SmokeMachine does multi dispatch have performance problem? 03:00
I am trying to fix the hash multidimensional slice, and im abusing of it... 03:01
and almost not using nqp... :(
samcv SmokeMachine, shouldn't have any performance change if it can determine the type on compiling. but if it's dynamic there can be some impact yes 03:04
SmokeMachine samcv: thanks! 03:05
samcv also depends how many you have. removing some can make a fair impact or restructuring so that you turn two multi's into one that has an if condition can also be more performant 03:06
usually that is not needed but in some cases that may be good to try if you think that may be causing an issue
SmokeMachine I was abusing of it... Ill change it... (I was testing outside of the code... I added it to the right file and didn't work... :( ) 03:08
samcv: that is what I was trying: www.irccloud.com/pastebin/fYK6sZos/ 03:10
samcv what error do you get SmokeMachine 03:12
it runs for me. but i'm not sure what the output should be or not 03:13
SmokeMachine compile error... P6opaque: must compose before allocating
samcv compiles for me 03:14
SmokeMachine samcv: it gives me error when I port it to src/core/hash_slice.pm
samcv can you paste the full text of the error?
everything 03:15
SmokeMachine sure, 1 sec
The last version was returning Seq... this returns List (as expected) www.irccloud.com/pastebin/CGDZ7gJw/ 03:16
samcv: the error: www.irccloud.com/pastebin/54Bx42Sp/ 03:17
samcv SmokeMachine, show me a diff of your changes 03:19
SmokeMachine samcv: the diff www.irccloud.com/pastebin/Ntrrr2GH/ 03:20
samcv SmokeMachine, i'm guessing something is not containerizing somewhere. or something 03:33
try commenting multi's out and see when it starts compiling again 03:34
SmokeMachine samcv: when I comment everything but the"multi recurse-at-key(\SELF, \indices, 0)" it gives error yes... 03:35
samcv: I think the problem is the WhateverStar! 03:37
samcv: I changed to a block and compiled! Ill uncomment everything and test again! 03:38
:(
another *... 03:39
compiled! 03:40
worked! :) 03:41
the spec tests passed... 03:49
samcv good :) SmokeMachine 03:58
yeah commenting things out until it compiles can be helpful because the messages it gives compiling rakudo are not the best
SmokeMachine samcv: github.com/FCO/rakudo/commit/39ef4...1da63c512a 03:59
samcv: yes, it helps...
samcv: thanks for the help! 04:00
samcv also yo should make sure to use `multi sub` for the postcircumfix
you don't have to for the subs inside the operator routine but you should call all multi's `multi sub` 04:01
SmokeMachine samcv: it is... wasnt on my test to be sure that would use the one i was testing...
samcv how is speed of the new function compared to the new one?
SmokeMachine ah! ok!
samcv kk np
SmokeMachine samcv: the old one wasnt working for cases as %hash{1,1; 2} 04:02
samcv not at all? or was it returning the wrong type of object?
SmokeMachine it should return (True, True) and was returning False 04:03
samcv that is no good 04:05
good work though 04:06
SmokeMachine but the fix is much slower... :( 04:08
samcv how much slower? 2x slower? 04:09
SmokeMachine much more... 04:10
samcv okay. then take out this multi multi recurse-at-key(\SELF, [$idx, +@indices where { .elems > 0 }]) {
and turn it into an if else statement
SmokeMachine samcv: www.irccloud.com/pastebin/T4piLZQX/ 04:11
samcv SmokeMachine, i would take out the inside multi's and write it into if else statements 04:12
SmokeMachine samcv: yes, Ill do that...
samcv: but ill do that tomorrow...
samcv kk. sounds good
SmokeMachine samcv: thanks for your help! 04:13
samcv anytime
SmokeMachine have a good *
samcv good * :)
lizmat Files=1181, Tests=55992, 197 wallclock secs (11.76 usr 4.57 sys + 1170.16 cusr 112.54 csys = 1299.03 CPU) 07:22
Zoffix I had RAKUDO_MAX_THREADS=100 on my VM 10:33
Zoffix tries a normal-sized run
timotimo i wonder if we already have the debugname set before that "need to compose before allocating" message happens. i'm expecting the debugname is set when we compose, though. 10:46
Zoffix ZOFVM: Files=1230, Tests=132967, 117 wallclock secs (22.42 usr 3.06 sys + 2441.50 cusr 122.97 csys = 2589.95 CPU) 10:47
that's a 5harness stresstest
Zoffix waits for 6
Unhandled exception in code scheduled on thread 7 10:48
Cannot look up attributes in a TAP::Reporter::Console::Session type object
in method summary at /home/cpan/CPANPRC/rakudo/lib/TAP.pm6 (TAP) line 508
lizmat ouch. been a while since I saw that one
timotimo wow, that's amazing 10:50
Zoffix why? 10:51
timotimo others have been trying to reproduce that error for multiple days
Zoffix ah
timotimo or am i thinking of something else?
