»ö« | perl6-projects.org/ | nopaste: sial.org/pbot/perl6 | evalbot: 'perl6: say 3;' | irclog: irc.pugscode.org/ | UTF-8 is your friend!
Set by Tene on 14 May 2009.
jnthn lichtkind: The problem is that the class is globally modified. 00:04
In place.
lichtkind which is unclean
jnthn A form of action at a distance.
Aye. 00:05
OK, sleep for me...
dobru noc
00:11 omega joined
omega hms, did I flood? 00:12
lambdabot omega: You have 1 new message. '/msg lambdabot @messages' to read it.
omega ahh, it seems not :p
00:14 amoc left 00:16 nihiliad1 joined
lichtkind dobrou 00:20
00:20 lichtkind left 00:21 nihiliad1 left 00:24 amoc joined 00:26 nihiliad left
snarkyboojum laku noc even 00:37
or dobar noc
00:49 nihiliad joined 00:50 kate21de1 left 00:55 unitxt_ joined, nihiliad left 01:01 eternaleye left 01:02 FurnaceBoy joined
FurnaceBoy The requested URL /pugs/docs/01Overview.html was not found on this server. 01:02
01:05 amoc left 01:06 unitxt left
TimToady looks like it should be docs/Pugs/01Overview.html 01:09
sjohnson hot day today 01:12
01:16 amoc joined 01:20 eternaleye joined 01:23 amoc left, amoc joined 01:27 Chillance joined 01:31 blas left
DanielC S22 says that YAML will be "available standard in Perl 6". What Synopsis talks about this? 01:34
JDlugosz Hi guys. 01:35
01:47 unitxt joined 01:54 synth left 01:57 unitxt_ left 02:04 DemoFreak left 02:06 justatheory joined
DanielC @tell pmichaud The problem with the k-nucleotide benchmark is definitely the ++. If you replace for loop by a "loop ($i=0; ...; $i+=1)" and change the "%table{...}++" to "+=1" the program runs 3 times faster. 02:10
lambdabot Consider it noted.
02:13 azawawi joined
azawawi hi 02:13
02:13 frew|work joined
DanielC o/ 02:15
02:17 DanielC is now known as DanielC_zzzz, jonathanturner joined 02:20 DanielC_zzzz left, justatheory left 02:23 araujo left 02:24 araujo joined 02:25 araujo left 02:26 araujo joined 02:31 synth joined, synth left 02:43 araujo left 02:44 araujo joined, araujo left 02:45 araujo joined, araujo left 02:46 araujo joined, araujo left, snarkyboojum left, azawawi left 02:47 araujo joined 02:49 araujo left 02:50 meppl left 02:54 alanhaggai joined 02:56 araujo joined 03:01 wknight8111 joined 03:07 Chillance left 03:11 alanhaggai left 03:38 justatheory joined 03:40 agentzh left, agentzh joined 03:44 wknight8111 left 03:50 xinming_ is now known as xinming 04:18 snarkyboojum joined 04:19 FurnaceBoy left 04:29 cotto left 04:46 cotto joined 04:56 justatheory left 05:01 charsbar_ joined 05:10 snarkyboojum left 05:11 fridim_ joined 05:13 cotto left
pugs_svn r26975 | lwall++ | [Cursor] switch to specced action api (tag separate from method name) 05:18
05:32 cotto joined 05:43 kate21de joined 05:52 xinming_ joined 05:59 brunov left
pugs_svn r26976 | lwall++ | [S11] introduce declarators "need" and "defines", components of "use" 06:09
06:09 xinming left 06:12 snarkyboojum joined
JDlugosz anybody around? 06:13
06:15 unitxt left 06:17 unitxt joined 06:22 alexn_org joined
Tene JDlugosz: I am. 06:38
JDlugosz I don't think I've run into you before.
Tene I've seen you around a bit. I haven't been very vocal lately. 06:39
Busy with work and the girlfriend.
JDlugosz Been with Perl 6 long? 06:40
Tene about two or three years, I think.
JDlugosz What have you worked on or been involved with? 06:41
Tene so, depends on what you cpmare it to. :)
Rakudo, Parrot, Web.pm
06:42 ejs1 joined
JDlugosz Well, I read the Apocolypses when they came out one at a time, and figured it would be available soon. Sooner than this! It was a little over a year ago that I was disapointed in the progress to date. 06:42
Tene I drift around... I did a lot of exceptions stuff a while back. Recently I've been focusing on HLL interop.
