Geth rakudo: AlexDaniel++ created pull request #1032:
Add ≤ ≥ ≠ unicode ops		 01:54
roast/≤≥≠: 91a781f207 | (Aleks-Daniel Jakimenko-Aleksejev)++ | S03-operators/misc.t
Tests for ≤ ≥ ≠ unicode ops
roast: AlexDaniel++ created pull request #249:
Tests for ≤ ≥ ≠ unicode ops		 01:55
naxieAlDle IOninja: hm, Geth reports the name inconsistently ↑ 01:56
TimToady back home in his own bed 06:45
unfortunately, furnace broke while we were away 06:48
fortunately, we live in California :)
lizmat Files=1178, Tests=55979, 196 wallclock secs (11.70 usr 4.65 sys + 1169.29 cusr 111.25 csys = 1296.89 CPU) 09:46
[Tux] This is Rakudo version 2017.02-167-ge114d5204 built on MoarVM version 2017.02-20-g773711e1 09:50
csv-ip5xs 2.875
test 12.538
test-t 4.995 - 5.154
csv-parser 13.214
MasterDuke IOninja: did you ever submit a PR for rt.perl.org/Ticket/Display.html?id=72820 ? 13:46
irclog.perlgeek.de/perl6/2016-07-06#i_12795893 suggests you did 13:47
IOninja Nope. The change interfered with interpolation of &foo(whatever) in strings
MasterDuke ok, mind if i mention so in the ticket (and link to here in the log)? 13:50
Geth rakudo/nom: d04c47fd63 | (Elizabeth Mattijsen)++ | src/core/metaops.pm
Make triangle reduce right op with 2+ params 1.5x as fast
MasterDuke so someone reading the ticket and log doesn't think it's already been fixed 13:51
lizmat m: sub infix:<f>(\a,\b,\c) is assoc<right> { a~b~c }; say [\f] ^5 13:52
camelia (4 234 01234)
lizmat ZofBot: hi
ZofBot_: hi
IOninja ?
lizmat there seem to be 2 ?
IOninja oh
MasterDuke m: my $a = class { has $.b }; say $a.new(b => 1).perl; 14:22
camelia <anon|65402160>.new(b => 1)
MasterDuke would `class { has $.a }.new(a => 1)` be a better output in that ^^^ case? 14:24
* s/a/b/g
moritz better, yes
but certainly not perfect, because it might reference outer lexical symbols 14:25
MasterDuke well, that's true for any .perl, right?
moritz m: my $x = do { class A { has $.x }; class :: is B { has $.y }.new(x => 1, y => 2) }; say $x.perl
camelia 5===SORRY!5=== Error while compiling <tmp>
'<anon|49526768>' cannot inherit from 'B' because it is unknown.
at <tmp>:1
moritz m: my $x = do { class A { has $.x }; class :: is A { has $.y }.new(x => 1, y => 2) }; say $x.perl 14:26
camelia <anon|69579408>.new(y => 2, x => 1)
MasterDuke m: my $a = class A { has $.b }; say $a.new(b => 1).perl;
camelia A.new(b => 1)
MasterDuke i was just comparing it to ^^^
moritz yes
and you're right, .perl typically reference a symbol which must be in scope to work
MasterDuke and fyi, i'm asking for rt.perl.org/Ticket/Display.html?id=77026 14:27
nine Given module A which requires module B. Should after "require A;" B be available through indirect lookup? 14:28
IOninja m: my $x = class {}; my @a = $x, $x; @a.perl.say
camelia [<anon|50504976>, <anon|50504976>]
IOninja MasterDuke: ^ in your version, those two will be different classes too
MasterDuke ah. could do something like `class <anon|50504976> { has $.a }.new(a => 1)`? 14:31
IOninja Don't really know how you're going to retrieve the guts either.
If you're just sticking public attributes into a `class {}`, it's really just as useless as <anon|432423432> thing we currently have 14:32
MasterDuke can always reject the ticket
anybody have an idea for what a better .perl would look like? fwiw, i'll at least link to this discussion in the ticket 14:41
Geth rakudo/nom: 9cec31afee | (Elizabeth Mattijsen)++ | src/core/metaops.pm
Make reduce chain op about 5% faster