cool, the debugname is set in new_type, which is separate from compose 10:53
so the error message about "compose before allocate" could get the debug name printed out, too
timotimo hacks
Zoffix ZOF6VM: Files=1216, Tests=132907, 241 wallclock secs 10:54
samcv: ^ you wanted to see the difference between 5 and 6 harnesses. Notice how along with being twice as slow, the 6 harness reports 60 fewer tests were run
And 4 fewer files :S 10:55
m: say 1230/30
camelia 41
Zoffix Maybe this line *is* buggy: github.com/rakudo/rakudo/blob/nom/...rness6#L60 10:57
m: dd <a> Z <a b>
camelia (("a", "a"),).Seq
Zoffix dunno if it'd always have the right size there or not.
timotimo i can imagine it'd fail when @tfiles / @slow isn't a whole number? 11:01
would probably be easy to check, no? 11:03
fantastic 11:10
the code SmokeMachine wrote gave "P6opaque: must compose before allocating"
with my diff it also mentions that it's about WhateverCode
MasterDuke nice 11:13
timotimo pushed 11:15
dogbert17_ anyone want ta continue the talk about the test harnesses? 12:23
the 'spectest.data' file contains 1253 rows. if I run a spectest which of these should be igonered? Only the ones marked 'stress? 12:25
if we ignore headers an S01 tests the number of files is 1253 - 36 = 1217 12:27
ok, so I have found the 8 files which differ in a spectest between harness5 (1168 files testes) and harness6 (1160 files tested) 12:43
MasterDuke always the same files?
Zoffix :S 12:44
My stresstest run only had 4 missing files
dogbert17_ these differ for me
t/spec/integration/role-composition-vs-attribute.rakudo.moar, t/spec/integration/rule-in-class-Str.t
t/spec/integration/say-crash.t, t/spec/integration/substr-after-match-in-gather-in-for.t
t/spec/integration/topic_in_double_loop.t, t/spec/integration/variables-in-do.t
t/spec/rosettacode/greatest_element_of_a_list.t, t/spec/rosettacode/sierpinski_triangle.t
Zoffix "differ"? Differ how? 12:46
dogbert17_ they are not run by harness6 when doing a spectest 12:47
I took the output from runs by harness5 and harness6, sorted it and ran diff 12:48
Zoffix does harness 6 run the file when you individually run it? HARNESS_TYPE=6 make t/spec/rosettacode/greatest_element_of_a_list.t 12:50
dogbert17_ yes
Zoffix Then the bug is likely in t/harness6 12:51
The part I linked to earlier is especially suspect
dogbert17_ your cmdline didn't work properly, see below
dogbert@dogbert-VirtualBox ~/repos/rakudo $ HARNESS_TYPE=6 make t/spec/rosettacode/greatest_element_of_a_list.t 12:52
/usr/bin/perl t/harness5 --fudge --moar --keep-exit-code --verbosity=1 t/spec/rosettacode/greatest_element_of_a_list.t
it used harness5 instead
I believe the makefile is hardcoded to use harness5 when you want to test a single file 12:53
Zoffix oh, well, try using harness 6
`make -n spectest` to get the command and then change the files to just 1 file 12:54
dogbert17_ then it runs the file, btw I used './perl6-m -Ilib t/harness6 --fudge --tests-from-file=t/spectest.data2' spectest.data2 contains the test file in question 12:57
dogbert17_ tests a vague theory, i.e. does the number of tested files change if the TEST_JOBS param is changed 13:00
Geth rakudo/nom: 4227b3824d | (Zoffix Znet)++ | docs/2017-IO-Grant--Action-Plan.md
Log changes to IO Plan at the top

Remove `other issues` from bottom and log changes due to discussions at the top so those who reviewed the plan already can easily see the new changes.
MasterDuke here's where it's hardcoded to HARNESS5 when making a single file: github.com/rakudo/rakudo/blob/nom/...re.pl#L324 13:01
Zoffix dogbert17_: just see if @tfiles here matches @files: github.com/rakudo/rakudo/blob/nom/...rness6#L60 13:02
well, the number of them
dogbert17_ ok, will check 13:04
well, @tfiles is 1160 13:06
Zoffix and @files?
ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­
ā­ā­ā­
ā­ā­ā­ Reminder: the deadline to submit your suggestions and corrections to
ā­ā­ā­ IO Action Plan is in 3 days, on April 1st. You can find the Plan at
ā­ā­ā­ github.com/rakudo/rakudo/blob/nom/...on-Plan.md
ā­ā­ā­ 13:07
ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­šŸŒ ā­ā­ā­šŸŒŸā­ā­ā­
dogbert17_ @files is 1139 ?
