libera/#clasp - IRC Chatlog
Search
15:30:49
Bike
i think i might have slowed down the compiler a bit with how i had to rewrite representation selection. kind of hard to tell at the best of times though
15:34:53
drmeister
yitzi: I see the debugger symbol in jupyterlab - what sort of debugging can I do with it?
15:40:33
drmeister
If I insert a (break "xxxx") into my code - does that interact with the debugger in any way? I see a REPL prompt in the output cell - and I can interact with it now - that's good.
15:44:16
Bike
hopefully the jupyter system uses *invoke-debugger-hook* so breaks end up in its debugger?
16:04:12
yitzi
drmeister: you will get a much better experience than a REPL if you install some extensions and use the jupyter debugger ui
16:05:08
drmeister
Bike: I'm looking forward to your floating point math improvements. I'm doing math on 3D points now.
16:05:34
drmeister
First up - determine the plane that minimizes the distance to a collection of points in a ring.
16:05:56
Bike
i'm a bit worried about the compiler slowdown, but i just ran the numerics ansi tests and other than one stupid bug it seems fine
16:09:55
Bike
compiler slowdown was the problem with type inference before... well, and just that it didn't help much, i guess, but things look better in that respect this time
16:11:04
yitzi
drmeister: You also need to turn on the Jupyter debugger. There is a switch in the upper right hand corner for it.
16:33:26
drmeister
I'm translating this code from: https://www.ilikebigbits.com/2015_03_04_plane_from_points.html
16:33:42
Bike
might need some more type declarations, e.g. i don't see anything saying xx is a single or that geom:x returns a single
16:35:14
Bike
(declaim (ftype (function (geom:vec3) double-float) geom:x ...)) should do it, i think
16:43:07
drmeister
Hmm, and we don't use a wrapped type for vectors - I'm translating into a list of three numbers. I should wrap a type.
16:45:39
drmeister
This is where we need to come up with something new. How do you manipulate things like vectors efficiently in Common Lisp?
16:46:15
drmeister
I don't want to write this crap in C++ - that's exactly the kind of thing I wrote Cando to avoid.
18:36:42
Bike
i think there are some deep flaws in how cleavir does graph modifications, but it should be okay for these basic optimizations
20:17:55
drmeister
Is this functionality exposed? Can we extend it easily? Or is it deep in the compiler?
20:51:15
Bike
if you mean what i'm doing, it's not that deep in the compiler, but it's not easy to write a transform
20:53:14
Bike
stupid question: can i include libm functions in the intrinsics list? like sin, cos, that kind of thing