Search
Friday, 29th of November 2019, 13:50:42 UTC
14:04:00
scymtym
stassats: i had that yesterday. i thought it was caused by profiler interrupts, but maybe it wasn't
14:04:18
stassats
yes, it was during profiling
14:05:28
stassats
trying to profile the pretty printer, which is really slow on nested forms
14:05:43
stassats
and there's zero comments on how the pretty printer works
14:05:55
scymtym
my idea was that maybe the signal handler should rebind the pprint circularity table, but i didn't investigate
14:06:11
stassats
is it printing anything?
14:12:21
stassats
does Profiler sample vector full count?
14:12:43
scymtym
also, pretty printing can take up ~ 20 % of the runtime of IR1-CONVERT-COMBINATION: https://techfak.de/~jmoringe/clim.flamegraph-ir1-convert-combination.png
14:12:57
stassats
yes, that's https://bugs.launchpad.net/sbcl/+bug/521153
14:16:47
scymtym
haha, the hash-table failure and the ensuing debugging session is also in the profile
14:16:56
scymtym
so the profiler probably didn't die
14:19:24
stassats`
i don't think i can muster the will to wade through the pretty printer code, so maybe sidestep the issue
14:19:49
stassats`
don't pretty print the forms, at compilation time, but only when needed
14:20:40
stassats`
but can't save just any lisp form, and reading back a string won't work
14:21:24
stassats`
wrap the unreadable stuff into a structure of some sort?
14:29:51
stassats`
(print-unreadable-readably *print-pprint-dispatch* ) => "#S(UNREADABLE :X #A((47) BASE-CHAR . \"#<SB-PRETTY:PPRINT-DISPATCH-TABLE {10000247C3}>\"))"
14:58:40
stassats`
how important is pretty printing for stepping anyway?
15:38:01
stassats`
i kinda see what makes pprint slow, iterating over a growing list
15:38:10
stassats`
but not yet clear what that can be replaced with
17:56:32
stassats
turns out, i just had to keep track of the length of a list
17:56:55
stassats
pretty printing is still slow, but a bit better on highly nested lists
18:38:20
phoe
I am trying to fix FORMAT.E.* ANSI-TESTs on SBCL
18:38:36
phoe
and porting the fixes from ECL still do not solve all the issues
18:39:19
phoe
rounding floats *does* bite one in the behind
Saturday, 30th of November 2019, 1:50:42 UTC