Zoffix ā­ā­ā­ If you already did, you can see the 2 new modifications and 2 changes to old modifications that have been proposed in the first section of the file: github.com/rakudo/rakudo/blob/nom/...ction-plan 13:08
Geth rakudo/nom: dc8124f1cd | (Zoffix Znet)++ | docs/2017-IO-Grant--Action-Plan.md
List .extension(:parts) is proposed to take a Range

  - This was originally suggested as possible future expansion,
but feels useful enough to include right now.		 13:10
dogbert17_ Zoffix: so why is tfiles 1160, shouldn't it be 1168? 13:35
interesting, if I omit the TEST_JOBS directive @tfiles suddenly becomes 1168 13:37
Zoffix The then bug is where I pointed to, in the Z 13:41
because it doesn't get run when TEST_JOBS is 1
Use the debug prints, Luke!
dogbert17_ that's so old school :) 13:42
brrt and super effective 13:46
dogbert17_ if TEST_JOBS > 1 then @slow.elems = 29 and @files.elems = 1139 while @tfiles.elems ends up being 1160 (that's eight short) 13:48
Zoffix m: say 29+1139 13:51
camelia 1168
Zoffix m: say 1139/29
camelia 39.275862
Zoffix m: say 1139 div 29
camelia 39
Zoffix m: say 39 29
camelia 5===SORRY!5=== Error while compiling <tmp>
Two terms in a row
at <tmp>:1
------> 3say 397ā5 29
expecting any of:
infix
infix stopper
postfix
statement end
statement modifier
ā€¦
Zoffix m: say 39 * 29
camelia 1131
Zoffix m: say (39 * 29) + 29
camelia 1160
Zoffix Yup. Z is the cause
Zoffix is pleased to have had guessed this to be the problem area a few weeks ago :) 13:52
dogbert17_ .oO (the Z(offix) operator)
Zoffix++, now all we need is to figure out a fix
Zoffix Godspeed
dogbert17_ will look when he manages to get home from $work 13:55
Zoffix m: dd <a> Z <b c> 13:57
camelia (("a", "b"),).Seq
Zoffix m: dd roundrobin <a>, <b c>
camelia (("a", "b"), ("c",)).Seq
Zoffix dogbert17_: ^ a hint :)
"c" is the missing files.
dogbert17_ Zoffix: interesting :) 13:59
[Coke] m: say "\c[IDEOGRAPHIC TELEGRAPH SYMBOL FOR JANUARY]" 14:01
camelia 拀
Zoffix m: gist.github.com/zoffixznet/a79f282...5a7897124d 14:30
camelia Wide (Wide)
No such method 'is-prime' for invocant of type 'Wide'
in sub foo at <tmp> line 8
in sub foo at <tmp> line 7
in block <unit> at <tmp> line 11
Zoffix Why is it even trying to call `.is-prime` on Wide? I gave it 41 to `nextwith`...
m: multi foo (Str) { nextwith 13 }; multi foo (Any $ where .is-prime) {say "ohai"}; foo "x" 14:32
camelia No such method 'is-prime' for invocant of type 'Str'
in sub foo at <tmp> line 1
in sub foo at <tmp> line 1
in block <unit> at <tmp> line 1
Zoffix Bug? 14:33
m: multi foo (Str) { nextwith 13 }; multi foo (Any $x where {dd $_; True}) {dd [ $x, $x.^name ]; say "ohai"}; foo "x" 14:34
camelia Str $topic = "x"
Int $topic = 13
[13, "Int"]
ohai
Zoffix Looks like the arg the `where` is trying is pre-loaded when dispatched is figured out, but when it actually runs the value is different :o 14:35
s/it actually runs/candidate actually runs/;
Oh, I just misread >_< 14:38
Zoffix needs moar coffee
SmokeMachine samcv: my last try yesterday on this computer that Im using now took 0.3894857, the original 0.02741633. removing all the multis and fancy sub signatures its taking 0.1327739 14:49
samcv: this is my test, if you are interested... :) www.irccloud.com/pastebin/aPL55Xzq/ 14:52
Zoffix Interesting. I initially assumed ALL candidates are sorted from narrowest to widest and THEN they're tried in order until first match is found. But my experemintation shows first ALL candidates are tried to see if they match and then they're sorted.... Kinda makes me wonder, wouldn't the first version be way more performant, since you only try enough candidates to find a match and not all? 14:53
Zoffix is probably talking shit, but I can see the readers of my article wondering the same
Oh nevermind. My experiment was wrong. 14:56
or is it... 14:57
m: multi foo (IntStr) { say "IntStr"; }; multi foo (Any $ where { say "Tried to match against $_"; .is-prime}) { say "Any" }; foo <43>;
camelia IntStr
Zoffix So it didn't try the Any... cool
m: multi foo (IntStr) { say "IntStr"; nextwith 13 }; multi foo (Any $ where { say "Tried to match against $_"; .is-prime}) { say "Any" }; foo <43>;
camelia IntStr
Tried to match against 43
Tried to match against 13
Any
Zoffix But now it did... twice 14:58
dogbert17_ - @tfiles = (@slow Z batch(@tfiles / @slow, @tfiles)).flatmap({ .map(|*) })
+ @tfiles = (roundrobin @slow, batch(@tfiles / @slow, @tfiles)).flatmap({ .map(|*) })
Zoffix ^^
Zoffix Is that .flatmap({ .map(|*) }) really needed? What about just .flat
dogbert17_ seems to work with .flat 15:00
Zoffix OK. I think I get it: it sorts candidates from narrowest to widest, then goes to find the first matching candidate. If you use `nextwith` and ilk, it will continue trying to find the next matching candidate *using original value* and then when it finds it, it'll try to bind the new value that was given to `nextwith` which makes `where` blocks run twice 15:01
I don't really get why it doesn't just continue trying with new value alone....
