Perl 6 language and compiler development | Logs at colabti.org/irclogger/irclogger_log/perl6-dev | For toolchain/installation stuff see #perl6-toolchain | For MoarVM see #moarvm
Set by Zoffix on 27 July 2018.
vrurg jnthn: It demoes that $.a could fail in methods too. 00:00
Anyway, I still don't see why submethod should be limited in using `$.a`? Maybe I'm missing something, but it's just a shortcut for self.a which just works.
jnthn: I noted it in the ticket: all tests and spectests are passing if I lift this limitation. I would consider allowing it as it definitely provides more good with no harm. 00:03
jnthn vrurg: That just means we're missing spectests covering the behavior. 00:04
Commented on the rationale in the ticket. This is not an accident. It was discussed and decided on way back.
It's a pity that it didn't receive spectests at the time, though. 00:05
Sleep time here; 'night
vrurg g'night! 00:07
AlexDaniel more operators??
vrurg AlexDaniel: extension for .+: .- for reverse MRO order. Because I didn't find any good analog for .* I came up with .?- which logically pulled in .?+ 00:09
AlexDaniel: tests are in github.com/perl6/roast/pull/581 00:11
AlexDaniel I'm just wondering, do these have to be operators? I understand that's kinda their nature, but I'll never in my life remember the difference between them 00:22
vrurg AlexDaniel: because you never used them. :) 00:23
More precisely: never needed.
AlexDaniel that's absolutely true, yes
do you think they'll be used often?
.? is ok, but everything else… uhhh 00:24
vrurg A MOP method? But there .+ already. I just slightly modified it and made all four share same base method. 00:25
AlexDaniel yeah, but maybe it's .+ that is in the wrong :)
vrurg .* is extensively used in spectest 00:26
AlexDaniel and .*…
vrurg They do make sense for submethods anyway. My best example is a custom app built from a number of roles/classes implementing certain functionality. Each role/class provide their own command line options. Instead of manually collecting them I can call cli-opt submethod and get a list of definitions. 00:29
AlexDaniel these remind me so much of perl5's $… vars
vrurg: right, IRC::Client is doing something similar if I recall correctly 00:30
vrurg Operator look very much natural here because $app.-cil-opts( ... ) is way much tidier than $app.^batch-call('cli-opts', \( ... args ... ))
AlexDaniel I like the second one more
vrurg Ah, worse: $app.^batch-call('cli-opts', \( ... args ... ), :reverse)
AlexDaniel it's longer, but we're fine with longer things :P 00:31
vrurg "To whom how" ;)
I'm usually anti-op too, but here I just like it.
AlexDaniel it's cute
but unreadable 00:32
vrurg No worse than $a (-) $b, anyway.
AlexDaniel a bit worse, actually 00:33
I mean, the unicode one kinda sucks too
but otherwise it's a well-known notation
and your example is just an ascii alternative 00:34
and we do have things like (elem) which are ok-ish
also, it's literally set subtraction, so that makes sense too 00:35
vrurg BTW, I hate it's unicode ∖ which with many fonts no different from \
If you need to work with method calls extensively, those will also make sense. ., .+, .- fail if no method found, respectively .?, .?+, .?- don't 00:37
I mean, I didn't want any linenoise. And initially I was more like kidding about .?-, but then it seemed like logical. 00:38
AlexDaniel that's nice, yeah
ok, well, I don't know :) 00:39
vrurg Actually, would there be a non-unicode reverse-order analog for .* – I would use it for shortness. But... 00:40
In either case, the PR is much more than those ops. Unless I miss something, it fully implements github.com/perl6/problem-solving/issues/103 and I didn't belive I could manage it. ;) 00:42
00:45 anatofuz left 00:46 anatofuz joined 00:53 anatofuz left 01:01 lucasb left 01:09 anatofuz joined 01:20 MasterDuke left 01:50 AlexDani` joined 01:51 Kaiepi joined 01:54 AlexDaniel left 02:20 squashable6 left 02:23 squashable6 joined, ChanServ sets mode: +v squashable6 02:51 anatofuz left 02:52 anatofuz joined 02:58 anatofuz left, epony left 03:08 anatofuz joined 03:28 softmoth left 03:35 anatofuz left 03:41 anatofuz joined 03:46 anatofuz left 04:43 ZzZombo joined 05:01 anatofuz joined 05:03 anatofuz left 05:08 ZzZombo left 05:09 ZzZombo joined 05:29 anatofuz joined 05:33 anatofuz left 05:41 anatofuz joined 05:45 anatofuz left 06:14 ZzZombo_ joined 06:16 ZzZombo left, ZzZombo_ is now known as ZzZombo 06:20 anatofuz joined 06:27 anatofuz left 06:37 epony joined 06:56 anatofuz joined 07:36 anatofuz left 07:37 anatofuz joined 07:42 anatofuz left 07:47 anatofuz joined 07:52 anatofuz left, epony left, anatofuz joined 07:54 epony joined 08:00 anatofuz left 08:02 anatofuz joined, anatofuz left 08:10 anatofuz joined 08:20 anatofuz left 08:21 anatofuz joined 08:36 ZzZombo_ joined 08:39 ZzZombo left, ZzZombo_ is now known as ZzZombo
