
16:56:35
karlosz
so delete-instruction can specailize on a liveness mixin to update the liveness information associated with the instruction