dogbert17_: "seems" is not very scientific. You can `dd` the result and see that it flattened right 15:02
dogbert17_ that will be done before a PR is submitted, running a spectest right now in order to se if anything explodes 15:04
Zoffix I really should read the historic docs of how our multi dispatch works. Even our docs just link to glossary instead of explanation :/ 15:07
dogbert17_ Files=1168, Tests=55436, 411 wallclock secs 15:09
Zoffix \o/
dogbert17_ Files=1168, Tests=55436, 340 wallclock secs ( 6.60 usr 4.22 sys + 903.76 cusr 64.94 csys = 979.52 CPU) # harness5 15:10
Zoffix multi foo { samewith 'blah' }; is the same as multi foo { foo 'blah' } is it not? 16:00
ugexe I think it depends on where the `proto` is? 16:10
Zoffix Like how? The where the proto is affects `foo` same as `samewith`, I'd think, on? 16:12
*no?
s: &samewith 16:13
SourceBaby Zoffix, Sauce is at github.com/rakudo/rakudo/blob/dc81...ol.pm#L136
Zoffix looks like it just calls foo() to me... 16:15
ugexe I guess my comment just affects dispatch in general, so just calling foo() is the same thing I suppose 16:24
could have swore i've gotten into spots where I was telling myself use `samewith(...)` and if it gets stuck in an infinite dispatch loop use `foo(...)` 16:26
SmokeMachine Any suggestion to make it faster? github.com/FCO/rakudo/commit/c5e8f...59aa3663fa 17:03
dogbert17 Zoffix: can you make sense of RT #131073? 17:06
synopsebot6 Link: rt.perl.org/rt3/Public/Bug/Display...?id=131073
Zoffix dogbert17: yes, it said "It's a spam email, possibly containing an infected attachment" 17:08
dogbert17 Zoffix: I see it was swiftly removed :) 17:10
Zoffix m: multi foo(Int) { .say for (nextcallee() xx *)[^3]; }; multi foo(Cool) {}; multi foo(Any) {}; foo 42; 17:14
camelia Nil
Nil
Nil
Zoffix m: multi foo(Int) { .say for eager (nextcallee() xx *)[^3]; }; multi foo(Cool) {}; multi foo(Any) {}; foo 42;
camelia Nil
Nil
Nil
Zoffix buggy xx still?
[Coke] note: clicking the big "S" is the "best" way to remove spam from the RT queue. (the one in the menu near actions)
Zoffix That's the one I used.
[Coke] I suspect it helps document what is spam versus what is a bad request.
Zoffix m: multi foo(Int) { nextcallee.say for ^3; }; multi foo(Cool) {}; multi foo(Any) {}; foo 42;
camelia sub foo (Cool $) { #`(Sub|52330576) ... }
sub foo ($) { #`(Sub|52330728) ... }
Nil
[Coke] Zoffix++ # I was expected to see it show up as red for me after that, which it didn't. 17:15
Zoffix dammit... I really wanted to show off the coolness of xx :(
[Coke] *expecting
Zoffix m: .say for ("hi".say xx 3)[^3]
camelia hi
hi
hi
True
True
True
Zoffix m: .say for ("hi".say xx *)[^3]
camelia hi
hi
hi
True
True
True
Zoffix :S
m: .say for (say("hi") xx *)[^3] 17:16
camelia hi
hi
hi
True
True
True
Zoffix Oh gee, it thunks nextcallee and calls it elsewhere >_<
fair 'nuf
dd (nextcallee() for ^*)[^3]; has the same problem :/ 17:20
Geth rakudo: dogbert17++ created pull request #1050:
Fix for RT #131075		 17:21
synopsebot6 Link: rt.perl.org/rt3/Public/Bug/Display...?id=131075
Geth rakudo/nom: 876637010e | (Jan-Olof Hendig)++ | t/harness6
Fix for RT #131075

It was noted by Zoffix++ that the number of tests run by harness6 differed from the number run by harness5 when spec- or stresstesting. The problem turned out to be the piece of code combining the normal test files in spectest.data with the ones marked as being 'slow'. That code used the Z operator on lists that were not the same size, changing to roundrobin instead solves the problem. Zoffix++ for help.		 17:23
synopsebot6 Link: rt.perl.org/rt3/Public/Bug/Display...?id=131075
rakudo/nom: b108c85105 | (Zoffix Znet)++ | t/harness6
Merge pull request #1050 from dogbert17/fix-for-rt-131075

Fix for RT #131075
synopsebot6 Link: rt.perl.org/rt3/Public/Bug/Display...?id=131075
dogbert17 oO 17:24
Zoffix holy crap, nextcallee is finicky 17:31
loop { dd nextcallee } gives Any instead of my callee, presumable cause it looks for that block instead of my routine >_< 17:32
TimToady huh, when did nextcallee officially get into the language?