Got Rakudo working together (kinda) with Cardinal, the Ruby compiler I wrote a year ago. 06:43
JDlugosz When Kodak closed, I took a sabatical and decided to make a serious contribution.
Tene What are you interested in working on?
JDlugosz What's the attraction of Ruby? Years ago it seemed to not hold up to Perl.
Tene I wrote a ruby compiler because nobody else had written one and I didn't know ruby. 06:44
I think. It was long enough ago that I don't remember my motivations too well.
It's been pretty abandoned for a while now, but it was an easy target for me to add HLL interop stuff to.
JDlugosz My ongoing interest is to develop a real formal specifcation. I'm also interested in type theory and the new "roles" idea, and ramifications of generics and interplay between typed and untyped code. Whew!
Care to read what I'm working on now? 06:45
Tene I'll take a look at it, and read it if it catches my eye. 06:46
JDlugosz <www.dlugosz.com/Perl6/web/info-mode...C.html> 06:47
Tene Ah, yes, I saw this on the ml.
06:47 eMaX joined
JDlugosz HLL interop is tricky business. Look at the hype of C#. They claim "write in any language" but they mean Java with your choice of syntax. You have to program to their wimpy capabilities, and totally loose what makes a language unique or powerful. 06:48
06:48 alexn_org left
Tene Yes, very tricky. 06:50
My attempts right now are definitely wrong.
I spent a long time thinking about it, and didn't really get anywhere at all, so now I'm trying to implement *something* in hopes that it will either help me work out something better or attract the interest of someone who can do a better job. 06:51
JDlugosz Ever use a program called Inkscape?
Tene Occasionally.
JDlugosz Not what it's cracked up to be, it seems. Too brittle, strange bugs, etc. 06:52
Tene Yeah, sounds familiar.
06:52 cotto left
JDlugosz Windows that don't behave, browse dialogs that are non-standard and don't remember the path you wanted... 06:52
I just discovered I lost some work, from a file I was fighting to copy/paste from. 06:53
Had to google discussions to figure out how to copy from one document to another!
Tene Sounds like some of my first HLL interop attempts. :) 06:54
06:54 LadyLunacy joined
JDlugosz the open dialog shows all the files twice. 06:54
Tene Interesting. 06:55
JDlugosz You might find my type stuff interesting. 06:57
06:59 cotto joined 07:01 skids left 07:05 viklund joined 07:09 LadyLuna1y left
Tene Nice... I now have PIR library loading working from Rakudo and Cardinal. 07:26
I need to figure out the makefile rules for this, though...
07:26 alexn_org joined 07:36 unitxt left 07:37 unitxt joined
JDlugosz PIR? 07:51
Tene Parrot assembly. 07:52
JDlugosz That's what I thought. Just misparsed your grammar there. Thought it was for Rakudo and for something else. 07:53
So what do you think of F-Bounded Polymorphism? Imagine getting that to interoperate with any lesser language! 07:54
07:54 ejs1 left
Tene use Foo:lang<parrot>; 07:54
JDlugosz I mean, how can Java for example use a class that doesn't follow it's "isa" relationship rules? 07:55
08:12 alexn_org left 08:15 iblechbot joined 08:20 unitxt left 08:36 M_o_C joined 08:38 ejs left 08:53 alexn_org joined 09:00 tulcod joined 09:02 abra joined 09:15 eMaX left 09:17 agentzh left, alexn_org left 09:24 DemoFreak joined 09:35 agentzh joined 09:43 fridim_ left 09:44 viklund left 09:57 viklund joined 09:59 pmurias joined 10:10 sjohnson left 10:13 amoc left 10:15 amoc joined 10:17 frew|work left 10:21 M_o_C left 10:23 sjohnson joined 10:36 iblechbot left 10:37 DanielC joined
DanielC moin 10:40
10:49 snarkyboojum left 10:54 amoc left 10:56 amoc joined 11:06 kane__ joined 11:13 masak joined
masak happy summer salutations, #perl6 people. 11:14
11:15 elmex_ joined
DanielC waves 11:15
11:15 Whiteknight joined 11:17 cotto left
masak hm, it all comes down to module recognition in Rakudo. 11:18
everything else comes after that.
