Geth rakudo: 18252b60d7 | (Daniel Green)++ | lib/Telemetry.pm6
Simplify Telemetry's new

Use nqp::splice instead of doing a push in a loop.
Also use nqp::p6bindattrinvres instead of nqp::bindattr + return.		 00:00
rakudo: b568510932 | (Elizabeth Mattijsen)++ (committed using GitHub Web editor) | lib/Telemetry.pm6
Merge pull request #1642 from MasterDuke17/simplify_Telemetrys_new

Simplify Telemetry's new
lizmat MasterDuke: is that an answer? :-) 00:01
lizmat gets some sleep
MasterDuke heh, good enough for me 00:08
lizmat++
lizmat Files=1236, Tests=76239, 277 wallclock secs (13.22 usr 4.89 sys + 1889.03 cusr 190.95 csys = 2098.09 CPU) 06:51
that's down from 316 on 20 March 06:53
12% faster :-)
lizmat thinks that should be visible in test-t also :-) 06:56
breakfast& 06:59
Geth rakudo: 723cbda796 | (Elizabeth Mattijsen)++ | src/core/List.pm6
Don't use private methods for utility methods
 08:45
rakudo: de1c6dc9e0 | (Elizabeth Mattijsen)++ | src/core/Array.pm6
Change some Array private utility method to normal methods

Avoid relatively costly dispatch:<!> runtime lookup. Also change names of methods to use uppercase only / underscore to "show" privateness.		 09:01
travis-ci Rakudo build errored. Elizabeth Mattijsen 'Change some Array private utility method to normal methods 09:58
travis-ci.org/rakudo/rakudo/builds/357722371 github.com/rakudo/rakudo/compare/7...1c6dc9e0dd
buggable [travis build above] ☠ Did not recognize some failures. Check results manually.
lizmat Files=1236, Tests=76226, 316 wallclock secs (15.05 usr 5.45 sys + 2180.38 cusr 212.10 csys = 2412.98 CPU) 11:37
hmmm... weird, back to 316 now
wonder what made it go down to 277 last night
afk again& 11:48
FROGGS it is known that require ::($module) explodes in async code? 13:34
is it*
nine is not surprised at least 13:44
Geth nqp: aa605d0ef6 | (Aleks-Daniel Jakimenko-Aleksejev)++ | tools/build/MOAR_REVISION
[MoarVM Bump] Brings 11 commits

MoarVM bump brought: github.com/MoarVM/MoarVM/compare/2...g67056c204 67056c204 Fix optimization of zombie conditional causing segfault a3bd17b4f Fix windows1252/1 decodestream replacement with last pos repl. dd2072da1 Fix missing `:` in ChangeLog, causing website not to generate right 41711ff26 Make sure shiftjis_codeindex.h is added to Makefile.in ... (7 more lines)		 13:55
¦ nqp: version bump brought these changes: github.com/MoarVM/MoarVM/compare/2...g67056c204
rakudo: 3b86ffd563 | (Aleks-Daniel Jakimenko-Aleksejev)++ | tools/build/NQP_REVISION
[NQP Bump] Brings 15 commits

NQP bump brought: github.com/perl6/nqp/compare/2018....gaa605d0ef aa605d0ef [MoarVM Bump] Brings 11 commits 02a2378f3 [js] Wrap native strings in NQPStr bb14087e0 [js] Fix nqp::encoderep with utf8 5bcb548b3 Extra decoding tests ... (11 more lines)
rakudo: version bump brought these changes: github.com/perl6/nqp/compare/2018....gaa605d0ef
6be586c580 | (Aleks-Daniel Jakimenko-Aleksejev)++ | tools/build/NQP_REVISION

