Geth MoarVM/main: 5 commits pushed by (Geoffrey Broadwell)++, (Will Coleda)++ 02:51
13:12 apogee_ntv joined 14:34 sugarbeet left 14:35 sugarbeet joined 15:10 MasterDuke joined
MasterDuke timo, et al.: we have some fast paths in MoarVM's string-to-numeric code that are only for `if (str->body.storage_type == MVM_STRING_GRAPHEME_ASCII) {`. those don't normally hit because most strings that aren't strands are MVM_STRING_GRAPHEME_8 or MVM_STRING_IN_SITU_8 15:14
tellable6 2026-11-02T20:00:43Z #raku-dev <timo> MasterDuke is the question about MVM_frame_find_lexical_by_name for $?CLASS still relevant? I see you merged a pull request that uses a lexical (that hopefully turns into a local) for $?CLASS in the regex code
MasterDuke i have forgotten what the difference in practice is between `*_ASCII` and `*_8` 15:16
lizmat IN_SITU_8 is for short strings only, using the header for storage ?
MasterDuke so in github.com/MoarVM/MoarVM/pull/2068 i take the fast path for all three
yeah, that describes the IN_SITU part, it's the 8 v ASCII i'm worried about 15:17
m: use nqp; say nqp::coerce_si("  123456789") # so in an example like this, or when building rakudo or running tests, it's never ASCII 15:19
camelia 0
lizmat timo might recall
MasterDuke also, what's up with camelia, that prints 123456789 (on both my branch and main) 15:21
locally that is
lizmat camelia runs about 2 months behind
MasterDuke committable6: HEAD use nqp; say nqp::coerce_si("  123456789") # so in an example like this, or when building rakudo or running tests, it's never ASCII 15:22
committable6 MasterDuke, ¦HEAD(95aa98b): «0␤»
lizmat I guess we need to find another place for camelia as nine is unresponsive :-(
MasterDuke committable6 is at HEAD though... 15:23
bisectable6: use nqp; say nqp::coerce_si("  123456789")
bisectable6 MasterDuke, Will bisect the whole range automagically because no endpoints were provided, hang tight
MasterDuke, Output on all releases: gist.github.com/44e9fd21115295b22f...3eb3f8d399 15:24
MasterDuke, Bisecting by exit code (old=2021.12 new=2022.02). Old exit code: 1
MasterDuke, bisect log: gist.github.com/0ccb7e1d6933a01409...d6ecc41e70
MasterDuke, (2022-01-27) github.com/rakudo/rakudo/commit/0c...1398b96b81
MasterDuke, Bisecting by output (old=2018.06 new=2018.08) because on both starting points the exit code is 1
MasterDuke ?!?!?!
bisectable6 MasterDuke, bisect log: gist.github.com/9c809b148aea4ad538...dc0c43ee7d 15:25
MasterDuke, (2018-07-09) github.com/rakudo/rakudo/commit/c2...356201cc21
MasterDuke, Output on all releases and bisected commits: gist.github.com/a471ddb3848549646d...810c55efb8
lizmat "a faster coerce_si" it saus in the commit
*says
MasterDuke what do you get locally? 15:26
15:28 MasterDuke8 joined 15:32 MasterDuke left
MasterDuke8 m: my int $i = +"  123456789"; say $i 15:33
camelia 123456789
lizmat same
MasterDuke8 weird 15:34
15:40 MasterDuke8 left 15:53 MasterDuke joined
MasterDuke nqp: say(nqp::coerce_si("  123456789")) 15:54
camelia
We trust you have received the usual lecture from the local System
Administrator. It usually boils down to these three things:

#1) Respect the privacy of others.
#2) Think before you type.
#3) With great power comes great responsibility…
MasterDuke why is it trying to sudo?!
lizmat no idea.. no idea 15:55
MasterDuke lizmat: btw, the rakudo build has gotten so slow. anything that can be done? this part by itself takes 2m: `/home/dan/r/install/bin/nqp-m --module-path=blib --ll-exception --target=mbc --output=blib/Perl6/BOOTSTRAP/v6c.moarvm --vmlibs=dynext/libperl6_ops_moar.so=Rakudo_ops_init gen/moar/BOOTSTRAP/v6c.nqp` 16:03
lizmat allow fewer LLM based commits in
MasterDuke same with tests and spectests
lizmat for make test: have fewer elaborate test files 16:04
(LLM based)
more seriously: since when did you notice this slowdown? 2026.09 ? 16:05
if so, than that's the RakuAST not being optimized as the legacy backend was
MasterDuke well, i *think* it was as soon as building with rakuast became the default.
lizmat yeah, so that'd be sometime between 2026.08 and 2026.09 16:06
RakuAST does a *lot* more when building QAST 16:07
and does extra work building the RakuAST tree
but yeah, I'm also frustrated by the slowdown 16:08
help with further optimizations would be appreciated 16:10
MasterDuke well, i can make micro-benchmarks faster... 16:12
the tricky part with rakuast, if nine is correct (any my memory of what he said is also correct), is that there isn't one or two parts that are slow. so looking at a profile and sorting by time isn't going to help. we have to find work being done that doesn't need to be. i.e., look at a profile and sort by number of calls to find something that's 16:14
being run more *times* than it needs to be
which means knowing how many times it *should* be
this is where some pointers from him or jnthn as to what parts might make sense to look at first would be helpful 16:16
lizmat I guess you will have to do with me in that respect, as jnthn and nine have moved on to other projects and/or are busy with their own startups 16:17
MasterDuke yeah, but i did see jnthn commented on a PR recently. maybe he could be persuaded to at least lay out some thoughts for where to start 16:18
lizmat thing is, a lot has happened in RakuAST after jnthn stopped contributing to it... I wonder whether he can provide insights without considerable effort from him 16:20
MasterDuke hm...true
maybe we need to do an effort to profile rakuast over time to see if that can point out anything 16:21
like benchable6 does, bisecting by time to run
build rakudo with rakuast at each release since that's been possible and see if there are any cliffs 16:22
lizmat that would be kewl.... also finer grained timing info would be great :-) 16:23
MasterDuke then bisect inbetween the release to find some individual commits or series of commits
timing info for what? the build process?
i used buildprof a couple times and it was very helpful for looking at where time went 16:24
lizmat build process indeed
MasterDuke i recommend trying buildprof 16:25
anyway, i gotta eat lunch and then afk. but i have some more coercing fast-paths i'm working on
16:30 MasterDuke left