11:18 meppl joined
DanielC What comes down to module recognition? 11:18
masak DanielC: module management, and CPAN or a CPAN replacement. 11:19
DanielC Rakudo doesn't support modules an packages yet?
11:19 cognominal left
masak DanielC: in my opinion, the discussion we're having about that is very useful, but it won't produce much code, because Rakudo doesn't implement modules as per S11 yet. 11:19
and won't for at least a few months.
DanielC :-(
masak DanielC: it does support modules, but not as per S11. 11:20
DanielC I was hoping for something tangible. You know.. round tuits and all.
masak that's what I'd like too.
that's why I'm thinking about what's blocking us right now.
11:22 elmex_ is now known as elmex
masak questions like "what's the least amount of coding that would allow us to store two modules with the same name, but with different versions? or different authors?" 11:22
at the very least, we could write some spectests for this. that would definitely help.
11:22 explorer__ joined
DanielC Daniel Ruoso says that the package format is the most critical step, so perhaps it is possible to make progress there. But I'm no expert on package formats. I just suggested some improvements on S22. 11:23
DanielC hopes they are improvements
masak I'm by no means an expert either. in what way does the package format affect the way we store modules so that Rakudo can access them? 11:24
DanielC I don't know.
I don't know a lot about either subject. 11:25
masak the big step for Rakudo, I think, is that modules might not be able to sit directily on the disk any more.
they might have to be stored in a database somewhere.
that's a big change.
DanielC Hmm... that doesn't sound like "the Unix way" or "the Perl way".
masak indeed not. 11:26
an alternative (perhaps) is something like URL encoding.
DanielC Why can't we use regular directories to organize modules? e.g. /<auth>/<version>/Foo/Bar.pm 11:27
masak DanielC: because modules can have arbitrary Unicode names.
and not all file systems appreciate that.
11:28 unitxt joined
DanielC That doesn't sound like a big problem. Just use some standard encoding like the URL encoding you said. 11:28
masak yes.
but it's exactly these small problems that I think we should start with.
they are concrete, they can be solved today, even before Rakudo gets there, and they can be made to work and show us what the next set of real problems are. 11:29
DanielC Most of the time, for most practical purposes, the module names are going to be ascii anyways.
masak sometimes amrchair discussions about frameworks bore me.
DanielC And even when they are not, lots of programming languages have tools for url encoding/decoding so the data would remain very accessible.
masak yes. 11:30
if it's at all practical not to involve a database, I think we should try to stay in the file system.
DanielC yeah
And I agree with picking small concrete problems we can start with today. 11:31
pochi does that mean I can't create a module 動物.pm and copy it into the module path without URL encoding it first?
DanielC pochi: What if the filesystem doesn't support unicode?
pochi what happens if one tries to copy a unicode named file to a non-unicode filesystem anyway? 11:32
DanielC I dunno
Do we know of any non-unicode filesystem? Can we test?
masak pochi: either the encoding would be uniform on all systems, and then you couldn't. or it'd be a matter of local configuration (when building Rakudo, say), and then you could if your system was configured like that. 11:33
DanielC: old DOS systems ought to be non-unicode.
DanielC ext3 seems to have no problem with Unicode. I just created a file called 動物.pm.
pochi masak: the latter sounds good
masak pochi: yes, but it has the drawback that you shouldn't indiscriminately copy Perl 6 libraries from one system to another. 11:34
they might be configured differently.
perhaps not such a large drawback, but still.
DanielC masak: Or Rakudo could try both paths... but then again, that could lead two conflicts (same module stored twice, once as unicode and once as URL encode).
masak DanielC: that sounds like a hack due to lack of planning. :) 11:35
I wouldn't do that if I didn't really have to.
