Search
22:15:56
stassats
so the finalizer for run-program has (sb-unix:unix-close fd), but if the fd is recycled it closes the wrong thing
22:16:20
stassats
for the whole make-fd-stream, but run-program has started finalizing its streams recently
22:17:08
drmeister
Keep talking like I follow what you are saying. yitzi will be back later - I'll point him to this discussion.
22:17:22
drmeister
...and I'll try and follow.
22:17:25
stassats
well, it's a bug in sbcl, nothing really to talk about
22:17:29
drmeister
Which run-program are you talking about.
22:17:34
stassats
how has it ever worked? nohow
22:18:25
drmeister
Oh - we are at that stage of programmer grief? Great - now I'm on firmer ground.
22:19:18
stassats
i didn't write, so i'm not grieving
22:19:57
stassats
broken for twenty years, never actually used until last month?
22:21:01
stassats
and i don't see how to make it work
22:23:57
stassats
as a first step, i can make it just not work horribly, by not doing anything
22:29:28
drmeister
How long until you might have a fix? I'm going to need to leave soon. When I get back I can pull sbcl and try it again.
22:29:44
drmeister
Or do you have a reproducer now?
22:30:00
stassats
i have a non-fix, by making it no-op
22:30:19
stassats
don't really have a reproducer, just can see where it doesn't work
22:30:30
drmeister
Good man. Is it pushed?
22:31:25
drmeister
Hmm, git says "Already up to date"
22:31:41
stassats
it's not the main thing
22:31:42
drmeister
You are pushing here: https://github.com/sbcl/sbcl.git ?
22:32:40
stassats
at least it exposed a problem with bad character encodings too
22:34:50
stassats
not being able to refer to the dead object when finalizing is a problem here
22:35:27
stassats
don't know, make a closure that closes the fd? so bad
22:36:12
drmeister
It doesn't crash now.
22:37:16
drmeister
Yep - I ran it three times.
22:37:46
drmeister
Is this something that will be in the next release?
22:38:17
stassats
well, it won't be closing random stuff for sure
22:38:48
stassats
but for working properly—closing the right stuff—probably too
22:39:27
drmeister
There's no other way around this in the short term is there? There is no alternative to run-program?
22:40:13
drmeister
I ask because I have a self-imposed deadline to release cando installed by package managers and this messes with that big time.
22:40:13
stassats
well, it works just as well as before
22:40:35
drmeister
We might be able to build sbcl from source.
22:40:47
stassats
can't retroactively fix it, though
22:40:59
stassats
the time machine was autoclosed by the finalizer
22:41:49
drmeister
Understood - that's life.
22:44:49
stassats
or a cons instead of a closure
22:46:27
drmeister
Well, thank you very much for taking the time to look at that. It was driving us crazy.
22:46:59
stassats
it's our bug, so kinda had to
22:48:38
Colleen
yitzi: kpoeck said 1 hour, 12 minutes ago: I did the requested change to the ansi-tests
23:00:03
stassats
although release-fd-stream-resources is supposed to cancel the finalizer
23:03:56
stassats
canceling before closing would probably be more correct
23:06:14
yitzi
So why would not running with `--script` make it break less? Just coincidence?