github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt \o 10:18
nwc10 o/ 10:34
brrt quiet here the last few days 10:43
well, nine++ fixed bugs
lizmat I think a lot of people are waiting for a MoarVM release 11:51
seems samcv is too busy IRL :-(
samcv yeah, sorry about that. i have the flu now as well. after being really busy IRL i got sick 12:45
if anybody else wants to do up the changelog this month, I can do the minting 12:46
lizmat samcv: get well soon! 12:57
MasterDuke is this something moarvm could use? github.com/zwegner/faster-utf8-validator 21:08
jnthn MasterDuke: Hard to say, since we don't actually want to validate it, but rather transform it into an internal NFG representation, and "validator" sounds like something that would be applicable if you wanted to keep the same representation and just check it first. 21:41
I mean, we do want to validate it along the way, of course, but it's not our whole process.
There may well be things to learn, though I've spent hours tuning that lot already :) 21:42
But patches with callgrind measurements welcome! :)
"random ASCII text at .09 cycles/byte" => this is impressive, though 21:43
Especially since we can potentially do a scheme whereby we walk through the thing until we find non-ASCII
MasterDuke hm. this is pretty far from my area of expertise. if anybody else wants to take a crack at it please do so 21:49
Voldenet decode_utf8_byte and add_codepoint_to_buffer could really use some vectorization 21:50
fwiw "reading utf8 string" may not be the best abstraction when dealing with xml/json 21:58
.net core introduced Utf8JsonReader which consumes stream and produces a reader - not even using utf8 encoding inbetween makes it a lot faster 21:59
(reading is around 2x faster, writing is around 1.5x faster) 22:00
jnthn We don't do JSON down at the VM level anyway; a Raku module could certainly do it at the Buf level 22:11
Voldenet: If you know how to do the vectorization and have time/motivation to look at it, I'd be very glad to see that. 22:15
lizmat jnthn: fwiw, I tried to do a from-json based on Uni, but couldn't get it to go significantly faster :-( 22:51
afk&
jnthn lizmat: Probably would have more luck on buf/blob, since nine++ optimized those a bunch when doing the bytecode writing 23:19