freenode/#clasp - IRC Chatlog
Search
2:22:00
drmeister
Bike: I mean to test it against the clasp source code - not to unilaterally incorporate it.
2:35:11
karlosz
drmeister: i think it would make sense to put it as the first optimization, right after converting to hir. that's the stage i applied the optimization to, and it should help speed up the other passes too
2:52:56
drmeister
We could postpone it until I get the kinematics working - but it helps to talk about it.
2:53:23
drmeister
When I first implemented it I made it way too complicated - I'm simplifying it now.
3:01:18
karlosz
Bike: you mentioned that generate asts is older. does that mean at some point it will be deprecated and a switch will be made to csts+ast?
8:25:12
scymtym
is (will) clasp (be) using the lambda list parsing facilities of the concrete-syntax-tree library? i'm asking because i noticed a few missing/suboptimal things when switching lambda list parsing to it in my toy
12:44:07
scymtym
as far as can tell, concrete-syntax-tree does not check for repeated parameter names, using constants as parameter names, etc. does clasp implement its own checks for these or how does this work?
13:24:27
drmeister
We don’t do any other checks. We recently switched to cst and haven’t run into problems
13:33:34
scymtym
i see. i noticed this because my toy accepted lambda lists like (a a pi) for which it should complain
13:34:26
scymtym
so i guess the question is whether to implement these checks (the environment-independent ones?) in the concrete-syntax-tree library
13:34:33
drmeister
We have just been compiling the clasp Common Lisp source code over and over again with cst. There probably aren't any instances of lambda-lists like that.
13:36:13
scymtym
yes. my focus is somewhat on invalid programs since i experiment with editor support
13:38:03
scymtym
i'm going to ask beach whether these things are in scope for the concrete-syntax-tree library
13:42:08
drmeister
And since we only ever write valid and correct programs - we didn't notice any problems.
13:55:29
drmeister
I'm shaking compiler bugs out of the kinematics code. I've ripped out my terrible idea of managing memory in vectors using reference counting and switched to Clasp's standard memory management.
14:07:07
Bike
i came up with a fix but i'm pretty weirded out, i should have hit this bug before and i thought i did
14:10:03
Bike
also i have an idea about precompiling clos stuff. i think it would reduce but not eliminate startup compilation, and we could move the delay when running the first compile to actual startup