DanielC yeah... it seems to cause a lot of problems. 11:36
Btw, I can think of another reason not to copy Perl 6 libraries around. The first computer could have the library stored as parrot bytecode and the second computer could have an implementation of Perl 6 that doesn't understand Parrot. 11:38
masak indeed.
std: use Dog:ver(Any):auth(Any);
p6eval std 26976: OUTPUT«ok 00:02 35m␤»
masak rakudo: use Dog:ver(Any):auth(Any);
p6eval rakudo 764684: OUTPUT«Can't find ./Dog in @*INC␤in Main (src/gen_setting.pm:438)␤» 11:39
masak hm, but it parses? interesting...
maybe it wouldn't be too far-fetched to fork Rakudo on github and try some stuff for realz.
pochi shouldn't "Any" be "*"?
masak now _that_ thought excites me!
pochi: I got that example from S11.
rakudo: use Dog:<1.2.1>; 11:40
p6eval rakudo 764684: OUTPUT«Can't find ./Dog in @*INC␤in Main (src/gen_setting.pm:438)␤»
masak yep, Rakudo already parses the colonpairs after the module name. 11:41
Rakudo++
phenny: tell mberends I'm toying with the idea of creating a Rakudo branch and adding the ability to load versioned/authored modules. it would lead to questions about module semantics, directory structure, perhaps distribution. what do you think? 11:45
phenny masak: I'll pass that on when mberends is around.
11:45 M_o_C joined 11:47 huma left 11:55 lichtkind joined, payload left 11:59 viklund left
jnthn hi all 12:02
masak jnthn: o/ 12:03
jnthn hej masak :-) 12:04
masak: Happy to see that you're interested in working on getting something solid together on versioned etc libraries. :-) 12:06
masak I am.
jnthn: is there a policy on branching of Rakudo?
jnthn Branches don't affect anyone who isn't explicitly interested in playing with the branch. 12:07
So just go ahead.
masak just goes ahead
lichtkind hello 12:09
is syntax infix:{+} { ... } legal 12:10
masak lichtkind: I wouldn't think so.
DanielC there is a syntax that is *similar*, I just need to remember what it is...
masak lichtkind: the <+> form autoquotes, and that corresponds to {'+'}
lichtkind so you have to use masak true thanks 12:11
DanielC masak: Is infix:<+> the same thing as infix:{'+'} ?
masak DanielC: if the latter is legal, yes.
DanielC and they mean the same...
DanielC can't imagine what else it could mean, but prefers to ask. 12:12
masak well, that's what 'the same thing' implies, that they mean the same. :)
lichtkind no you got it entiely i understood now
DanielC masak: I missed the "if" in your reply, so I parsed your sentence wrong :)
masak phenny: tell mberends github.com/rakudo/rakudo/tree/ver-auth-modules -- jnthn++ gave the go-ahead. :) 12:13
phenny masak: I'll pass that on when mberends is around.
lichtkind is prelude and STD.pm the same thing?
masak lichtkind: no.
lichtkind: prelude is a set of 'core' subs and methods written in Perl 6.
lichtkind: STD.pm is a Perl 6 grammar. 12:14
masak jacks out of the Intertubes for a while
12:14 M_o_C left, masak left
lichtkind masak: that i read but i cant imagine where one thing ends and other begins 12:14
12:14 M_o_C joined 12:16 |MoC| joined, M_o_C left 12:18 Whiteknight left
DanielC masak: On the topic of versions and file systems... do you know of ':' and '.' are valid file name character in most file-systems? The package Foo::Bar version 1.2 could be stored literally as 'Foo::Bar-v1.2.pm' 12:18
masak: Earlier I said /<auth>/<version>/Foo/Bar.pm but clearly that doesn't make any sense. Two unrelated modules with the same version number would end up together. 12:19
12:20 payload joined
jnthn DanielC: No, they're not. 12:21
DanielC: I don't know anywhere . doesn't work, but : is a problem. 12:22
DanielC Alternatively: /<auth>/Foo/Bar-<version>.pm
Avoids the ::
jnthn (On Windows at least.)
DanielC (and still has version after "Bar")
jnthn DanielC: I don't know. I'm not sure if auth, name and version are the only meta-data we'll ever care about.
So something more extensible is probably wise. 12:23
12:24 mikehh left
DanielC jnthn: I can easily imagine us caring about the platform (Perl 6 vs Parrot). 12:24
And given that Perl can even use stuff written in other languages, even today ("use ruby"), we might even have true Ruby modules stored. 12:25
jnthn: I'm having trouble thinking of something extensible that doesn't make the modules inaccessible (e.g. storing them in a database). 12:27
inaccessible = difficult to access to third-party tools, in particular Unix commands like 'find'. 12:28
jnthn Yes because no way can we build the tools we'd need! 12:29
:-S
FWIW, as a Windows user, I don't really care about accessibility via Unix commands.