nine On the one hand one needs to be very careful to not have things like Failures serialized, on the other hand I constantly run into "Cannot invoke this object (REPR: Null; VMNull)" 08:43
Even more frustrating since the things that definitely cannot be called are subs like "say" and "note", which robs me of my primary debug tools 08:44
08:46 ZzZombo_ joined 08:48 ZzZombo left, ZzZombo_ is now known as ZzZombo 08:50 anatofuz left, anatofuz joined
nine Apparently there's an issue with EXPORT here. Can be easily reproduced with: echo 'sub EXPORT() { Map.new("&foo" => sub () { note "foo!" }) }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test' 08:54
Same happens with: echo 'my package EXPORT::DEFAULT { our sub foo() { note "foo!" } }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test' 09:01
This however works just fine: echo 'our sub foo() is export { note "foo!" }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test' 09:02
As does this: echo 'my package EXPORT::DEFAULT { }; BEGIN EXPORT::DEFAULT.WHO<&foo> := our sub () { note "foo!" }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test' 09:08
Note that the BEGIN is essential in this case
And a "no precompilation;" fixes all cases 09:09
09:14 epony left 09:15 epony joined 09:19 anatofuz left 09:29 ZzZombo left 09:41 anatofuz joined 09:45 anatofuz left 09:59 anatofuz joined 10:03 ZzZombo joined 10:04 anatofuz left 10:24 softmoth joined, AlexDani` is now known as AlexDaniel, AlexDaniel left, AlexDaniel joined 11:43 squashable6 left 11:47 squashable6 joined, ChanServ sets mode: +v squashable6 13:50 vrurg_ joined 13:53 eaterof joined, tony-o_ joined, DrForr_ joined 13:55 literal_ joined, eater left, tony-o left, vrurg left, kawaii left, literal left, DrForr left, [Coke] left, mornfall left, vrurg_ is now known as vrurg, [Coke] joined, [Coke] left, [Coke] joined, kawaii joined 13:58 mornfall joined, epony left 14:04 lucasb joined
[Coke] . 14:32
|Tux| Rakudo version 2019.07.1-386-g7e76762c5 - MoarVM version 2019.07.1-257-g39f577438
csv-ip5xs0.754 - 0.783
csv-ip5xs-206.420 - 6.566
csv-parser20.994 - 21.812
csv-test-xs-200.428 - 0.429
test7.153 - 7.321
test-t1.709 - 1.735
test-t --race0.775 - 0.777
test-t-2029.505 - 30.145
test-t-20 --race8.888 - 9.053
 14:51
nine Wait....what?! 15:42
nine@sphinx:~/test/precomp12> echo 'BEGIN Rakudo::Internals.EXPORT_SYMBOL(q<&foo>, ["DEFAULT"], sub () { note "foo!" })' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test'
foo!
nine@sphinx:~/test/precomp12> echo 'BEGIN { Rakudo::Internals.EXPORT_SYMBOL(q<&foo>, ["DEFAULT"], sub () { note "foo!" }) }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test'
Cannot invoke this object (REPR: Null; VMNull)
TreyHarris statement BEGIN vs block BEGIN would differ in this case in... hm, they should be semantically equivalent in the end. but they aren't semantically equivalent in terms of what the phaser must do, right? Because of the closure? 16:14
nine Apparently, everything's fine as long as an exported sub is declared in the top level lexical scope of the comp unit.
But then, so is BEGIN { sub foo() is export { note "foo!" } } 16:17
TreyHarris I am right here, yeah? Been awhile since I've looked at rakudo internals, but phaser BLOCK vs phaser STATEMENT do actually work by constructing a closure of the block and executing it, it's not some special-case macro thing that immediately ingests the block and treats it just like a phaser statement?
nine AFAIK yes 16:18
16:34 patrickb joined
nine Looking at the block vs. statement versions with --target=ast, I see definitely more fixup code for the statement version. 16:55
Ok, I'm beginning to form first vague suspicions 17:01
There's this comment in Perl6::Actions for begin_time_lexical_fixup: 17:02
When code runs at BEGIN time, such as role bodies and BEGIN blocks, we need to ensure we get lexical outers fixed up properly when deserializing after pre-comp. To do this we make a list of closures, which each point to the outer context. These survive serialization and thus point at what # has to be fixed up
I think these fixups are not propagated down to nested blocks. And they are missing entirely for EXPORT subs which are also run at BEGIN time (of the caller) 17:03