[Coke] the language or the compiler? 17:34
Zoffix TimToady: Before Christmas apparently, tho looks it might've been accidental: github.com/perl6/roast/commit/d88e...0526b489ca
That commit is the only instance of it in roast; though I see it mentioned in the docs too. 17:35
TimToady: I take it it shouldn't be?
TimToady looks like it's just in there to test wrapping 17:36
take a reference to the innards of a dispatcher seems...fraught...
[Coke] github.com/rakudo/rakudo/commit/b6...bc13b125cc is the commit that put it in rakudo
TimToady okay, I vaguely remember jnthn++ asking about calling something repeatedly 17:37
[Coke] rt.perl.org/Ticket/Display.html?id=125783 is the ticket
TimToady but that's for .wrap, dunno why you'd expect it to work in a loop 17:38
it might work in a multi, or in a method dispatch
Zoffix Well, I wanted to loop through candidates and find one on which `callwith` wouldn't fail to bind with 17:39
And the "loop through" part puts nextcallee into a loop and makes it give wrong results :) 17:40
TimToady nextcallee doesn't do that
nextcallee will always give you the same next one
unless you actually call down
it's not for iterating the candidate list
Zoffix Hm, then there's a bug. Because it shifts the candidates
m: gist.github.com/zoffixznet/80955d5...8e0c8c7d3a 17:41
camelia Middle (Middle)
Non-Prime 42
We're back! The return value is from NotPrime
TimToady hmm, well, maybe it could be used that way then
Zoffix Without `nextcallee` call, `callwith` ^ would've called the `Wider` candidate and failbind
Now... if only I could stick it in a loop šŸ¤”šŸ¤”šŸ¤” 17:42
s: &nextcallee 17:43
TimToady try something non-closurey, like a hyper?
SourceBaby Zoffix, Sauce is at github.com/rakudo/rakudo/blob/dc81...ol.pm#L131
TimToady well, but what to hyper it with, since it's a function...
Zoffix hehe 17:44
TimToady obviously you just need a goto 17:45
Zoffix :D 17:46
TimToady anyway, hyper would have its own dispatcher, obviously, which would interfere 17:48
so I suspect you can't do this in Perl 6 without dropping into nqp
dispatchers are, by necessity, considered very low level 17:49
well, or you could implement goto :D
or we could implement an opcode that would give you the nth dispatcher going outward 17:50
Zoffix I'm gonna take the easy way out... Just not mention anything about looping and nextcallee in my article :) 17:52
TimToady the intent was obviously to only deal with the next candidate, not the one after that :) 17:53
Zoffix seems like it :)
TimToady I suppose we could add a loop inside nextcallee, with an optional :nth() arg, since nextcallee itself is perhaps not in anyone's hotpath, but that wouldn't help your article 17:55
or add a nextcallees function that returns them all 17:56
that would maintain better inlineability for nextcallee itself 17:57
interesting that shift_caller deconts after pulling from an nqp list, so something (probably .wrap?) pokes containerized routines into it 18:04
I wonder why we don't decont when poking things in, rather than when pulling candidates out...
or maybe it's superstitious?
seems like a heck of a lot of unnecessary deconts, if so, so probably is there for a reason 18:05
I guess we need to be able to store wrappable candidates 18:06
TimToady wonders how much faster we'd run if we didn't allow such late wrapping, and what would break 18:08
maybe it type-specializes away somewhere
Zoffix Well, I *think* I succeeded in my quest: my $r = &?ROUTINE; my @nextcallees = .[.first(* === $r, :k) ^.. *] with $r.dispatcher.candidates;
TimToady I guess .dispatcher will return the nextsame list, not the whole list 18:09
Zoffix of wait, It also needs a .grep: *.cando: \(.signature) or whatever 18:10
Zoffix moves on...
TimToady huh, p6finddispatcher is heavier than I expected, seems to act more like p6createbrandnewdispatcher... 18:21
probably a consequence of not having cons lists...
or maybe I'm reading it wrong 18:22
[Coke] gets ready to get laid off Friday!