DanielC jnthn: I was thinking in terms of "the Unix way" (plain text, etc)
One thing I like about Unix/Linux is that the system is generally accessible. I can hack it. 12:30
No registry. Config files are plain text. Things like that.
jnthn You can still do something database-y and keep it generally accessible/hackable.
Think Sqllite. 12:31
*Sqlite
Already lots of languages with bindings to it.
DanielC If Rakudo used a database, sqlite would be my top choice.
It is the most accessible of the databases I know.
jnthn From the (limited) amount of thought I've givne this, it's felt to me that something database-ish is the way to go, and I'd much rather see us use something existing like sqlite than create yet another third party format. 12:32
DanielC absolutely (reusing sqlite). 12:33
Last thing this world needs is a vanity database just for Perl. 12:34
jnthn: If we use sqlite, then sqlite support must be part of the standard library. You can't install with Sqlite support module without sqlite support... 12:35
jnthn DanielC: Aye, though it'd need to be part of Rakudo's rather than Perl 6's. 12:37
DanielC Is it ok for Rakudo to have a standard library that differs from Perl 6?
jnthn I figure all implementations will have some extra little bits specific to their needs. 12:38
And it depends how you define standard library too. 12:39
They'd still have a consistent setting, and I guess that's what really matters.
DanielC I guess that the Rakudo distribution could come with DBI and DBD::SQlite by default... 12:40
jnthn Dunno, this is details... 12:41
I'm mostly hoping that @other will fill in the gaps on this stuff.
DanielC k
jnthn So I don't have to think about it too much. :-)
DanielC :-)
The alternative to including DBD::SQlite is for Rakudo to come with its own, separate support for SQlite, but I can't imagine how that could possibly be useful. 12:42
pmurias is there anything interesting happening on August 6th (other than the trainings)? 12:43
DanielC jnthn: Here is one thing I like about the SQlite idea: It resolves the unicode problem we were discussing earlier (did you see the discussion). 12:44
jnthn DanielC: Glanced over it.
DanielC If you use SQlite you don't have to worry about what characters the filesystem supports.
jnthn pmurias: Is that the day after YAPC::EU?
pmurias: If so - potentially a Parrot Developer Summit also.
DanielC masak: Have you followed this discussion about SQlite? What do you think? 12:45
jnthn pmurias: Beyond that not aware of anything. 12:46
lichtkind macro summe { 3 + 4 } does it return an AST containing 7 or 3+4 ?
12:46 amoc left
lichtkind i mean i i dont quote will it executed? 12:46
DanielC lichtkind: I don't know, but if the compiler is smart, I would hope that it would just use 7. 12:47
12:47 sri_kraih joined
lichtkind DanielC: thanks but i dont mean optimasition but whot syn says 12:47
DanielC Ok. I don't know what the syn says. 12:48
12:49 ZuLuuuuuu joined 12:50 mberends joined
DanielC o/ ZuLuuuuuu mberends 12:50
mberends waves: wow, lurking on the backlog is interesting, as always :) 12:51
phenny mberends: 11:45Z <masak> tell mberends I'm toying with the idea of creating a Rakudo branch and adding the ability to load versioned/authored modules. it would lead to questions about module semantics, directory structure, perhaps distribution. what do you think?
mberends: 12:13Z <masak> tell mberends github.com/rakudo/rakudo/tree/ver-auth-modules -- jnthn++ gave the go-ahead. :)
mberends wahey!
12:53 Chillance joined
pmurias jnthn: yes, that's the day after YAPC::EU 12:53
jnthn: is the Parrot Developer Summit open to all or only parrot developers? 12:54
ZuLuuuuuu hello DanielC 12:55
12:55 sri_kraih_ left
mberends before opting for anything sql-flavoured to manage module metadata, be aware that sql language comptatibility slows the inner data access routines by an order of magnitude. Google avoids SQL and XML for anything performance sensitive, and we definitely need to be that here. 12:55
jnthn pmurias: I'm not 100% sure it will actually happen yet, I didn't yet book my flights etc. If it does, I think it will be mostly for discussing Parrot-level issues and planning Parrot dev generally rather than just stuff specific to Perl 6. I doubt it'd be a secretive-cabal-ish style meeting though...if you're interested in Parrot and want to come I'd not see a probem with that. 12:57
And of course, Parrot issues relavent to Rakudo will certainly be some of what is discussed. 12:58
DanielC mberends: I thought Google used MySQL. 12:59
mberends DanielC: your're joking, right? 13:00
DanielC No, I'm not.
