% IllvilJa has left IllvilJa!~jilves@emea-netcache1.oracle.co.uk % c9s has joined #parrot % darbelo has left #parrot % IllvilJa has joined #parrot % Dave has left Dave!~dave@pool-141-150-16-73.mad.east.verizon.net % Davemon has joined #parrot % Davemon is now known as Dave % contingencyplan has left contingencyplan!~contingen@cpe-76-186-27-146.tx.res.rr.com % ruoso has joined #parrot % AndyA_ has joined #parrot % AndyA has left AndyA!~andy@062249178237.customer.alfanett.no % AndyA_ has left AndyA_!~andy@062249178237.customer.alfanett.no % iblechbot has joined #parrot % tetragon has joined #parrot % tetragon has left tetragon!~seneca@CPE0040d001f62f-CM000a736592a8.cpe.net.cable.rogers.com % mire__ has joined #parrot % skids has left skids!~bri@c-71-233-204-100.hsd1.ma.comcast.net % rdice has joined #parrot jonathan, the fpw is in the south of France, and has been decided quite late so it will be a small fpw, so they have place for any good presentation even at a lame time. The attendance may be small but there will be the same regulars who are delighted to see you and hear your talks % IllvilJa has left IllvilJa!~jilves@emea-netcache1.oracle.co.uk Infinoid++ thus commences the first (reported) smoke on netbsd diakopter++ I guess make -j doesn't work sorry, that wasn't helpful. make -j; outputted the syntax for calling make. says it expects [-j numJobs] or something [-j max_jobs] ok so, does "make -j 4" work on netbsd? :) it's smoking now... I did make -j ; make smoke (should've done make -j && make smoke) cognominal ~~ s/lame/late/ make -j smoke the smoke rule depends on the build rules, so it should all work out with one make there are so many undocumented dependencies when installing Test::TAP::HTMLMatrix via cpan % ambs has joined #parrot Test::TAP::Model depends on Petal and Test::Harness::Straps, neither of which is installed by cpan before it attempts to install HTMLMatrix hmm netbsd smoke: http://tinyurl.com/3vnuqf diakopter: hurm. I don't recall having that issue. :| did you report the bug? OBTW, chances are good that parrot will be getting a smolder setup. I read that on use.perl Test::Harness::Straps isn't part of the perl install? not on every vm I've set up diakopter: O RLY? YA RLY they must have split it up and expected you to install another port for it, then kinda like freebsd does with php. CPAN wouldn't have known about that :( centos5.1, debianSid, solaris10, netbsd4 because of that, the Makefile.PLs shouldn't assume the perl core is complete..? well, Test::Harness::Straps is part of the default install for perl 5.8.8 and 5.10.0 if you didn't get it, the install wasn't complete is this only happening on netbsd4, or elsewhere too? I know it happened on sol10, and I know it happened on one of the other two, but I don't recall which I could check the file times to be sure hmm sol10 came with perl 5.8.4 the linuces had 5.8.8 Test::Harness::Straps is in 5.8.4 too... http://search.cpan.org/src/NWCLARK/perl-5.8.4/lib/Test/Harness/Straps.pm Infinoid's url is at http://xrl.us/bi7v8 odd yeah oh, I have an idea; brb never mind Okie dokie purl, mind now! Infinoid: excuse me? anyway, how do I get the output of these (2) failing tests on netbsd (3) I mean oh, they're just codingstd if you run them under "prove -v", or just under "perl", you will see their full diagnostic output nopaste? nopaste is, like, at http://nopaste.snit.ch/ (ask TonyC for new channels) or http://rafb.net/paste or http://paste.husk.org/ or http://poundperl.pastebin.com/ or http://paste.scsys.co.uk/ or don't bother me while I'm eating "diakopter" at 209.9.237.164 pasted "./t/codingstd/c_cppcomments.t" (11 lines) at http://nopaste.snit.ch/12671 diakopter: huh. that should be telling you the file that it failed on. cool, looks like the test errored out. Can you try "perl t/codingstd/c_cppcomments.t"? (and that passes for me here.) Segmentation fault (core dumped) awesome diakopter: what version of perl do you have? This is perl, v5.8.8 built for i386-netbsd-thread-multi are the other tests failing in the same way? yep. c_operator outputs 1..1 and c_parens outputs 1..3 and c_cppcomments outputs 1..1 hmm. I think it has to be a perl bug next I was going to try NetBSD4.0/amd64 we'll see how that goes I guess % skids has joined #parrot % ambs has left ambs!