Zoffix :( 18:23
[Coke] (new job literally starts Saturday, everything is fine)
Zoffix say $new-job cmp $old-job
TimToady m: class Foo { method x { say &?ROUTINE.dispatcher.WHICH } }; class Bar is Foo { method x { say &?ROUTINE.dispatcher.WHICH; nextsame } }; Bar.new.x 18:25
camelia Mu|U36394168
Mu|U36394168
TimToady okay, it doesn't clone dispatchers for nextsame, whew! :)
just uses the current dispatcher destructively 18:26
SmokeMachine is it just for me or is t/spec/S05-modifier/ignorecase.rakudo.moar broken? 18:30
and t/spec/S09-hashes/objecthash.rakudo.moar seems to have a redeclared %h there too... 18:31
Zoffix I saw ignorecase. pass a TODO once in a while recently...
TimToady SmokeMachine: are you running on HEAD? 18:32
maybe you need . config.status or so? 18:33
SmokeMachine TimToady: I think so... I just rebase my changes and tried again...
maybe...
Ill see that..
TimToady could be the tests outracing the bumps... 18:34
SmokeMachine this is the problem for objecthash: github.com/perl6/roast/blob/master...hash.t#L87 18:35
redeclaring the %h
TimToady I'm guessing whoever added that didn't run the test individually, and the warning got hidden 18:47
SmokeMachine the ignore case test worked after I ./Configure again...
TimToady I can never remember the Configure options, which is why I do . config.status :) 18:48
SmokeMachine :) 18:50
Zoffix m: say DateTime.now > DateTime: :2016year 18:57
camelia Cannot resolve caller Real(DateTime: ); none of these signatures match:
(Mu:U \v: *%_)
in block <unit> at <tmp> line 1
Zoffix :/ I thought we added these :\
m: say DateTime.now after DateTime: :2016year
camelia Use of uninitialized value of type DateTime in string context.
Methods .^name, .perl, .gist, or .say can be used to stringify it to something meaningful.
in block <unit> at <tmp> line 1
True
Zoffix m: say DateTime.now after DateTime: :2022year
camelia Use of uninitialized value of type DateTime in string context.
Methods .^name, .perl, .gist, or .say can be used to stringify it to something meaningful.
in block <unit> at <tmp> line 1
True
Zoffix :(
terrible
Oh, I'm just missing a .new \o/ 18:59
m: say DateTime.now > DateTime.new: :2016year
camelia True
Zoffix m: say DateTime.now > DateTime.new: :2018year
camelia False
samcv Zoffix, those todo's pass magically for unknown reasons
just flapping
Geth rakudo: FCO++ created pull request #1051:
Fix multidimensional hash slice
roast: FCO++ created pull request #259:
Add tests for multi dimensional hash slice
Zoffix samcv: why magically? It uses .pick, so randomly it picks examples that avoid the bug, don't it?
samcv which is kind of weird in and of itself.
maybe 19:00
that may be true
i will adjust it
i see why it could pass wrongly. fixing 19:03
SmokeMachine this PR fixes the multi dimensional hash slice... but without it, the only feature thats right takes 0.02741633 on my machine... with it it takes 0.1098307, but it fix what didn't work without that...
samcv gonna do ('a'..'r','u'..'z').flat.pick
SmokeMachine its much slower... but works... is that ok?
samcv is there a way to do ranges like this without havng to make two and then flatten it? 19:04
Zoffix SmokeMachine: yes. Wrong answer that's fast is worse than right answer that's slow :) 19:05
m: sub postcircumfix:<{; }> (\a, \b) { dd b }; my %h; %h{1;1...5} 19:06
camelia (1, (1, 2, 3, 4, 5).Seq)
Zoffix SmokeMachine: ^ looks like it's possible to pass a Seq as a slice. Does your code handle it right? 'cause Seq isn't a List 19:07
and you have nqp::istype(blah, List) up in there
SmokeMachine Zoffix: yes... I didn't thought about the Seq... 19:08
fixing
Zoffix SmokeMachine: I think people usually use nqp::istype(blah, Iterable)
OTOH Hash/Map are also Iterable, so *shrug*
SmokeMachine Zoffix: great@
!
Zoffix: Positional? 19:09
Zoffix SmokeMachine: Seq isn't positional
SmokeMachine :(
Zoffix :)
SmokeMachine: I think Iterable is the best bet.
SmokeMachine ok! fixing!