However, I might be wrong.
But hasn't Google even submitted patches to MySQL? 13:01
mberends If Google could use MySQL, they'd have bought the company. But Sun did, instead.
DanielC www.mysql.com/customers/view/?id=555
They are listed as a customer.
www.informationweek.com/news/intern...=199201237 13:03
mberends DanielC: interesting, I shall dig into what they use it for.
13:04 amoc joined
mberends my concern if from another angle, several database engines were superbly fast until they got an SQL language interface on top, that slowed things down considerably. The reason, general purpose accessibility, is not warranted for Rakudo. 13:05
DanielC jnthn: Were you thinking of storing the actual module in SQL or just some reference to a file on disk? (e.g. the name of the file on disk could be a hash sum of its metadata). 13:06
jnthn DanielC: Just meta-data. 13:07
DanielC ok
I think that's the most reasonable. 13:08
mberends DanielC: the SQLite alternative is much faster and more appropriate, but I'm still doubtful about making Rakudo depend on such a complex subsystem
jnthn The other thing to figure into all of this is pre-compilation.
That is, for efficiency the module installation process should probably compile modules to PBC.
DanielC mberends: I understand. masak expressed similar concerns, and so did I. But jnthn makes good points. 13:09
jnthn well, *an* other, not *the* other.
DanielC PBC = Parrot Byte-Code right?
jnthn DanielC: Yes.
Not a big deal, just something to keep in mind.
pmichaud good morning, #perl6 13:10
lambdabot pmichaud: You have 1 new message. '/msg lambdabot @messages' to read it.
pmichaud @messages
lambdabot DanielC said 11h 7s ago: The problem with the k-nucleotide benchmark is definitely the ++. If you replace for loop by a "loop ($i=0; ...; $i+=1)" and change the "%table{...}++" to "+=1" the program
runs 3 times faster.
DanielC yet another thing to think about is that the user must be able to specify multiple sqlite databases (so the can store modules in his home directory)
jnthn mberends, DanielC: I'm not so much attached to Sqlite, more that we don't re-invent a wheel we don't have to.
pmichaud (sqllite versus filesystem) 13:11
mberends DanielC: that would be no problem
DanielC jnthn: ok
pmichaud Last year at YAPC::NA I spent an hour or so talking to a representative from $large_financial_company about module access in Perl 6 (more)
In his company, they absolutely need the capability to have versioned modules
Which was why he cornered me for an hour to discuss it 13:12
One of the points he wanted to make was that the filesystem was almost certainly not fast enough to be able to handle it
mberends jnthn: a Berkeley DB level of database (essentially a disk based hash) would perform far more efficiently
pmichaud at least, the filesystem alone wasn't enough to do it.
mberends but the WinFS disaster should have lessons for us 13:13
jnthn mberends: *nod*
pmichaud This wasn't speculation on his part, it was actual experience. In fact, they came up with their own set of module management for Perl 5 (basically abusing @INC and all sorts of other things) because the facilities Perl 5 and CPAN provide weren't nearly capable of scaling to handle it.
mberends pmichaud: *very* interesting 13:14
DanielC pmichaud: Interesting. But if an SQlite database told you the direct path of the file you need, then it would be fast enough, right?
pmichaud DanielC: Likely, yes.
DanielC pmichaud: Do you still have his contact email? It sounds like he could contribute much to this discussion.
mberends DanielC: SQLite itself will internally be slower that we are likely to expect
13:16 cognominal joined
DanielC mberends: Do you know anything better? 13:16
mberends our own future custom solution ;)
DanielC :)
pmichaud DanielC: Yes, I could likely contact him; OTOH, he and his company wasn't much interested in being publicly known for working on Perl
DanielC pmichaud: Sure, but he can join IRC, we won't know his name or his company. 13:17
pmichaud I might let him know that discussions are taking place and invite him to monitor 13:18
This person was sufficiently high level that I doubt he'd be able to spend much time hanging around on IRC.
