Search
23:18:04
selwyn
CST used to build in 1h 52m now it builds in 42 minutes (!) well done karlosz
23:18:47
stassats
if you apply the patch twice does it build in 20 minutes?
23:21:04
drmeister
ACTION wishes software worked that way
23:21:43
drmeister
stassats: did you see that hacker news raycast thing?
23:22:28
drmeister
https://news.ycombinator.com/item?id=19750507
23:22:51
Bike
ha, was that the one in #sbcl
23:23:05
stassats
if that's the one, then yes
23:23:21
stassats
as i said "if that's the quality of the lisp code what can be said about the whole benchmark?"
23:24:03
karlosz
thanks! that's good news then, because i have other patches that sped my build up more :)
23:24:05
stassats
but i guess it's too much to ask for well written lisp code, but in how many other languages are they not experts?
23:26:19
stassats
and it feels like the exercise in "make a wrong statement on the internet and wait for it to be corrected"
23:26:37
stassats
except for "publish poorly optimized code and wait for someone to prove you wrong"
23:40:33
drmeister
While building the latest 'dev' I hit this - selwyn you didn't see this?
23:40:46
drmeister
https://www.irccloud.com/pastebin/nLeknrLD/
23:41:06
drmeister
UNBOUND-VARIABLE . cleavir-cst-to-ast:client . ?
23:41:31
selwyn
yeah i got that earlier. i thought it went away, but it turns out that i actually built the latest clasp with AST - i just realised
23:42:08
drmeister
Ah - you didn't build the CST version then?
23:42:29
selwyn
no.. forgot to copy wscript.config ..
23:42:51
selwyn
sorry team. i'm building again but i will probably hit the same thing
23:43:21
drmeister
No problem - there are a lot of things to get right.
23:43:55
selwyn
could it be due to not having distclean'd? it's what i assumed when i saw it and indeed i hadn't
23:43:57
Bike
well none of my commits today seem to involve that variable
23:44:04
drmeister
Good 'ol safe-backtrace...
23:44:06
drmeister
https://www.irccloud.com/pastebin/2i8funUA/
23:44:25
Bike
BAD-TAGGED-POINTER, that bodes poorly
23:44:32
karlosz
:( but more than 2x speedup is *not* expected
23:45:48
Bike
oh hey, it's actually beach's mistake
23:45:56
Bike
https://github.com/robert-strandh/SICL/blob/87eb4fead5c65eaa5b8954ca2f7fa47cf5dd08e8/Code/Cleavir/CST-to-AST/convert.lisp#L17
23:46:50
Bike
i think that should be system, not client
23:46:56
selwyn
karlosz: yes i should have been more sceptical :s
23:47:33
drmeister
But the bad-tagged-pointer?
23:47:47
drmeister
here's a complete backtrace
23:48:04
drmeister
https://gist.github.com/drmeister/b5517aa128cef15aeaa42d9e9293f4d1
23:48:08
Bike
yeah i don't know what that shit is, but the error seems straightforward
23:48:21
Bike
it matches the backtrace too, that method is the last function before the error basically
23:48:34
drmeister
But yeah - it might be safe backtrace screwing up printing and thus harmless.
23:49:02
Bike
_CONVERT^CLEAVIR-CST-TO-AST^((T T T))^METHOD^^
23:49:55
Bike
i could commit a fix for sicl from here, but i'm not sure i want to do rebase stuff
23:50:02
drmeister
This is in cleavir - so you'll have to fix that right?
23:51:05
karlosz
this is why i stuck with cherry picking the patches i made - even loading sicl in clisp broke a couple of things recently
23:51:32
karlosz
so theres definitely some cleavir bugs lurking about
23:59:23
Bike
well i was going to make more cleavir changes tomorrow anyway, so i can update it with that
23:59:41
Bike
related, if anyone knows what name i should use for the multiple-value-setq-prog1 thing, now's the time
23:59:52
drmeister
I made the change by hand and it's building again.
0:15:07
drmeister
After changing client to system it gets past inline.lisp
0:26:41
drmeister
The build is proceeding well
2:16:13
drmeister
Here is the flame graph
2:17:22
drmeister
https://usercontent.irccloud-cdn.com/file/PJVIqJHy/out-9149.svg
2:19:57
drmeister
The ASDF compilation appears to hang however.
2:20:56
Bike
this is a stupid question but what are all the TOP-COMPILE-FILE functions exactly
2:23:36
drmeister
I don't recall - something with top level forms.
2:23:55
Bike
i'm just wondering why they show up everywhere
2:24:00
Bike
seems like we're using a ton of stack space
2:24:04
karlosz
weird - ASDF has never hung for me, but i havent updated sicl
2:28:53
karlosz
flamegraph looks good. there's no more r-u-i visible on top of do-inlining now
2:29:53
karlosz
and reinitialize-data on top of the inlining function bar can go away, and so can set predecessors next to inlining-function
2:31:05
karlosz
patch for getting rid of both forthcoming after this build finishes, but its at 90/445 and hasn't died yet
2:31:54
drmeister
My laptop ran out of power in the middle of ASDF build - and it hung up - so I'm starting it again.
2:37:13
drmeister
ASDF built fine after restarting it.
4:33:54
beach
Good morning everyone!
11:01:04
drmeister
Yesterday when I built CST it took about 150min