~ambs@siglab.di.uminho.pt someone just told me they're giving me an UltraSparc5 with Solaris10 on it diakopter: awesome. eh, I think I'll wait until Sol10 2008-05 is available April 16 to do the 64-bit one. okay, I guess I'll start on some Windows ones since those are fun, and MSFT is unlikely to want to smoke all the combinations % gryphon has joined #parrot there's 32-bit vs 64-bit, there's 2003, 2008, XP SP1,SP2,SP3, Vista SP0,SP1, there's strawberry perl, activestate perl, microsoft(!) perl, cygwin perl, where do I start (augh)? and there's MSVC, cygwin, and mingw and strawberry perl. it has been said that strawberry perl is found at http://strawberryperl.com/ (since that has a compiler built in.) % Andy has joined #parrot yeah, strawberry implies mingw hey, I mentioned cygwin and strawberry, and I figured mingw was covered under strawberry you're right :) % contingencyplan has joined #parrot now, which ones will Microsoft likely prefer to smoke I'm guessing VC++ and Microsoft perl or perhaps ASperl what is "microsoft perl" ? what used to be called Services For Unix microsoft bundles a distribution of perl with windows server 2003 it's on the second CD Server 2003 R2, I should say. for pre-R2, it was available for download looks like SFU includes ASPerl. (and some other perl, but if has asperl... hm here's the one that includes "UNIX perl": http://www.microsoft.com/downloads/details.aspx?familyid=93ff2201-325e-487f-a398-efde5758c47f&displaylang=en diakopter's url is at http://xrl.us/bi7zm Utilities and SDK for UNIX-Based Applications is an add-on to the Subsystem for UNIX-Based Applications (referred to as SUA, hence forth) component that shipped in Microsoft Windows Vista / Windows Server 2008 RTM. in the least, I could help them set up the smoking environments by preparing some install instructions and smoke scripts * diakopter installs Server 2008 and UaSfUBAforSUA it's likely perl 5.6 does parrot require perl 5.8? yes. but mainly out of laziness. (Given our long term plan to avoid having perl required at all, forcing ourselves to limit ourselves to older perl versions now is too limiting.) does that need to happen before 1.0? (I don't know where it sits on the roadmap.) % uniejo has left uniejo!~uniejo@langebro.adapt.dk the original plan was yes. the current plan doesn't require it, I think. We would need to have some kind of bootstrapping plan to replace it, and while that's theoretically possible, it's better for us to concentrate on VM features. % IllvilJa has joined #parrot yeah, it doesn't sound easy. we're going with perl 5.8 for 1.0 Unless of course someone gets very motivated. diakopter: smoking strawberry would be fabulous mmm, strawberry * pmichaud wonders what the DEA would say about "smoking strawberry" we're likely going to require icu for parrot 1.0, too ....we are? * particle installs icu (finally) yes I was hoping that we'd have the new string code in place before then i.e., the Strings PDD sure. but transcoding ops? transcoding isn't all that difficult, really I mean, what are we going to transcode between? utf8, ascii, and iso-8859-1, yes? sorry, i meant upcase/downcase particle: are these declarations the result of your private conversation with allison this weekend? and grapheme cluster equivalence grapheme cluster equivalence might take more work, yes coke: yes. she should be writing something up ... That would be handy. upcase/downcase isn't even that difficult either the tricky part is properties 'u' :umlaut :uppercase but I've already done (in PHP) upcase/downcase conversion tables for codepoints up to U+055F it's really not that bad and I don't know that there's much upper/lowercase stuff beyond U+055F ok, well, if we can get it done, great! but allison and i are ok with requiring icu for parrot 1.0 if we have to okay, if we can get it done :-) if someone can focus on getting an efficient string implementation for Parrot according to whatever the new strings pdd looks like, I can help with getting the unicode stuff to work w/o icu jonathan++ # pint % peeps[work] has joined #parrot wow, it's 10:40 already think I'll do an early lunch then come back with more pct updates :-) ooh. breakfast. good idea % AndyA has joined #parrot nearly 12 here. bother. % sjansen has joined #parrot % ambs has joined #parrot r26902 | tewk++ | trunk: : [c99] independent C pre-processor diff: http://www.parrotvm.org/svn/parrot/revision?rev=26902 r26903 | tewk++ | trunk: : [MANIFEST] diff: http://www.parrotvm.org/svn/parrot/revision?rev=26903 particle++ # comment about BEGIN blocks on p6l "cognominal" at 82.67.232.89 pasted ".perl for Bool, handling of my WHAT for classes derived from PMC" (38 lines) at http://nopaste.snit.ch/12675 can someone review and apply this patch? % AndyA has left AndyA!