TreyHarris that sounds rather plausible 17:04
17:04 epony joined
nine Yes, yes indeed! The issue is definitely limited to subroutines in lexical scope. Calling methods OTOH is just fine. 17:22
I can get it to work by turning this: [flat $self, |@args] into: List.new($self, @args.Slip).flat.Array 17:23
Like this example demonstrates, it's just hard to find all the implicit subroutine calls in Perl 6 code... 17:24
17:34 looker joined 17:37 looker left
vrurg nine: can you get it fixed? 17:44
nine vrurg: given enough time...probably. Do I have that time? That's the more difficult question 17:48
vrurg :( :( :( It's a kind of issue affecting lots of code around. I'm not well-enough known with with area to do something about it. 17:50
github.com/rakudo/rakudo/issues?ut...en+closure 17:51
nine vrurg: given your endeavours into the MOP and role composition, I don't doubt your ability to fix something like this. It's just a different area that you'd have to dig into. My whole understanding of this part of the code comes from making BEGIN time EVAL work. Which I started without knowing anything really. 17:53
It usually comes down to time, effort and patience 17:54
vrurg nine: I was able to track similar case down to a lost closure long ago. But no more. Serialization is still some magic for me. ;) 17:56
Maybe you're right and it's time to start digging this fields...
nine And yes, this area really could use some love. Our support for BEGIN time code really feels half baked still 17:57
vrurg *field
My biggest problem is lack of understanding how exactly contexts/frames work. No docs, no papers... Tracing MOP is much easier because it's mostly introspectable. 17:58
nine No docs, no papers, but you got the implementation instead :) 18:05
Serialization is actually not that magical. There's a serialization context for the currently compiled comp unit. That's just a pile that objects get added to (via $*W.add_object or $*W.add_object_if_no_sc mostly). At the end of compilation, this pile is dumped into the precomp file. 18:10
What makes issues surrounding serialization so hard to debug is usually just that the objects on this pile don't have any connection anymore to the code that caused their addition. So it's hard to say, _why_ something got serialized. 18:11
18:12 epony left 18:17 epony joined
nine And just for completeness: there's also something called "repossession". This comes into play, when during compilation, we load an object from another comp unit (read: from a module we loaded) and this object gets modified in some way. We then have to dump this modified version into the precomp file as well, even if it didn't originate in this comp unit. 18:18
vrurg nine: Thanks! That sorts out some bits in my head. 18:19
afk&
nine So the current comp unit sort of takes possession of the modified object. The VM tracks modifications to objects from other serialization contexts (modules) and does this repossesion thing automatically for us. In some cases we don't want that and that's what the obscure scwb ops are about. 18:20
18:31 MasterDuke joined 19:04 Kaiepi left 19:24 patrickb left 19:26 committable6 left 19:27 committable6 joined
vrurg nine: perhaps, you've convinced me to give it a try. R#2897 is a good and simple candidate for the starting point. 20:05
synopsebot R#2897 [open]: github.com/rakudo/rakudo/issues/2897 VMNull in add_method in indirect use
20:06 leont joined
nine vrurg: yeah, looks exactly like the issue with echo 'sub EXPORT() { Map.new("&foo" => sub () { note "foo!" }) }' > A.pm6 && echo 'use A; our sub test() is export { foo }' > B.pm6 && perl6 -I. -e 'use B; test' 20:36
21:03 Kaiepi joined
Geth_ rakudo: patzim++ created pull request #3224:
Add BuildTools installation assistant		 21:10
rakudo: f7b4d71f46 | (Patrick Böker)++ | 2 files
Add BuildTools installation assistant

Add a script to help with installing the MSVC compiler needed to install binary packages on Windows. Also adapt the documentation to explain when a compiler is needed and how to install it.		 21:12
rakudo: 1fef78388c | (Patrick Böker)++ (committed using GitHub Web editor) | 2 files
Merge pull request #3224 from patzim/msvc-installer

Add BuildTools installation assistant
21:16 leont left 21:45 leont joined 22:09 anatofuz joined 22:17 anatofuz left 22:37 anatofuz joined 22:39 softmoth left 22:44 anatofuz left 23:16 Kaiepi left 23:33 lucasb left 23:58 leont left

Powered by Cro and the Raku® Programming Language.

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

Thank you!