This reverts commit 3b86ffd563c77da210c98b02e9aaf1539ecfdcd5.
AlexDaniel wonders why Nick used RT to submit these bugs 14:04
travis-ci NQP build failed. Aleks-Daniel Jakimenko-Aleksejev '[MoarVM Bump] Brings 11 commits 14:10
travis-ci.org/perl6/nqp/builds/357782068 github.com/perl6/nqp/compare/02a23...605d0ef6da
dogbert11 AlexDaniel: I have problems with your latest bump 14:12
AlexDaniel dogbert11: maybe that's why nobody bumped before me! :) 14:13
fwiw I just woke up and didn't even run the spectest, but really wanted to play with the segfault fix
dogbert11: what's the problem exactly? 14:14
dogbert11 give it a go, perhaps I'm the only one experiencing problems
rakudobrew is struggling 14:15
===SORRY!===
Missing or wrong version of dependency 'gen/moar/BOOTSTRAP.nqp' (from '/home/dogbert/.rakudobrew/moar-master/lib/CompUnit/Repository/Staging.pm6 (CompUnit::Repository::Staging)')
had to do a make realclean in order make my non rakudobrew install to build 14:16
AlexDaniel yeah, travis is unhappy too
dogbert11 tries make realclean on rakudobrew as well ... 14:17
AlexDaniel hah, it's green on osx 14:19
dogbert11 aaand rakudobrew is happy again 14:20
AlexDaniel wait what 14:21
Oracle JDK 8 is NOT installed.
am I reading this right?
travis-ci.org/rakudo/rakudo/jobs/357782093
dogbert11 a lot of test failures as well
AlexDaniel so a moar build attempts to download openjdk? why? 14:22
or, well, whateverjdk
MoarVM oops: Trying to compile if_o/unless_o, should never happen 14:23
¦ rakudo: version bump brought these changes: github.com/perl6/nqp/compare/2018.......2018.03
nqp: 2aea8c9030 | (Aleks-Daniel Jakimenko-Aleksejev)++ | tools/build/MOAR_REVISION
Revert "[MoarVM Bump] Brings 11 commits"