Typical use case: 'say "numericall sorted" if [<] @a'		 15:22
IOninja nqp::chars got some sort of magic for strings fewer than 15 chars? 15:45
lizmat IOninja: not that I know, but that's generally out of my view :-) 15:49
timotimo that sounds unlikely. what symptoms are you experiencing? 16:03
IOninja Got a 60% perf improvement by avoiding using nqp::chars, but it kicks in only for strings with 15+ chars. I'm thinking maybe those strings can fit into a CPU register and that's why? 16:04
I tested with using 'a' as the char FWIW 16:05
timotimo how do you create the longer strings?
IOninja oh.... 'a' x 15
timotimo haha
IOninja :) 16:06
How to make it use proper string instead of the rope stuff?
timotimo you can use nqp::indexingoptimized to create a proper string from a rope
(be sure to use the return value)
m: "a" x 1000000000
camelia WARNINGS for <tmp>:
Useless use of "x" in expression "\"a\" x 1000000000" in sink context (line 1)
timotimo m: use nqp; nqp::indexingoptimized("a" x 1000000000)
camelia MoarVM panic: Memory allocation failed; could not allocate 4000000000 bytes
IOninja m: use nqp; say nqp::indexingoptimized("a" x 10) 16:07
camelia aaaaaaaaaa
IOninja m: use nqp; my @s = 'a' x 10; @s .= map: {nqp::indexingoptimized($_)}; say @s
camelia [aaaaaaaaaa]
IOninja same thing 16:08
timotimo what do you mean?
IOninja I got a smaller boost: 50% instead of 65% now, but it's still kicks in only for strings 15+ chars 16:09
ZOFVM: Files=1227, Tests=132953, 122 wallclock secs (21.86 usr 3.24 sys + 2412.20 cusr 247.86 csys = 2685.16 CPU)
timotimo interesting
but "a cpu register" isn't going to have much to do with it 16:10
IOninja hm
timotimo cache lines, however. there's something that could help you
but L1 cache ought to be big enough for 15+ chars? maybe?
Geth rakudo/nom: 22f43d20ef | (Zoffix Znet)++ | src/core/Str.pm
Make Str.Bool 30%–65% faster for strings 15+ chars long

Boost sunk results by ~65% and unsunk results by ~30% Improvement is 5-10 percentage points smaller for strings with flattened ropes (as opposed to one roped char).
Makes Tux's bench ~5% faster.
Unsure why no improvement is seen for strings smaller than 15 chars:
  irclog.perlgeek.de/perl6-dev/2017-...i_14208922		 16:13
IOninja This is my benching script FWIW: gist.github.com/zoffixznet/1b1d001...03c1357c48 16:15
and here's the result of the run: gist.github.com/zoffixznet/e60c029...b99ef0ec90
Before 15chars results flop around zero, but 15+ steadily in 30-65 percent 16:16
timotimo impressive difference
huh 16:17
how can the difference between 100, 1000, 100000 1000000 and 10000000 be so insignificant?
i.e. why does the time hardly increase? 16:18
also, what happens if you randomize the order of the tests?
IOninja m: say [<a b c d e f>].shuffle 16:19
camelia No such method 'shuffle' for invocant of type 'Array'
in block <unit> at <tmp> line 1
IOninja ...thought we did have it :/
timotimo nope, it's .pick(*)
IOninja Thanks.
timotimo m: .chars.say for '.pick(*)', '.shuffle()'
camelia 8
10
timotimo look it's even shorter
m: .chars.say for '.pick(*)', '.shuffle'
camelia 8
8
timotimo or the same length if you drop the parens 16:20
IOninja rendomization don't do anything: gist.github.com/zoffixznet/cb4b6db...5f2ba01617 16:22
timotimo that's fucking weird yo. 16:23
i'll look at it in more detail with a profiler or something, but for now i've gotta AFK
Geth rakudo/nom: bdd472bd8d | (Zoffix Znet)++ | src/core/IO/Spec/Unix.pm
Remove Perl-5-isms in string comparisons

