Perl 6 - the future is here, just unevenly distributed

IRC log for #perl11, 2015-12-10

| Channels | #perl11 index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
00:28 travis-ci perl11/cperl#448 (smoke/config_too_big_ints_fix - 8509f4b : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95916725
00:29 travis-ci perl11/cperl#448 (smoke/config_too_big_ints_fix - 8509f4b : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95916725
01:42 willthechill joined #perl11
02:05 travis-ci perl11/cperl#451 (master - 8509f4b : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95941101
02:10 bulk88 grr, I kindda accidentally pushed to master
02:11 travis-ci perl11/cperl#451 (master - 8509f4b : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95941101
02:14 travis-ci perl11/cperl#451 (master - 8509f4b : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95941101
02:42 travis-ci perl11/cperl#452 (master - f5fd78d : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95945788
02:46 travis-ci perl11/cperl#452 (master - f5fd78d : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95945788
02:48 travis-ci perl11/cperl#452 (master - f5fd78d : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95945788
03:23 travis-ci perl11/cperl#454 (smoke/config_improve_dotpm_loader - e18c526 : bulk88): The build was canceled. https://travis-ci.org/perl11/cperl/builds/95947913
03:37 sten joined #perl11
03:57 travis-ci perl11/cperl#455 (smoke/config_improve_dotpm_loader-2 - 1cf72d0 : bulk88): The build was canceled. https://travis-ci.org/perl11/cperl/builds/95951960
03:57 travis-ci perl11/cperl#455 (smoke/config_improve_dotpm_loader-2 - 1cf72d0 : bulk88): The build was canceled. https://travis-ci.org/perl11/cperl/builds/95951960
03:57 travis-ci perl11/cperl#455 (smoke/config_improve_dotpm_loader-2 - 1cf72d0 : bulk88): The build was canceled. https://travis-ci.org/perl11/cperl/builds/95951960
03:59 travis-ci perl11/cperl#455 (smoke/config_improve_dotpm_loader-2 - 1cf72d0 : bulk88): The build was canceled. https://travis-ci.org/perl11/cperl/builds/95951960
04:28 willthechill joined #perl11
04:53 travis-ci perl11/cperl#457 (smoke/config_improve_dotpm_loader-4 - b5eaab8 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95959051
04:56 travis-ci perl11/cperl#457 (smoke/config_improve_dotpm_loader-4 - b5eaab8 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95959051
04:58 travis-ci perl11/cperl#457 (smoke/config_improve_dotpm_loader-4 - b5eaab8 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/95959051
06:10 sten joined #perl11
07:53 basiliscos joined #perl11
09:18 rurban joined #perl11
09:29 rurban bulk88: for smoke/appveyor_find_alternate_maketool you need to change the image from Default Azure to MingW. at least gmake. dmake no idea.
09:30 rurban I did try some Mingw builds and probes, but failed previously.
09:31 bulk88 I want non-nmake with MSVC since I want to see if a parallel build will reduce that 30 mins per job
09:31 bulk88 I also suspect a couple 180 second timeouts are happening during the testing, which are race conditions or SEGvs
09:31 bulk88 or "VM" timing issues
09:32 bulk88 the perl test suite often has "watchdog
09:32 bulk88 " set to 180 seconds
09:33 bulk88 im trying a Win64 build for the first time right now to see if I cna fix any more test fails
09:34 rurban then you probably you need to install a dmake from somewhere
09:35 rurban `cinst dmake` maybe
09:35 rurban git log -p appveyor.yml
09:38 rurban dmake is not on nuget. https://www.nuget.org/packages?q=dmake
09:38 bulk88 if it takes 2 minutes to install dmake off the web, to do parallel building, it would waste time more time to download it and extract vs serial building
09:39 bulk88 the box has only 1 core since I guess you are using the free plan, so only disk io would cause the other thread/process to run for a bit
09:40 rurban we should restrict appveyor to smoke and master only, right?
