02:48 ilbot3 joined 04:21 brokenchicken joined 06:07 geekosaur joined 06:45 TimToady joined 09:22 TimToady joined 11:09 domidumont joined 11:15 domidumont joined
jnthn I'm guessing wildly about phones here, but it may be that multiple slower cores is more power-efficient (or easier to do power-management on) than one really fast core. Perhaps ditto for heat. Both matter plenty for phones. 11:34
Geth arVM/master: 6 commits pushed by jnthn++ 11:45
arVM: cc8a133a28 | MasterDuke17++ | 5 files
Fix format strings that warn during compile

Change to the suggested types+modifiers.
arVM: ac79dfb2a6 | (Jonathan Worthington)++ | 5 files
Merge pull request #501 from MasterDuke17/fix_warnings_during_compile

Fix warnings during compile
timotimo a veritable deluge of mails directly after the release :) 11:51
jnthn :) 11:54
timotimo didn't read the PR description of inactivate-async-tasks until now 12:20
interesting
managing memory is difficult, yo 12:23
jnthn Indeed.
Anyway, with that merged we should see a lot less leaks in thing using Proc::Async and friends.
timotimo i mean ... virtual memory means that if the array grows unboundedly and we never touch the slots near the beginning, and make sure to null them out properly, they'll just be dedup'd :P 12:24
jnthn haha 13:15
That's horrible, but yes :P 13:16
We'll still fix it though, it's good for locality :)
timotimo aye 13:17
interesting. that "riptide" gc does its "draining" (aka DFS to set marking bits) in parallel without considering what thread an object belongs to, and they CAS the mark bits in parallel
(extra bonus: the words "parallel" line up in my terminal right there
)
dogbert17 o/ 14:37
deafening silence :) 14:38
m: class SQLString { }; my $stringy = Str.^find_method("Stringy"); my $handler = $stringy.wrap(method () { SQLString.new(:str(callsame)) }); say "foo".Stringy # should this loop indefinitely? 14:39
camelia rakudo-moar 8a6bfc: OUTPUT«(timeout)»
14:57 Ven joined
dogbert17 and another question, what does this message mean? MoarVM panic: Zeroed owner in item added to GC worklist 15:16
Geth arVM: MasterDuke17++ created pull request #510:
Change tabs to spaces		 15:25
samcv guys so our strings aren't always stored in one long string or they are? looking at how we find a needle's inside haystack 15:40
we just call MVM_string_substrings_equal_nocheck which does memcmp(a->body.storage.blob_8 + starta,..)
so it would seem as if they are continuous regions of memory? or is there something i'm not seeing
timotimo oh hey samcv :) 15:43
samcv hey
timotimo strings can be "ropes"
MasterDuke samcv: i don't know much about the internal representation, but if you're planning on working on string searching you might be interested in github.com/MasterDuke17/MoarVM/tre..._for_index
timotimo which is a list (can be a tree) of references to other strings including offset and length
looking forward to you commenting on my PR on UCD, samcv :) 15:45
samcv yeah about tho check it out now
timotimo but the next ~hour&half will be spent on the road and i should really rather not stare at screens, lest i projectile-vomit people to death
samcv hah
timotimo we're still waiting on one person, though 15:46
samcv i think switching to memmem instead of memcmp would bring us a pretty big improvement 15:48
also i'm not sure how we get different locations comparing how we are doing, in case it uses ropes or something
timotimo i wasn't aware of memmem, actually 15:52
substrings_equal_nocheck will fall back to grapheme iterators if one of the strings is a rope 15:53
i'd expect memcpm to be as fast as memmem, as memcmp can already abort when a single byte differs 15:54
oh 15:56
"C's memcmp is actually pretty slow because it tries to preserve useful string comparison semantics" - sounds like the writer of this SO question confused memcmp with strcmp?
oh wait a minute 15:58
substrings_equal_nocheck isn't about "where in this string is another string", it's about "is the n graphemes after offs equal to the first(?) n graphemes in the other" 15:59
but yeah, finding a string in another would be right for memmem
okay, time to drive 16:00
samcv see ya soon 16:03
timotimo re 17:28
i see you didn't comment on my cod e:S 17:29
samcv timotimo, will soon. have a question, is it able to parse multiple 0 values for one base 40 int? 18:34
like if we have 3 codepoints with no name in a row, that would be value 0.
so we can save a lot of space by not having to use three 0's and just have one
18:58 geekosaur joined
timotimo how often do codepoints not have names? 19:04
it doesn't currently handle that, but it totally can
iirc i keep the state around for the Decoder
so it keeps the queue, and there's the eos_signaled bit that could perhaps have something to do with this
samcv: ^
samcv uhm 19:05
timotimo i'm currently attending the Anime Season Kickoff Frenzy, in which many first episodes of anime shows are being shown
not really able to communicate a bunch 19:06
samcv ah kk. well looks like we're ignoring blank characters. but i think there are enough that we should just treat it as a stream of data to be decoded
shouldn't require too much change, but basically if we see a 0 in the highest place, then we still need to retain the smallest place which is a number, or if there's multiple 0's places then we need to assign to a diff cp 19:07
timotimo right. we already don't have "places" any more when we're looking through them 19:08
i.e. the queue already contains the up-to-40 numbers as individually addressable things in the queue 19:09
and we can pull individual ones out
samcv yeah
timotimo so when we get three 0s, it'll be fine to take a single one for each call to nom_a_string 19:10
of course it'[l have to go into the encoding part, too 19:13
maybe it'd
be interesting to adopt a queue-like approach to the code that encodes
samcv yeah. gotta adjust the encoding part
timotimo so we'll split the code that wants to emit up-to-40-numbers from the code that takes care to take triplets of it and makes 16bit numbers
it'll also handle setting up the very last bit 19:16
samcv i will make a function which we just push our strings to, and then we call at the end to get our hash table 19:17
err not hash table, but uh, array
timotimo that sounds doable 19:18
it'd be nice if we could generalize the code for generating compressed stuff so we could also use different tables to generate the sequence names 19:28
samcv ah
yeah
timotimo we can put the character table and the level one (and more) tablet behind pointers that live in the decoder in the C code 19:29
so we can just switch them out to decode other stuff, like sequence names 19:30
samcv unless $no-empty # yeah this is the variable that needs to be true to begin generating null names
yea
should also factor in the memory we lose by storing the string, and maybe those strings should be encoded? (the shift levels?) 19:31
timotimo could definitely do that 19:32
we might have to make the queue longer then, so that we can just decode a whole second-level string into it at any time 19:33
otherwise we'd have to store more state in the decoder to tell when we're in a second-level string and how long it'll go for
*shrug*
samcv yeah 19:34
timotimo but since the Decoder will live on the stack (since it doesn't have to be long-lived ever) it can be big 19:35
20:28 domidumont joined 21:41 zakharyas joined 22:10 BinGOs joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!