The Str.Bool is also ~2x faster than doing Str.defined && Str.chars		 16:31
timotimo maybe that code was from back when "0" would boolify to false 16:32
IOninja Ah, right. 16:34
That was changed pretty close to Christmas, I recall
timotimo i think it was changed much earlier than that
IOninja Hm
Well, I was already around when it changed :) 16:35
c: all say "0".Bool
committable6 IOninja, gist.github.com/36dee8c548522b0275...3643c5f7c1
timotimo um
wordpress are you okay
IOninja 2015.04
timotimo it's doing a roulette right now
RIP my ram 16:36
i wanted to type "string bool" into the search field 16:37
and it started putting four different values into the search field over and over and over and over and over
also, it became slower the more often it did that
"putting in" as in "replaced the content with" 16:38
and it also updated the URL to go along with the search term
but what the flying fuck was that all about
how did it end up in an infinite loop like that?
IOninja You owe a dollar into the swear jar. 16:39
timotimo ho-hum. 16:42
Geth rakudo/lexical_require: 175151e0a8 | (Stefan Seifert)++ | src/Perl6/Actions.nqp
Create lexical stub package for nested required module.

require Foo::Bar; will now create a lexically scoped package "Foo" containing the nested stub package "Bar" to be replaced by the loaded module's package at runtime. With this, "require Foo::Bar; Foo::Bar::baz();" works again.		 16:46
MasterDuke IOninja: do you see numbers with unicode chars in the strings? 17:56
timotimo: 15 is just one off of 16. are small strings stored differently? inlined directly into a data structure? 17:58
Geth roast/≤≥≠: ce5e3baab5 | (Aleks-Daniel Jakimenko-Aleksejev)++ | S03-operators/misc.t
Tests for ≤ ≥ ≠ unicode ops
 18:05
roast/≤≥≠: 2b19c48da3 | (Aleks-Daniel Jakimenko-Aleksejev)++ | S03-operators/misc.t
Tests for ≤ ≥ ≠ unicode ops
 18:08
timotimo no, we don't have that kind of optimization yet 18:18
i wanted to build something like that for something in the past. it didn't work out well yet, but i forgot which datastructure it was for 18:19
MasterDuke thought i remembered you working on something, wasn't sure exactly what for or if it was merged 18:21
IOninja: * different numbers
lizmat hmmm... looks like IO::Socket::INET.get could use a lot of love 18:22
feels to me the first 3 lines should only be done once for each socket, instead of for every get 18:23
IOninja m: .bytes.say for 'a', '♥' 18:25
camelia No such method 'bytes' for invocant of type 'Str'
Did you mean '.encode($encoding).bytes'?
in block <unit> at <tmp> line 1
IOninja m: .encode./bytes.say for 'a', '♥'
camelia 5===SORRY!5=== Error while compiling <tmp>
Malformed postfix call
at <tmp>:1
------> 3.encode.7⏏5/bytes.say for 'a', '♥'
IOninja m: .encode.bytes.say for 'a', '♥'
camelia 1
3
IOninja MasterDuke: same numbers with ♥
m: .ords.say for 'a', '♥' 18:26
camelia (97)
(9829)
MasterDuke oh well, would have been interesting if the inflection point had been at less than 15 chars then 18:27
Geth roast/≤≥≠: b15122f1df | (Aleks-Daniel Jakimenko-Aleksejev)++ | S03-operators/misc.t
Tests for ≤ ≥ ≠ unicode ops
 18:39
rakudo/nom: d2c5684e50 | (Aleks-Daniel Jakimenko-Aleksejev)++ | 2 files
Add ≤ ≥ ≠ unicode ops

  🙌
Tiny IRC discussion:
  irclog.perlgeek.de/perl6/2017-03-01#i_14183585
  ⩵ (and especially ⩶) were left out intentionally. For now, let's focus
on operators that were actually requested many times.		 18:50
rakudo/nom: 5c68ea6828 | (Zoffix Znet)++ | 2 files
Merge pull request #1032 from AlexDaniel/≤≥≠

Add ≤ ≥ ≠ unicode ops
roast: c4ff5ce7d4 | (Aleks-Daniel Jakimenko-Aleksejev)++ | S03-operators/misc.t
Tests for ≤ ≥ ≠ unicode ops (#249)
nine lizmat: I concur 18:58
Geth rakudo/nom: e2c62cfaf4 | (Elizabeth Mattijsen)++ | 10 files
Change use of nqp::chars as bool to nqp::isne_s

This turned out to be a lot cheaper.		 18:59
rakudo/nom: 4bc826da3d | (Elizabeth Mattijsen)++ | src/core/IO/Handle.pm
Streamline IO::Handle.get