09:41 bulk88 I think so, since its 2 hours per branch, your mass rebase requires alot of manual cancelling each time
09:41 rurban I'll merge smoke/config_improve_dotpm_loader-4 soon, rebased to the int as string change
09:47 rurban https://chocolatey.org/packages?q=dmake also nothing
09:50 rurban Config.t 3-4 need some TODOs or exceptions on windows. but at least 32bit doesn't fail with the byteorder issue
09:51 rurban https://ci.appveyor.com/project/rurban/cperl/build/5.22.2.661/job/98g2l5tyipxor8qt#L4315
09:52 rurban I'll update the memory graphs now with the superior Config
09:54 bulk88 if I have gperf in PATH config.t passes on Win32, I wonder whether to bundle gperf in cperl, and build it for win32 but never install it (like miniperl's status), I have plans to release XS Config on CPAN, so config.t for no-gperf will need to be solved by adding Win32-only keys to the template like you already do for some other vars
09:55 bulk88 Config.t will need 2 different tests, depending on whether gperf was available when config.dll was built, if no gperf, for every PP key, check that the XS key is identical
09:55 bulk88 if gperf was available, the 2 hashes must be identical in keys and values
09:56 bulk88 no gperf XS Config will always be more bloated than gperf available XS Config
09:57 bulk88 my plan is that the template will be rebuilt, but not used, if gperf is available, so 2 gperf runs happen when Config_xs.PL runs, 1 generated the commited template, the other generates a "no extra keys" version specific for that particular perl
09:58 bulk88 the maintainer just needs to remember to commit the template
09:58 bulk88 there is also the question of soft deleting keys in the template to make keys disappear on a Perl level (but obviously they cant disappear on a machine code/disk level without running gperf)
10:00 rurban We also need to harmonize Config.t and t/Config.t. Maybe: mv Config.t t/XS.t
10:01 rurban and the version number is currently bound to perl_core. there needs to be a better $VERSION > "5.02"
10:02 rurban and the regen funcs need to be extracted.
10:02 bulk88 I'd leave the 2 Config.t files alone, one is from cperl, the other is from p5p, the p5p one should stay unchanged as much as possible
10:02 rurban yes, but they need to be in t/
10:03 rurban and Config XS is not really cperl specific. It should be on CPAN
10:03 rurban same for warnings and later unicode
10:04 rurban Encode did it right by letting CPAN maintaining it, because they know how to do it. unicore is horror compared to Encode.
10:06 bulk88 http://paste.scsys.co.uk/502593 the Win64 failures I reproduced, since it is the end of my day right now I dont think Ill be doing much till later today
10:06 bulk88 alarm.t is a race condition
10:10 rurban alarm is interesting, the rest is known. thanks
10:11 rurban 2126 secs!!! I have 240secs
10:12 rurban 10x times slower doesn't make me enthusiastic having this to support. is this a superslow machine with no RAM?
10:12 rurban appveyor needs 1544 wallclock secs
10:15 bulk88 each SEGV is a 180 timeout timer
10:16 bulk88 so you run the test for 1 second, it SEGVs, then you sleep for 179 seconds
10:16 bulk88 https://ci.appveyor.com/project/rurban/cperl/build/5.22.2.655 one of them took 1 hr, it timed out appveyor wise
10:17 bulk88 the test that froze was kill0.t, I actually wrote it, perhaps it needs limits to its sleep loops
10:17 travis-ci perl11/cperl#459 (smoke/config_improve_dotpm_loader-4 - 3934194 : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/95992039
10:21 travis-ci perl11/cperl#459 (smoke/config_improve_dotpm_loader-4 - 3934194 : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/95992039
10:21 rurban memory went down from 595302 byte to 586879
10:22 rurban oh, the SEGV's
10:22 rurban will leave now for lunch. bye!
11:01 rurban Updated memory graph for the upcoming cperl5.22.2 release. With warnings and Config_heavy https://imgur.com/HhsOM1G
11:02 rurban esp. with warnings as XS (cperl mem) it's shining
11:03 rurban more than 2x less memory!