Geth roast: bacc9cb41c | (Samantha McVey)++ | S05-modifier/ignorecase.t
ignorecase.t don't generate random strings containing the needle

This *may* have been the reason some of these tests were passing sometimes		 19:13
samcv ok should maybe be fixed now. now it'll repick if it picks a string that contains 'st' in it 19:14
SmokeMachine Zoffix: changed... 19:18
if the PR is ok, can I accept it? 19:19
Zoffix Probably :) 19:21
samcv SmokeMachine, how commonly used is this in perl 6 code? 19:22
Zoffix m: use nqp; my \indices = [1..*]; dd nqp::iseq_I(indices.elems, 1),
camelia P6opaque: get_boxed_ref could not unbox for the representation '20' of type Failure
in block <unit> at <tmp> line 1
SmokeMachine samcv: I don't think its very used...
samcv ok then making it correct is probably better than making it fast 19:23
Zoffix
.oO( when is making it fast is better than making it correct :D )
m: sub rand { 4 }; say rand
camelia 4
Zoffix I guess .is-prime falls into that category 19:26
Since it's not 100% right
SmokeMachine Zoffix: why isn't it 100% right?
Zoffix SmokeMachine: because it's probabilistic :) It uses a formula that most of the time gives right results, but not 100% all the time. There are some inputs that it'll say they're prime/not-prime 19:27
SmokeMachine 'make' is not recognized as an internal or external command? 19:28
samcv Zoffix, which inputs are those?
and are they in roast?
Zoffix sudo apt-get install build-essential?
SmokeMachine what does that mean? ci.appveyor.com/project/moritz/rak...5gea2bqd59 19:29
Zoffix: ^^
Zoffix Man, what';s the name of a popular Perl guy? 19:30
samcv msdn.microsoft.com/en-us/library/1kay26wa.aspx
dunno
i have seen it before
jnthn, ?
Zoffix Damn. Damian Conway
*sigh* was trying to remember the name to find the talk about the .is-prime stuff
This one: www.youtube.com/watch?v=Nq2HkAYbG5o 19:31
I *think* it's it; I attended the "extended version"; but maybe regular version has is prime stuff too
And well, it names what we use for is-prime and using that name you can google what inputs it's not right for :)
Zoffix guesses you'd have to wait a lot of time for it to complete to process them :) 19:32
samcv i think i've seen that talk. was good. 19:33
Zoffix SmokeMachine: no idea. Something's up with appveyor again. It's not just your commit, all the previous ones have that problem too 19:34
jnthn samcv: Sounds like the issue that was fixed in github.com/MoarVM/MoarVM/commit/36...14c82e37b5 19:36
samcv cool
jnthn Just means that it doesn't cope with sizeof(void *) or so 19:37
Or more like, won't let you do pointer arith on void*
Since it doesn't know the units
I guess other compilers assume the unit is bytes
timotimo SmokeMachine: i think you can get away with only calling SELF.EXISTS-KEY when the outermost and second-from-outermost if hits its else-branch 19:51
SmokeMachine: also, check how indices[1 .. indices.end] compares to indices[1..*] and indices.skip(1)
m: say (1, 2, 3, 4).skip(1)
camelia (2 3 4)
SmokeMachine m: my @a = ^10; $ = @a.skip(1) for ^10000; say now - INIT now 20:06
camelia 0.0568199
SmokeMachine m: my @a = ^10; $ = @a[1 .. @a.end]for ^10000; say now - INIT now
camelia 0.6056935
SmokeMachine changing to skip!
m: my @a = ^10; say = @a.skip 20:07
camelia 5===SORRY!5=== Error while compiling <tmp>
Preceding context expects a term, but found infix = instead
at <tmp>:1
------> 3my @a = ^10; say =7ā5 @a.skip
SmokeMachine m: my @a = ^10; say @a.skip
camelia (1 2 3 4 5 6 7 8 9)
SmokeMachine m: my @a = ^10; $ = @a.skip for ^10000; say now - INIT now
camelia 0.0490078
SmokeMachine timotimo: thanks!
m: my say ^10 .skip 20:10
camelia 5===SORRY!5===
Type 'say' is not declared
at <tmp>:1
------> 3my say7ā5 ^10 .skip
Malformed my
at <tmp>:1
------> 3my7ā5 say ^10 .skip
SmokeMachine :(
timotimo haha "my say"? :) 20:11
SmokeMachine m: say ^10 .skip 20:12
camelia (1 2 3 4 5 6 7 8 9)
SmokeMachine (sorry)
m: my $a = ^10; $ = $a.skip.cache for ^10000; say now - INIT now
camelia 0.09728743
SmokeMachine with skip it got from 0.1098307 to 0.0547416 20:17
timotimo *nice*
SmokeMachine thanks!