This reverts commit aa605d0ef6da958fd87f1c7342e361cf21c12d01.		 14:24
¦ nqp: version bump brought these changes: github.com/MoarVM/MoarVM/compare/2......2018.03
dogbert11 heh, that bump didn't survive very long :) 14:25
dogbert11 hmm the revert didn't work that well either 14:30
AlexDaniel what? 14:36
dogbert11: can't be? 14:37
travis is failing because of dependency on oracle.com, it's alright 14:38
dogbert11 I reinstalled my rakudo and now all seems to be well
AlexDaniel good.
dogbert11 so what's wrong with the reverted code I wonder 14:40
AlexDaniel dogbert11: looks like the segv fix is at fault 14:41
at the very least it broke the build on windows
nine AlexDaniel: pushed a fix for "'optimize_iffy' : not all control paths return a value" 15:07
travis-ci Rakudo build errored. Aleks-Daniel Jakimenko-Aleksejev 'Revert "[NQP Bump] Brings 15 commits" 15:09
travis-ci.org/rakudo/rakudo/builds/357788456 github.com/rakudo/rakudo/compare/3...e586c580a6
buggable [travis build above] ☠ Did not recognize some failures. Check results manually.
nine But that was just a warning. I've got no idea about the "MoarVM oops: Trying to compile if_o/unless_o, should never happen". I don't see how my patch should cause that. The code was clearly wrong before (operating on a deleted instruction) and is clearly more right now and the failure appears in some travis builds with no apparent system 15:10
timotimo if we had an MVM_SPESH_LOG of the crashing invocation, we'd probably be able to tell what's going on 15:11
travis-ci Rakudo build errored. Aleks-Daniel Jakimenko-Aleksejev 'Revert "[NQP Bump] Brings 15 commits"
travis-ci.org/rakudo/rakudo/builds/357788456 github.com/rakudo/rakudo/compare/3...e586c580a6
buggable [travis build above] ☠ Did not recognize some failures. Check results manually.
AlexDaniel nine: cool! feel free to bump if everything looks alright 15:14
nine Why don't we always run travis builds with MVM_SPESH_LOG?
timotimo spesh logs are huge :) 15:15
nine Is that an issue?
timotimo let me check exactly how big they are
though of course if we set the same filename for all, rather than put a %d in there for a pid ... 15:16
we'd only get the last file output, which would usually be the one for any crash
888 MB for an nqp build, not counting "make test" 15:17
1.6 gigs when adding "make test" 15:18
nine That's with an individual log file for each process?
timotimo yes
nine What's the largest one's size? 15:19
timotimo 84 megs
nine Now that sounds less intrusive :)
timotimo smallest is empty, second-smallest is 262K
let me bzip2 it
lol. 15:23
bzip2'd the whole shebang and it's also 84 megs big
we could probably build a nice dictionary of spesh log pieces and zstd-compress the stuff 15:25
AlexDaniel timotimo: can you try lrzip-ing it? 15:33
timotimo what's that?
timotimo installs
AlexDaniel long range zip
timotimo hm, "command not found"; what provides it? 15:34
AlexDaniel are you on something debian-based? lrzip is the package name
timotimo fedora
with zstd it's only 44 megs instead of 84, but that's with -19 and takes 6 minutes :D 15:35
AlexDaniel what version of zstd?
it had long range mode also
timotimo *** zstd command line interface 64-bits v1.3.3, by Yann Collet ***
you know, you can just get the files locally, too 15:36
set MVM_SPESH_LOG=nqpbuild.%d.txt and make && make test in nqp's folder
AlexDaniel :)
timotimo since you got a bunch of experience with compression stuff 15:37
AlexDaniel well, this is what whateverables use to put a bunch of rakudo builds into the same archive (otherwise you can compress them individually with exactly the same ratio) 15:38
timotimo i see 15:39
AlexDaniel dunno if spesh logs will benefit from that
AlexDaniel tries
timotimo any experience with pre-built dictionaries?
AlexDaniel yes, no luck with rakudo builds 15:40
and it's only supposed to improve things with small files 15:41
timotimo ah, that does make sense
since the compressor needs a bit of data to "learn" at the beginning
AlexDaniel timotimo: wait uh, so what's the ratio in your case? 15:44
I just compressed two spesh logs to 3% of their size 15:45
timotimo 1.6 gigs to 44 megs
using 6 minutes of compression work
AlexDaniel (≈80M original, <2M result)
ah ok
I mean… can you really want it to be better than that? :D
timotimo maybe!
the information density is really low in spesh logs 15:46
nine is out for the evening 15:52
AlexDaniel timotimo: ok it doesn't sound right, it should be smaller 15:54
one second I'll figure out what you're doing wrong… 15:55
timotimo i could unlock the higher compression levels :D
60 megs of ram usage is nothing
AlexDaniel no-no
higher compression levels generally just make it slower 15:56
you need long range mode if you're thinking in terms of ram
timotimo: try zstd -15 --long=31
that should work about 60% better than without --long
timotimo wow 15:57
AlexDaniel 1.78% of the original size
timotimo: but it's only useful if you're applying that to whole 1.6 gigs :) 15:58
you can try -19 too, it's not any slower I think 15:59
and it will save 10% on top of that
AlexDaniel ♥ zstd
timotimo nice, only 1:30m 16:00
27 megs, cool.
in theory we could re-order the files to have plans in one file and before/after spesh dumps in another, with little markers for reconstruction, but maybe --long does that almost as good as we could, but automatically 16:01
AlexDaniel if you want it to be smaller we'll probably need another breakthrough in compression methods :D
timotimo to be honest, now that we have messagepack built into moarvm itself, i'd like to have a mode for spesh logs that uses msgpack 16:02
that's for later, though
it will however let me put in more information that'll let me figure out 1:1 which instructions were where before
like, this instruction has been added, these instructions have been kicked out, this instruction has been turned from one opcode into another
currently the spesh log outputs the BB with their memory addresses so even though we renumber the bbs at every step, you can tell which one was which 16:04
messagepack will make it compress much less, but it'll be smaller to begin with 16:06
it'll also need a custom viewer, though
that'll be a bit of development time all in all, so i'll probably do it either when my grant is done or near the end
unless i find it'd help me with the grant, of course
so in the travis.yml we would probably want to only create the compressed stuff if anything actually fails 16:20
if we do that, we'll also want to know the PID of everything we run
because that way we can see which spesh file corresponds to what command 16:21
stmuk . 17:26
Geth rakudo/js: 6 commits pushed by pmurias++ 17:53
[Tux] Rakudo version 2018.03-64-g6be586c58 - MoarVM version 2018.03-3-g2bd691551
csv-ip5xs0.825 - 0.848
csv-ip5xs-207.846 - 8.003
csv-parser37.953 - 38.629
csv-test-xs-200.448 - 0.454
test9.080 - 9.481
test-t2.427 - 2.556
test-t --race1.012 - 1.157
test-t-2044.845 - 46.206
test-t-20 --race16.217 - 19.936
 18:02
:) 18:03
2018-01-30 08:39:24 test-t 2.509
2018-01-29 17:52:01 test-t 2.502
2018-01-29 09:46:21 test-t 2.497
2018-02-17 11:21:49 test-t 2.491
2018-03-24 16:49:18 test-t 2.427
timotimo very nice 18:04
japhb I am curious what's up with test-t-20 --race. Those times have really high variance. 18:07
MasterDuke what's the bot command to create the graphs? 18:21
huggable: test-t
huggable MasterDuke, nothing found
MasterDuke buggable: test-t
AlexDaniel tux.nl/Talks/CSV6/speed4.html ? 18:22
buggable: speed
buggable AlexDaniel, ▄▄▅▆▃▃▇▄▃▄▄▄▅▂▆▂▃▁▄▆↑▂▃▂▂▂▃▃▄▆▅▂▂▁▇█▂▃▁▂▄▂▂▂▃▄▃▂▃↑ dates: 2018-02-26–2018-03-24; range: 15.487s–20.488s; speed: 4% faster (widths: 9/3) 18:23
AlexDaniel o I guessed it
MasterDuke buggable: speed test-t-20 --race
buggable: speed test-t-20
buggable MasterDuke, ▃▆▇▆▄▃▇▇▃▄▄▃▄▄▇▄▂▃▅▇↑↑▃▃▂▄▄▄▅▃▅▆▂▂▇▇▂▃▂▃▃▄▃▂▆▂▄▄▁▃ dates: 2018-02-26–2018-03-24; range: 44.845s–49.791s; speed: 4% faster (widths: 9/3)
Geth nqp: d6e7d9364c | (Zoffix Znet)++ | tools/build/MOAR_REVISION
[MoarVM Bump] Brings 12 commits

