IRC channel logs


back to list of logs

***Server sets mode: +nt
***snape` is now known as snape
<wingo>good morning
<civodul>hello wingo!
<manumanumanu>good morning!
*wingo didn't get to jit over the weekend
<wingo>on this machine i see GUILE_JIT_COUNTER_THRESHOLD=50000 as about right, as a value that increases performance when JIT time is taken into account. if we can speed up the JIT we can lower it
<manumanumanu>wingo: I am going to try getting it up and running this week. Is there anything you want to have tested?
<wingo>fixed the optargs JIT issue. now all tests pass except those requiring hooks
<wingo>civodul: i think maybe for debugging purposes (code coverage, tracing, precise call counts in statprof, breakpoints, stepping) maybe we should just always fall back to the interpreter (i.e. fall back from jit to vm, never re-enter jit until debugging is done)
<wingo>maybe in 3.0 then the interpreter can always run hooks, dunno, maybe that's too much overhead
<wingo>would be nice to remove the --debug / --no-debug distinction!
<civodul>wingo: sounds good; i suppose it would be very hard to do all that when there's jitted code
<civodul>though it'd be nice to keep the "regular" engine by default
<civodul>(without hooks)
<wingo>it's certainly possible to do in jit code fwiw
<civodul>by regenerated JIT code with calls to the hooks?
<civodul>i'm just guessing it's more difficult
<civodul>also more difficult to inspect local variables, etc.
*civodul has to go
***lavaflow_ is now known as lavaflow