(but I had to .skip.cache)
if/when you guys think its ok to merge the PR, please tell me 20:21
Zoffix yeah merge it 20:48
lizmat m: my %h{Any} = now => 42; dd %h # guess this is a case of DIHWIDT 21:01
camelia Hash[Any,Any] %h = (my Any %{Any} = :now(42))
SmokeMachine lizmat: sorry, what means DIHWIDT? 21:04
lizmat Doctor, It Hurts When I Do This
(then don't do that :-)
timotimo yas
[Coke] buggable: DIHWIDT? 21:05
lizmat m: my %h{Any} = now,42; dd %h
camelia Hash[Any,Any] %h = (my Any %{Any} = Instant.from-posix(<1265707485887/849>, Bool::False) => 42)
[Coke] ETOOMANYBOTS
lizmat ah, better :-)
SmokeMachine Zoffix: create a merge commit, squash and merge or rebase and merge? 21:06
Zoffix SmokeMachine: dunno, just click some button :) 21:08
I'd say it's a case of DWIM. You asked for quoted key, so you got it. It'd be pretty weird if your keys got automatically changed the second you imported some term or constant. 21:09
m: my %h{Any} = (now) => 42; dd %h
camelia Hash[Any,Any] %h = (my Any %{Any} = Instant.from-posix(<1862036359552/1249>, Bool::False) => 42)
Zoffix m: my %h{Any} = 42 R=> now; dd %h
camelia Hash[Any,Any] %h = (my Any %{Any} = Instant.from-posix(<2512034655006/1685>, Bool::False) => 42)
Geth rakudo/nom: a758c0bbf9 | (Fernando Correa)++ | src/core/hash_slice.pm
Fix multidimensional hash slice

  &postcircumfix:<{; }> wasnt prepard to receive a List as one (or more) position(s)
of the argument list (eg: %h{"bla", "ble"; "bli"})
rakudo/nom: dc812a4a5a | (Fernando Correa de Oliveira)++ | src/core/hash_slice.pm
Merge pull request #1051 from FCO/multidimensional-hash-slice

Fix multidimensional hash slice
lizmat SmokeMachine++
Geth roast: a25592a533 | (Fernando Correa)++ | S09-hashes/objecthash.t
Add tests for multi dimensional hash slice
 21:10
roast: fc4e2c7b25 | (Fernando Correa de Oliveira)++ | S09-hashes/objecthash.t
Merge pull request #259 from FCO/multidimensional-hash-slice

Add tests for multi dimensional hash slice
SmokeMachine lizmat: thanks!
Zoffix SmokeMachine: what does the .skip.cache do tho? 21:11
lizmat the .skip skips the first element of the sequence, the .cache makes sure that you can .AT-POS it without losing the iterator
SmokeMachine Zoffix: what lizmat sad 21:12
Zoffix Ah
SmokeMachine Zoffix: and @a.skip is 10 times faster than @a[1 .. @a.end] 21:13
Zoffix Sweet
Geth rakudo/nom: a8c6eca78d | (Elizabeth Mattijsen)++ | src/core/set_operators.pm
Make Str (elem) Map / Map (cont) Str 50x faster

At least. Because this timing is for a one element hash. Since these candidates don't need to coerce anything to a Set, the gain for larger Maps / Hashes would be even bigger.		 21:17
lizmat now, if I would have a way to do MMD on object hashes... 21:18
I could do the same for Any (elem) ObjHash
Zoffix Holy shit. 21:19
lizmat++
DrForr lizmat: Before I forget, Perl6::Tidy does some very primitive indentation.
SmokeMachine lizmat++
lizmat: sorry again, but whats MMD? 21:20
lizmat multi method dispatch
SmokeMachine thanks!
b2gills That commit does something slightly different to calling .Set, as ļ½¢'a' (elem) { 'a' => 0 }ļ½£ will now return True but used to return False. 21:36
I think it is an improvement though, as it makes more sense.
lizmat argh 21:37
yes
hmmm
fwiw, that wasn't picked up by spectesting 21:38
m: say 'a' (elem) { 'a' => 0 }
camelia True
lizmat argh, too late :-) 21:39
nebuchadnezzar lizmat: isn't bisectable6 here for that purpose? 21:53
I confess I'm a little bit lost will the bots 21:54
Geth rakudo/nom: 5aebaf1fe4 | (Elizabeth Mattijsen)++ | src/core/set_operators.pm
Fix Str (elem) Map / Map (elem) Str for False values

As b2gills++ pointed out, we still need Set semantics. So we actually need to check the value truthiness. This reduces the speed increase from 50x to 15x. Again, for a 1 elem hash. Larger maps / hashes should be much faster still.
lizmat has done enough damage for today 22:02
good night, #perl6-dev!
Zoffix night 22:10
samcv night 23:31
Zoffix night 23:37
SmokeMachine m: my %a = 1 => {2 => {3 => 42}}; say %a{1;2;3}:exists 23:43
camelia True
SmokeMachine m: my %a = 1 => {2 => {3 => 42}}; say %a{1,1;2;3}:exists
camelia (True True)
SmokeMachine m: my %a = 1 => {2 => {3 => 42}}; say %a{1;*;3}:exists # I think I forgot this case... what should it return? 23:44
camelia False
SmokeMachine m: my %a = 1 => {2 => {3 => 42}}; say %a{1;*;3} 23:46
camelia (42)
Zoffix m: .say with do with 42 { $_ } 23:51
camelia 42
Zoffix loves it

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!