Search
Monday, 23rd of November 2020, 1:43:27 UTC
3:38:38
pfdietz
stassats: which functions should be tested for the division-by-constant optimizations?
12:04:24
flip214
stassats: Well, I'd be happy with a simple byte-stream, too... but it seems that SBCL closes all fds before execve, is there an (official?) way to turn that off?
12:06:19
stassats
pfdietz: i'm changing truncate on signed integer / signed constant
12:06:27
stassats
but that's after the freeze
12:07:01
stassats
flip214: you're not passing them as streams?
12:08:47
flip214
well, there's only :input, :output, and :error-output... I'd like to pass in additional pipes or sockets
12:09:01
stassats
i guess could add an interface for which fds to not close
12:14:18
stassats
do you have a simple test case?
12:18:00
flip214
stassats: how about this? https://paste.debian.net/hidden/43e4c356/
12:18:45
stassats
osicat, cffi, uiop, not so simple
12:19:02
stassats
so will uiop take up the new interface? do other implementations have it?
12:20:16
flip214
wait, here's one with LOOP instead of iterate, and an ASSERT
12:20:17
flip214
https://paste.debian.net/hidden/0eec189c/
12:20:39
stassats
right, no to mention iterate!
12:27:47
flip214
stassats: here's a "plain" version: https://paste.debian.net/hidden/0629fc68/
12:28:46
flip214
ah, I forgot an UIOP there
12:29:00
stassats
i'll take the pipe part
12:31:58
flip214
stassats: https://pastebin.com/1nst9A08
12:32:28
stassats
i'm going by launching a new sbcl and writing to the pipe
12:32:33
flip214
pipe(2) shouldn't set the close-on-exec flag for the fds; but for testing I clear this flag for one of them.
12:32:56
stassats
we don't use close-on-exec
12:33:05
flip214
I think there's some circumstances where fds are set to close-on-exec by default
12:33:18
flip214
but I'm not sure, so I just put that in for testing
12:33:37
flip214
yeah, right... write to the passed-in fd, and if it fails, it didn't work ;)
12:41:14
stassats
ok, made a working test case by disabling all closing
12:41:26
stassats
will add an interface o selectively disabling later
Monday, 23rd of November 2020, 13:43:27 UTC