Diggsey has quit [Quit: Connection closed for inactivity]
infernix has quit [Ping timeout: 276 seconds]
infernix has joined #pypy
itamarst has quit [Quit: Connection closed for inactivity]
_whitelogger has joined #pypy
otisolsen70 has joined #pypy
jcea has joined #pypy
jcea has quit [Remote host closed the connection]
jcea has joined #pypy
otisolsen70 has quit [Quit: Leaving]
_vertig0 has joined #pypy
itamarst has joined #pypy
<mgorny>
cfbolz: sorry to bother you, but did you have any chance to look at the OrderedDict problem? it's biting us quite hard right now, because we either end up with pytest crashing or lots of other packages crashing on weakref issues
_vertig0 has quit [Remote host closed the connection]
<cfbolz>
mgorny: sorrrrrryyyy :-(. I started debugging it and reproduced it nicely and everything, but my attempts at finding a smaller reproducer have completely failed so far. I fear I just need to bite the bullet and take a proper day to get it done
<mgorny>
cfbolz: no problem. it's always useful to know that it's not just me :-)
<mgorny>
this thing is damn hard
<cfbolz>
yeah, the dict really ends up in an "impossible" state, and my attempts at reproducing the same state without needing all these libraries went nowhere
<cfbolz>
I think it might be time for rr
<mgorny>
hmm, maybe i can use a lesser hack in Gentoo, and instead of patching pypy -- just use pytest to block loading the rpython variant
<mgorny>
s/use/patch/
<LarstiQ>
would that be interesting for a livecoding session or too boring a slog? (Also, too distracting to cast?)
<cfbolz>
LarstiQ: I don't know, do you think that is interesting for anyone?
<LarstiQ>
I would be interested
<cfbolz>
like saturday evening CEST for example?
<LarstiQ>
it's an issue that stumped both you and mgorny. I've read some rr posts before so am aware of its theoretic usage but's that still a bit different than really seeing it in action. But well, might be just me
<LarstiQ>
cfbolz: should work! I'm in Finland so a little ahead
<mgorny>
LarstiQ: don't look at me, i'm stumped because it's rpython and i can't printf debug it xD
<LarstiQ>
mgorny: don't sell yourself short :)
<mgorny>
but seriously, i don't have any time to look at it
<mgorny>
it's an itch i can't reach
<cfbolz>
mgorny: that makes perfect sense, it seems neither fixable on the C level nor purely on the Python level, so really annoying
<cfbolz>
mgorny: about the workaround: indeed, making *just* pytest use the pure python version might be workable for a while?
<mgorny>
yeah, i'm going to test that in a minute
<mgorny>
now, how do i actually block loading it...
<mgorny>
sys.modules["_collections"] = None
<mgorny>
import collections
<mgorny>
del sys.modules["_collections"]
<mgorny>
ugly but works
<cfbolz>
not pretty, but sounds like a workaround that could unstuck gentoo?
<mgorny>
only if i manage to get it to work :-)
<mgorny>
okay, that's not going to work
<mgorny>
too hard to get it right, especially when it can be imported earlier
<mgorny>
and we must not override the one used by the actual package
<mgorny>
my next best idea would be to provide a separate import-name for the pure python version, and use that directly in pytes
<mgorny>
btw if i remove the move_to_end() bit from pytest, it doesn't crash anymore
<cfbolz>
yeah, move_to_end() is definitely involved
<mgorny>
hmm, let's try something dumb
<mgorny>
oh wait
<mgorny>
they're using move_to_end() to move to the beginning
<mgorny>
and i was investigating last=True codepaths :-)
<mgorny>
oh my, that logic is too compelx to eyeball debug it