~andy@062249178237.customer.alfanett.no * diakopter installs Microsoft perl on Win2008 cognominal: $I0 = $P0 is wrong could perhaps use $I0 = istrue $P0 but even better is simply: if $P0 goto true yes (one cannot use numification to decide if a particular value is true/false) as for the 'WHAT' method, I think the 'Ref' type will disappear when we get the copy op conversion done ok waht is the copy op? * Coke supposes he should re-join p6l after a multi year hiatus. given two pmcs, put the guts of $2 into $1. two pmcs of the same type? no, they can be different types essentially, make $1 into a copy of $2 what makes it different from morph morph simply changes an object's type ok currently rakudo ends up doing the equivalent of morph+assign, which is what generates the Ref replacing this with 'copy' will hopefully fix that I'm checking to see what would be needed to convert hi, I'm working on TapTinder http://tt.perl6.cz/ , so no time to pct-html-dump http://mj41.cz/static/parrot/pct-html-dump/output-06/ , patch is for r24855, if anybody has free time ... :-) and sorry for my English again I speaks engrish too :) * pmichaud tries the copy op conversion pmichaud: Erm, you're going to hate me for what I did to assignment in Rakudo. ;-) jonathan: why, what did you do? But it did provide a way to try out the copy op. And a bunch of tests failed with it. :-( I just tried converting to copy op now and it seems to work pmichaud: Since we have to do a type-check on assignments, assignment isn't done through the copy past node any more. at least, it passes the basic rakudo tests A next step was to, when we can statically know there weren't any type constraints, just emit a standard copy node. Which is an optimization. I somewhat think we need/will end up with a Scalar type and that type checking would be done via the Scalar To try this out for real, you'll need to change TYPECHECKEDASSIGN. Well, we can do that route if we have two PMCs per variable - a container PMC and a value inside it. i.e., typing is a property of the container, not of the value s/typing/type checking/ Sure, but at the moment we just have one PMC that is acting kinda like both. well, if I knew how to efficiently write a Scalar PMC, I'd do it. But I don't. :-| Which means you have to check the type and pass it on over assignments. *other anyway, I did expect the assignment code to change for rakudo, yes. OK. Well, I changed it. :-) But I did in passing try the copy op in doing so. I'm not sure I planned for type checking to be this soon in development, but we'll leave it for now so, are you saying copy doesn't work with what is in rakudo now, or ... ? % Theory has joined #parrot If you replace the morph/assign in the type checked assignment routine with copy, you get some test failures. oh, assignment is doing its *own* morph/assign? I was focusing on getting the type checking stuff working rather than trying to trace down those ones. Right, it seemed the neatest implementation. we really need to get rid of morph from rakudo, though See src/builtins/guts.pir Change the morph/assign sequence to a copy in there. jonathan: that doesn't look right okay, will do thanks When I did, I got a few failures. why does the assignee always need a 'vartype' attribute? We don't have a scalar PMC to hold onto the type. Thus we need to attatch it to the value. Somehow. is the type restriction dynamic? Yes. Constraint types (declared with subset) have to be in the general case. I agree with you but don't follow the constraint types/subset example preparing to smoke i586-pc-interix3 Constraint types, declared with the subset keyword, allow any condition to be used to create a type. And trying to type-check those at compile time involves solving the halting problem. Which may delay Perl 6 a little... We may well be able to detect some easy things at compile time. with copy, WHAT indeed works without my patch But for now, fully dynamic checks get us to meet the spec, I believe. jonathan: okay. But this means we really need a Scalar type. I'm afraid if we keep trying to attach things to the values we'll really clutter up the entire implementation especially because there are large number of properties that we'll start to have to keep track of % peepsalot has left peepsalot!