Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
00:07 bloatable6 left, greppable6 left, bisectable6 left, sourceable6 left, benchable6 left, shareable6 left, quotable6 left, Techcable left, ab5tract left, mst left, gfldex left, jjatria left, ugexe left, SmokeMachine left, patrickb left, evalable6 left, sugarbeet left, jnthn left, japhb left, xiaomiao left, vrurg left, tbrowder left, camelia left, leedo left, tellable6 left, unicodable6 left, releasable6 left, linkable6 left, committable6 left, timo left, rakkable left, kjp left, Woodi left, ingy left, JRaspass left, Geth left, rba left, gfldex joined, jjatria joined, ugexe joined 00:08 rakkable joined, kjp joined, camelia joined, leedo joined, ingy joined, JRaspass joined, Woodi joined, timo joined, committable6 joined, linkable6 joined, releasable6 joined, unicodable6 joined, tellable6 joined, greppable6 joined, bloatable6 joined, bisectable6 joined, sourceable6 joined, benchable6 joined, shareable6 joined, quotable6 joined 00:09 SmokeMachine joined, Techcable joined, ab5tract joined, mst joined, japhb joined, xiaomiao joined, Geth joined, rba joined 00:11 vrurg joined, tbrowder joined 00:16 patrickb joined, evalable6 joined, sugarbeet joined, jnthn joined 03:49 japhb left 03:51 japhb joined 06:37 japhb left 06:41 japhb joined 06:59 [Coke] left, [Coke] joined 09:14 sena_kun joined 10:09 sena_kun left
lizmat and yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2025/03/03/2025-...ting-down/ 12:57
timo once again i am confused by the internal structure of NFAs 17:18
lizmat do we have docs on that ? 17:21
timo never mind, I was just Doing Pointers Wrong
lizmat heh... there was a time when that acronym was used a lot 17:22
timo i once had a commit that turned "charlist" edges with just a single character into a "codepoint" edge, that would also come in handy for what i'm currently looking at 17:28
right now i'm seeing if i can make a fast path for when the NFA would immediately return "no match" and compile that into the code 17:29
unclear if that's a win or not 17:31
"nfa engine startup" is not really an expensive "operation", and the fast check would always have to happen and can't really take work away from the real engine
the fast check can actually be a bunch simpler than a full nfa run, as it doesn't have to care about fates at all, and therefore the acceptable characters from all targets can be collapsed into a short list of deduplicated characters 17:42
for example, the NFA for "proto token term" gives me 259 lines of output for "interesting edges" for the first character; subtract 43 from that for synthetic edges and epsilons, then when i deduplicate all codepoint and codepoint longlit together i'm down to 66 unique codepoints, two character classes, and four charlists that i haven't looked at yet 17:51
the charlists expand to 9 characters 17:52
and they are all already there as codepoint edges, too 17:54
ok, this particular example has char classes 4 and 8 which are "alphabetic" and "numeric" respectively, which covers a quite wide variety of stuff 17:57
18:40 sena_kun joined 23:45 sena_kun left