freenode/#sbcl - IRC Chatlog
Search
21:38:29
pfdietz
stassats: I don't have a good test case, but I think randomly generated forms with lots of PROGVs tend to make the compiler's space usage blow up. Turning off PROGV makes the random tester work better for large size form.
21:41:28
pfdietz
Not very new, but I don't remember this years ago. I tweaked the tester to do "swarm testing" at one point, which would have made it produce forms loaded with more of particular kinds of subforms. That could have made it start happening more.
21:42:45
pfdietz
swarm testing = on each iteration, select a random subset of the possible ways to generate random exprs, and only use those productions. So if PROGV is selected, it would then show up a lot on that iteration.
21:43:53
AeroNotix
stassats: it behaves kind of strangely. I'm guessing when the function is ran, it takes a lock.
21:48:28
AeroNotix
stassats: makes no difference, the code I initially triggered it from used :thread t as well
21:49:40
AeroNotix
stassats: that's what the original question was regarding, does unschedule-timer require that the scheduled function is not currently running
21:50:10
stassats
well, if it's in the same thread it's irrelevant, cause they can't run at the same time anyway
21:54:38
AeroNotix
Well, either way my question is answered: you can't cancel a timer if the function you scheduled is running
22:25:11
stassats
it actually looks like my ir1-phases thingy actually improves this (but not as dramatically as compared to 1.3.14)
22:32:50
stassats
(that was my motivation, not only fixing a bug but reducing redundant optimization passes)