~peeps@cpe-67-9-161-48.austin.res.rr.com Well, we can always resolve that by a level of indirection - every value has one place to hang off all of these things. and just reassign it? yeah, that might work % peepsalot has joined #parrot * Coke whispers, "hll interop" Right, just as I'm doing with types, but generalize it to do everything. also, as a general rule I try to make sure all of my attributes have sigils (e.g., 'vartype') *but*, it doesn't win us much in memory usage. Because then we have another PMC holding the "hung off" data. Well, it does if there's no data to put there, I guess. OK, can change that. % mj41 has left mj41!chatzilla@pc-jurosz.ro.vutbr.cz I think in the long run we're going to end up effectively having that extra PMC around anyway it might not _be_ a PMC, but it'll certainly act a lot like one :) having a Scalar type would also be nicer because then we can make type and other constraint checking a part of the copy op instead of a separately called sub (constraint types) can we do this with a hook on setting the value of the pmc that invokes some method? Coke: that'd be nice to have I think having a Scalar type is the neater implementation. (that would address one of the requirements tcl has.) or, for that matter, why not just override the copy op for Perl6Object ? You can't override ops? we can't? not sfaik. Ops aren't MMD at that level. you can add a new dynamic op. Only register types. INSP I'm confused INSP? INSP is "integer, number, string, pmc" Right, the different register types. ops *can* be mmd, though see, for example, cmp_str ops definitely can be mmd they're not fun to debug that way, but they do work ... it's not the same mechanism, izzit? and is 'copy' just an op or is it also a vtable entry? No, it's just an op pity. It calls multiple (at least two) v-table entries. cmp_str isn't "an mmd op", it's "an op that invokes mmd". okay, then could copy become 'an op that invokes mmd'? sure. The copy op does 3 v-table method calls. It's quite a scary op... It doesn't actually call any kind of set_pmc or assign_pmc vtable methods. it's not supposed to. :-) Right, but I'm not seeing anywhere we can hook it. well, maybe we just need to bite-the-bullet and develop a Scalar PMC type I fear so. Wonder how hard this will be to get right... :-) unfortunately, I have no clue where to begin. There still isn't a good "here's how to create a new PMC" document that I'm aware of all of the PMCs I've created are just cargo-culted (bugs and all) from other implementations 'ack -a pmclass docs' is probably your best bet atm. i designed exporter pmc from scratch doesn't use vtable functions though a scalar pmc should really take on the vtable of its contents, except for a few special operations didn't audrey start a Scalar pmc for pugs? (audrey) yes, but I could never quite understand it or get it to work like I thought it should what is the real_self in a PMC? read pdd17_pmc fwiw, if we have a real Scalar PMC, then the operation should become 'assign' instead of 'copy' cognominal: If you have an instance of a PMC, but actually it's been subclassed by a high level object, this stores the subclass, so we can dispatch calls on the Right Thing. it seems that the copy makes a lot of stuff shared between the copy and the copeid object pmichaud: Agree. I can have a crack at it, perhaps. jonathan: I'd really appreciate that otherwise I think the rakudo code is going to get progressively messier in this area Yeah. pmclass Scalar provides scalar { ... } /* first cut */ (which is one reason why I wasn't in a rush to do it -- I wanted to take care of 'cleaner' things first :-) As per usual, my first implementation of a feature reveals the way not to do it. anti-patterns++ * jonathan has a feeling getting the Scalar PMC right is going to hurt this also tells me that what I was planning to use to test the copy opcode isn't really going to test it :-| I have a dread that we're goign to finish two different languages and then have to redo them to get HLL interop to work in any sane fashion. :| coke: I think that's a near certainty no matter what we do Coke: Scalar PMC should give us a better change of success on interop with Perl 6. can we write some tests for Scalar PMC before we write the pmc? % mj41 has joined #parrot first crack at a scalar PMC: (1) assign should cause the scalar in $1 to reference a COW version of $2 (2) all other operations should be delegated to the reference if COW isn't possible, then a copy is okay Let's do a copy first. trickier is figuring out how to do a 'bind' operation And optimize later. Well, if you bind to a value, then it's ro It's just assignment of the PMC reference, right? ah, that's the other problem darn $a[5] = 4; currently uses "copy" aw drat; interix6.0 has perl5.6.1. oh well. on to strawberry. strawberry++ * ambs licks his lips a lot of this stuff seemed more doable to me when we had setprop/getprop attributes don't work for you? sure, if we're giving every value in rakudo a set of attributes but setprop/getprop didn't really copy from one pmc to the next, iirc whereas attributes do i.e., PMC acted more like containers than values *PMCs Right, and the properties stayed the same when you changed the type/value. exactly That's exactly what I've gone and faked up. :-) But more seriously, we need to allow people to write their own container