11:11 sten1 joined #perl11
13:10 rurban joined #perl11
16:29 travis-ci perl11/cperl#461 (bugfix/gh59-dump - 92c39ee : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058007
16:33 travis-ci perl11/cperl#461 (bugfix/gh59-dump - 92c39ee : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058007
16:36 travis-ci perl11/cperl#461 (bugfix/gh59-dump - 92c39ee : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058007
17:02 bulk88_2 joined #perl11
17:03 travis-ci perl11/cperl#463 (master - 3934194 : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058079
17:04 travis-ci perl11/cperl#463 (master - 3934194 : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058079
17:07 travis-ci perl11/cperl#463 (master - 3934194 : Reini Urban): The build passed. https://travis-ci.org/perl11/cperl/builds/96058079
18:04 basiliscos joined #perl11
18:39 rurban signatures are getting close. call-by-ref done, auto-conversion from sigs to protos done, just invocant and compile-time type checking missing.
18:43 bulk88 the byteorder bug on Win64 is getting very interesting, on p5p perl it is "byteorder='1234'", yeah that is wrong, but its been wrong forever, IDK how p5p perl was passing the p5p config.t, I am trying to find out now
18:43 rurban really? oh my
18:43 rurban maybe the pack template for P is wrong then
18:44 rurban ivsize vs ptrsize. win64 is famous from deviating there
18:45 bulk88 J is not the pack letter for P
18:46 bulk88 the logic I use  https://metacpan.org/source/BULKDD/Win32-API-0.82/W32ATest.pm#L108
18:47 bulk88 ivsize and ptrsize are both 8 on Win64, they have to be because IVs must store PTRs per perl api
18:58 bulk88 this is getting stupid, $Config{byteorder} in the config DB is never used, instead whe config_heavy is loaded, byteorder is recalculated using perl code
18:58 rurban but sizeof(int) is different
18:59 rurban yes, this is dynamic. there must be some old bug or usecase for this
18:59 bulk88 sounds like technical debt from people refusing to touch metaconfig
19:00 bulk88 ill got and git blane it
19:00 bulk88 *got
19:00 rurban I'm not so sure
19:00 rurban I'm leaving now for the evening
19:00 bulk88 git blame says it was for fat packing compilers
19:01 bulk88 your .exe is both 32 bit machien and 64 bit machien code, all in the same file
19:01 bulk88 I dont think that that is sane
19:02 bulk88 same perl bin, run it on one machine its 32 bit pointers, run it ont he other, and its 64 bit pointers
19:03 bulk88 http://www.nntp.perl.org/group/perl.perl5.porters/2000/07/msg15277.html
19:03 bulk88 use for cross compiling :?
19:03 bulk88 *:-/
20:38 travis-ci perl11/cperl#468 (smoke/config_byteorder_bugfix - 07a10e2 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/96115726
20:38 travis-ci perl11/cperl#468 (smoke/config_byteorder_bugfix - 07a10e2 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/96115726
20:42 travis-ci perl11/cperl#468 (smoke/config_byteorder_bugfix - 07a10e2 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/96115726
20:55 rurban I see. What a hack. It should be only dynamic for 0xffff(unknown)
21:09 rurban merged smoke/config_byteorder_bugfix, thanks
21:09 jevin joined #perl11
21:15 rurban now we just need to ignore or add the special win keys in the test
21:18 bulk88 http://paste.scsys.co.uk/502608 my win64 build after "*Config get "byteorder" from Config, dont computer it ourselves"
21:19 bulk88 the remaining issues are specific to the 2 Visual C versions being used by appveyor, not Win64/Win32 in general
21:41 travis-ci perl11/cperl#469 (master - 2bc8ab2 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/96131152
21:45 travis-ci perl11/cperl#469 (master - 2bc8ab2 : bulk88): The build passed. https://travis-ci.org/perl11/cperl/builds/96131152

| Channels | #perl11 index | Today | | Search | Google Search | Plain-Text | summary