MoarVM bump brought: github.com/MoarVM/MoarVM/compare/2...2-gb5b6968 b5b6968 Fix "'optimize_iffy' : not all control paths return a value" 67056c2 Fix optimization of zombie conditional causing segfault a3bd17b Fix windows1252/1 decodestream replacement with last pos repl. dd2072d Fix missing `:` in ChangeLog, causing website not to generate right ... (8 more lines)		 18:47
¦ nqp: version bump brought these changes: github.com/MoarVM/MoarVM/compare/2...2-gb5b6968
rakudo: 2c882d9601 | (Zoffix Znet)++ | tools/build/NQP_REVISION
[NQP Bump] Brings 17 commits

NQP bump brought: github.com/perl6/nqp/compare/2018....7-gd6e7d93 d6e7d93 [MoarVM Bump] Brings 12 commits 2aea8c9 Revert "[MoarVM Bump] Brings 11 commits" aa605d0 [MoarVM Bump] Brings 11 commits 02a2378 [js] Wrap native strings in NQPStr ... (27 more lines)
¦ rakudo: version bump brought these changes: github.com/perl6/nqp/compare/2018....7-gd6e7d93
rakudo/js: 77c462b571 | pmurias++ | src/core/Iterable.pm6
[js] Make .hyper and .race just return the unmodified iterator

Most JS VMs lack the threading support needed for that
dogbert11 hmm, a new bump 18:55
dogbert11 wonders if that'll work better
Geth roast: 39d4ac68ef | (Zoffix Znet)++ | S03-metaops/hyper.t
Fix incorrect nodality tests for reduce/produce

Closes github.com/rakudo/rakudo/issues/1633
The tests' description say they're testing that reduce/produce
  are nodal, and it makes sense for them to be nodal, however, the
  $expected specified in the tests indicates the non-nodal behaviour.
Likely, the $expected was simply copy-pasted from the output of running the $got code and plugged into the test, without noticing that the result is incorrect.
roast/6.c-errata: b8dca86f45 | (Zoffix Znet)++ | S03-metaops/hyper.t
Fix incorrect nodality tests for reduce/produce

Closes github.com/rakudo/rakudo/issues/1633
The tests' description say they're testing that reduce/produce
  are nodal, and it makes sense for them to be nodal, however, the
  $expected specified in the tests indicates the non-nodal behaviour.
Likely, the $expected was simply copy-pasted from the output of running the $got code and plugged into the test, without noticing that the result is incorrect.
travis-ci Rakudo build passed. Aleks-Daniel Jakimenko-Aleksejev 'Revert "[NQP Bump] Brings 15 commits"
travis-ci.org/rakudo/rakudo/builds/357788456 github.com/rakudo/rakudo/compare/3...e586c580a6
AlexDaniel Zoffix++ this is freaking awesome 19:24
japhb AlexDaniel: What is freaking awesome? 21:16
AlexDaniel japhb: github.com/MoarVM/MoarVM/commit/06...47211b874b 21:17
japhb Ah, nice 21:24

Powered by Cro and the Raku® Programming Language.

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

Thank you!