classes too. jonathan: unless you have a breakthrough idea, I'd hold off on Scalar for now my $x is Persistent; let's just give every object a %!properties attribute or something and we can store constraints in there And copy it on assign? well, reference reference is now stringified or perhaps I should say "restore" restore is the word I think I used in my comment ;-) i.e., make sure that assignment doesn't destroy the PMCs %!properties OK, so that's just taking what I did already and generalizing it. but I'd still prefer to see it handled as part of a copy operation we can generalize it with a role does rScalar I think that such an op would need to be so specific to Perl 6 it'd be a dynop. or rProperties or whatever I'd really prefer to find a way to get copy $P0, $P1 to dtrt particle: (role) does something that has to be on every PMC really need to be a role? gotta get allison involved on that sorry, I mis-stated Well, I can hack it to keep any %!properties around, but it will be a Perl 6 specific op then... I _really_ wish that assign would dtrt pmichaud: it's a matter of composition strategy % Theory has left Theory!~Theory@c-76-115-2-225.hsd1.or.comcast.net oh, hey, is assign a vtable entry? we have roles for a reason. reuse. yes oh, then let's just overload ' 'assign' and use that 'assign_string_native' and 'assign_pmc' % ambs has left ambs!~ambs@255.92.54.77.rev.vodafone.pt we're unlikely to use 'assign_string_native' in rakudo assign can be built in terms of the 'copy' op Hmm, interesting idea. i heard interesting idea was to put two of the bots in the same channel, from different game servers so, whenever we want to do assignment, we generate an 'assign' instruction, and our overloaded 'assign' (for all rakudo PMC values) will do a copy and check any constraints and make sure the copy retains the constraints Worth a try. other languages wanting to hll interop with rakudo also use 'assign', which dtrt Right, and PCT would not emit a morph and assign any more, just an assign? correct er, not exactly How will that impact other HLLs using assign? Rakudo would generate assignments as assignment ops i.e., it'll just do :pirop('assign') instead of :pasttype('copy') Oh, OK. So we add a third pasttype? no bind, copy, and now assign? no, assign is a PIR operation oh pirop *right* Sorry, I thought they were both pasttype Inverse typo. it doesn't affect other HLLs unless they happen to call 'assign' on a rakudo PMC In which case, it should do the Right Thing. exactly. OK. So which one of us gets to try this? :-) whoever gets to it first? ;-) I'd rather we don't both try at the same time. pmichaud: (assign, HLL), which is entirely possible, if you generate a PMC and pass it in to some other HLL. well, I'm not doing it soon. I'm happy to know that we have *a* potential solution without needing a ton of change er, to a method defined in some other HLL. if I start on the assign op version I'll let you know first otherwise feel free to do it whenever you feel like getting around to it % ruoso has left ruoso!~ruoso@195.23.92.2 including not doing it at all anytime soon I'm kinda keen to get this worked out, just because until we have a solution I'll feel uncomfortable building a load of stuff on top of it. Just feels like one of those fundemental things we should try and get right early. then feel free to start. I don't plan to be doing anything anytime soon (i.e., next seven days) that might interfere on it OK, I will see what I can do. most of what I expect to do will be code review and finding things like this that I can carp about :-) Finding Rakudo hacking vaguely addictive at the moment. and I'm very happy to be feeding that addiction :-) % barney has joined #parrot I agree, it's a fundamental thing I'd like to get right early * PerlJam hopes someone is saving this conversation. Review of what I'm doing is certainly a Good Thing. The last few lines from pmichaud and jonathan seem like they should be on a slide for someone's YASPC talk :) er, s/YAS/YA/ perljam: so, stick it in your talk! What... you didn't submit one?!?! PerlJam: which lines, exactly? I'll store them away for my talk or, even better, I'll put them in a use.perl article pmichaud: the addictive/feeding lines especially :) OK, I'm hungry. Time to go cook something. % rdice has left rdice!~richard_d@CPE0014bfafbbd5-CM0011e6ecf48a.cpe.net.cable.rogers.com (hll interop) -- yes, I think we'll end up having to do some tweaking to get hll interop to work between some of the languages % rdice has joined #parrot "diakopter" at 72.202.142.227 pasted "strawberry failures so far" (123 lines) at http://nopaste.snit.ch/12676 r26904 | bernhard++ | trunk: