freenode/#clasp - IRC Chatlog
Search
15:14:40
Bike
o-k, i'm pretty much set up for fast method calls except i don't know how to codegen the actual call.
15:55:01
ecraven
I'm not an expert, but I think npm is more of an example of how to *not* do things than anything else... there's been another incident recently with deleted packages that were re-uploaded by someone else. there's no verification of the source, so people would just have pulled that
15:57:39
Shinmera
The whole web ecosystem is a mess that you can't trust a single bit of: https://hackernoon.com/im-harvesting-credit-card-numbers-and-passwords-from-your-site-here-s-how-9a8cb347c5b5
16:30:51
Bike
Just did a quick survey of how effective methods shake out in clasp. On startup there are 772 generic functions. After having used cleavir a bit and looking through all call histories: 799 (33%) emfs are slot readers, 197 (8%) are writers, 1089 (45%) are leaf methods, and the 320 (16%) remaining do something more complicated.
16:31:13
Bike
"leaf methods" meaning they don't call-next-method or anything so their method functions can be used directly as the emf.
16:36:58
Bike
of course, some frequently called generic functions are in the last category, like compile-ast
16:45:27
Shinmera
I'd estimate that in most libraries accessors will dominate both in definition and usage frequency
16:47:21
Bike
i can't measure actual calls without some slowing instrumentation, though it would still be interesting
16:47:48
Shinmera
Can't you get pretty low overhead instrumentation with linux tools like perf or what?
16:48:42
Bike
prrrrrobably, but i'm not immediately sure how to use that with the clos infrastructure
16:50:41
Bike
just thought i'd try to get a bit of evidence for the idea that most effective methods only involve one method, really