And also fix a few cases where nqp::chars were used as a boolean.
lizmat wonders what this all will bring for test-t
IOninja hmm 19:07
m: say 5.780667/5.710333
camelia 1.01231697
lizmat IOninja: is that good or bad ?
IOninja That's commit right before Str.Bool and second is HEAD. 19:08
lizmat so a small improvement
IOninja lizmat: when I measured against Str.Bool commit alone, it was 5% improvement. So if we assume there's no noise, the latest commits brought a 4% loss. Lemme build a few prior commits
the nqp::isne_s thing seems to be 0%-3% slower than nqp::chars for empty strings tho 19:09
jnthn I'm very confused by all of this, because nqp::chars is O(1) 19:10
lizmat jnthn: but if a str is a rope, it would need to walk the rope, no?
whereas nqp::iseq_s would stop immediately on the first different/missing char ?
jnthn lizmat: No, even if it's a ropey string we calculate the number of chars during the op 19:11
lizmat: Since it's always almost free anyway
IOninja jnthn: that's really weird, considering nqp::iseq_s('', $str) stars to be faster than nqp::chars() for 15+_char strings: gist.github.com/zoffixznet/e60c029...b99ef0ec90
jnthn Like in a concat it's just an addition
lizmat
.oO( only 10 more tests for 56K :-)
 19:12
IOninja m: say 2.93532595188/1.80599481741
camelia 1.62532357434
IOninja That's the diff for timings between 10000000-char string and 0-char string
jnthn IOninja: What's the test code?
IOninja jnthn: gist.github.com/zoffixznet/1b1d001...03c1357c48 19:13
timotimo oh, wait 19:14
we do actually store the size of a string in the string's header 19:15
why didn't i realize that obvious fact
we don't have to strlen at all
IOninja m: say 5.8725/5.7745 19:16
camelia 1.016971
lizmat IOninja: and that is ?
IOninja that was 5c68ea6828e131e1e2365d3a6d05ca4d2b55d184
IOninja builds Str.Bool one now...
robertle 1
IOninja The numbers are pretty noisy tho.
5.707, 5.703, 5.888, 5.800 <-- all the same commit 19:17
lizmat: errr wait
lizmat: that's that commit, but it's 1% SLOWER than before the Str.Bool (if we assume no noise issues) 19:18
lizmat ok, I will revert
IOninja So 5.8725 is the average of 4 runs on 5c68ea6828e1
lizmat: no wait
lizmat waits
:-)
IOninja m: say 5.934/5.82625 19:23
camelia 1.018494
IOninja So on Str.Bool commit, it now shows I made test.t 1% slower... I want to saw there's just too much variance to get a reliable result, but I don't know if that's just 'cause I want to believe my Str.Bool improvement affected it more than it really did
It was a 5% improvement when I ran the benching (3 runs each) before committing :/ 19:24
IOninja builds HEAD again
s/saw/say/
jnthn fwiw, here is the MVM_string_euqal impl: github.com/MoarVM/MoarVM/blob/mast...ops.c#L634 19:25
You'll line 5
*notice
Which actually is first checking if the strings have different length
Which is the very same thing that nqp::chars calls
So I'm very confused about the results :) 19:26
lizmat jnthn, yeah, ok, so reverting :-)
jnthn There's lies, darned lies, and benchmarks... :-) 19:27
IOninja :)
lizmat: would you revert mine too then? The Str.Bool one 19:28
lizmat IOninja: will do
lizmat spectest to be sure she didn't revert too much 19:33
Geth rakudo/nom: f11752cc76 | (Elizabeth Mattijsen)++ | 11 files
Revert all of the nqp::chars -> nqp::isne_s stuff

There does not appear to be a reason why this would make a positive difference. And there is some evidence it actually caused a negative difference.
This reverts: 4bc826da3d8ab3 e2c62cfaf451ef9b 22f43d20ef038ea7a23		 19:45
rakudo/nom: 9019a5b8ba | (Elizabeth Mattijsen)++ | src/core/IO/Handle.pm
Streamline IO::Handle.get/getc

Making IO::Handle.lines about 10% faster, and IO::Handle.getc about 3% faster.		 20:19

Powered by Cro and the Raku® Programming Language.

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

Thank you!