These are chat archives for dropbox/pyston

15th
Sep 2015
Sun
@Daetalus
Sep 15 2015 09:43
@kmod You are working on #918?
Never mind. I still working on #914
Kevin Modzelewski
@kmod
Sep 15 2015 10:54
oh sorry, yeah I started looking into that myself
about to put a PR up and I think I'll move on after that
Sun
@Daetalus
Sep 15 2015 10:54
A question, please.
d = {1:2, 2:4}
s = set(d)
What function will the set(d) called? I put breakpoint inPySet_New. But it will not be called. setInit and setNew seems should not be called.
Kevin Modzelewski
@kmod
Sep 15 2015 11:00
why not?
Sun
@Daetalus
Sep 15 2015 11:05
setNew doesn't take the container argument. I put breakpoint in setInit, but could not comfirm it will called by set(d).( The pyston interpreter will called it many times). Maybe I did something wrong.
Kevin Modzelewski
@kmod
Sep 15 2015 11:06
I think those are supposed to get called
you can do PYTHONIOENCODING=utf8 gdb --args ./pyston_dbg -S test/tests/t.py to get it to not run any code other than your file
(or PYTHONIOENCODING=utf8 make dbg_t ARGS=-S)
Sun
@Daetalus
Sep 15 2015 11:07
Ok, trying. Thanks!
Marius Wachtler
@undingen
Sep 15 2015 12:31
I noticed that we we use a single edgecounter inside the interpreter to count the number of backjumps. Inside the LLVM tier we use different counters for every OSR exit point.
Is this because the multiple counter approach would be more accurate but is more difficult to implement inside the interpreter (if we don't want the embed the counter inside the jump node) and that's why they are different or is there some other reason?
Kevin Modzelewski
@kmod
Sep 15 2015 20:16
I don't think there's any particular reason... :P
Marius Wachtler
@undingen
Sep 15 2015 20:54
I just came home from the office a few minutes ago but unfortunately I could not get the cache in a state for reviewing :-(. But I will make sure that you can look at it tomorrow.
Marius Wachtler
@undingen
Sep 15 2015 21:37
I'm not a friend with the gitter android app...