DanielC pmichaud: Maybe he can delegate to an underling who knows what they need.
pmichaud Good point.
13:19 cotto joined 13:20 amoc left 13:21 amoc joined
mberends just knowing that the @INC and filesystem solution *can* be improved successfully, is a very important fact. It sets standard Perl 5 up as a target for us to outperform. 13:21
btw, I think that burying our heads in the sand about performance is not going to last much longer, so optimizing is no longer premature in whatever we add to Rakudo. 13:23
13:23 wolverian joined
pmichaud mberends: oh, we're already on it. 13:23
mberends :) I see that too.
pmichaud I suspect that by the end of the day Rakudo performance will have doubled over what it was this time last week. 13:24
(as measured by "time to execute spectests")
jnthn Wow. 13:25
mberends pmichaud++: I know you care a lot about performance.
pmichaud and perhaps even more if jnthn+'s dispatch improvements land today :-)
jnthn pmichaud: I need to find the energy to debug segfaults. 13:26
pmichaud jnthn: no pressure on this end -- was just a comment :-)
jnthn pmichaud: I'm suffering a little with some kind of cold-ish virus that seems to be sapping a lot of my energy at the moment. :-| 13:27
jnthn has slept for 10 hours two nights running...
pmichaud jnthn: ouch, sorry to hear that. Hope things improve very soon. 13:28
mberends jnthn: get well soon, what else can we say?
jnthn Anyway, I'm very much wanting to land the dispatcher soon. :-)
Excluding some exit-time segfaults, I'm down to two test files with issues (one easy, one hard).
So it's *close*. :-) 13:29
(If anyone is curious to know what's taking so long, I've got a 1500+ line diff against head.) 13:30
(Which should give an idea of the order of magnitude of the refactor.)
lichtkind is there any difference between macro summe { '3 + 4' } and macro summe { 3 + 4 } ?
mberends lichtkind: surely just the same as if you placed the macro contents in the code calling the macro, so the first could not be used as an array subscript for example, but the second could. 13:33
bloonix rakudo: say $*OUT "foo"
p6eval rakudo 764684: OUTPUT«Statement not terminated properly at line 2, near "\"foo\""␤in Main (src/gen_setting.pm:0)␤»
bloonix rakudo: say $*ERR "foo" 13:34
p6eval rakudo 764684: OUTPUT«Statement not terminated properly at line 2, near "\"foo\""␤in Main (src/gen_setting.pm:0)␤»
bloonix is the syntax wrong?
lichtkind mberends: so basically would first return the compiled string and seconf 7?
mberends bloonix: yes, you're using Perl 5 style 13:35
lichtkind: yes, I think so
lichtkind mberends: but your not shure ? :)
mberends lichtkind: I mainly understand what I use in Rakudo, and macros have not arrived yet. I haven't read the macro sysnopis recently, that's what I mean. 13:37
lichtkind mberends: i do read but have problems anyway :)
DanielC mberends: SQLite may not be fast, but the database table shouldn't be very large. I can't imagine someone having 10,000 modules installed.
lichtkind mberends: thanks 13:38
bloonix rakudo: $*OUT.say("foo") 13:39
p6eval rakudo 764684: OUTPUT«foo␤»
mberends DanielC: if we specify what goes into and come out of the database, SQL is not part of the requirement
DanielC mberends: ok...
mberends: So you want to use a custom non-SQL database? 13:40
mberends DanielC: not absolutely, but they are likely to be more efficient 13:41
and I'm desperate to find the most efficient solution for the computer here, not for the Rakudo developer ;) 13:42
DanielC The upshot is that it may be hard to write your own script to read the "Rakudo Registry". 13:43
mberends we can discuss abstract designs in terms of tables, columns, indexes and so on.
DanielC: yes. it will be enough for Rakudo to be the interface to this "Registry" 13:44
If anything else can get access, it will be nice but not essential. 13:45
DanielC ok
pmichaud DanielC: fwiw, the person I was talking with did have thousands of modules installed. 13:47
maybe not 10,000, but definitely at least 4-digits. 13:48
DanielC pimchaud: Ok. And since Perl 6 will also have module versions, the same customer might have the full 10,000 in Perl 6.
mberends as masak++ wrote a few hours ago, the full implementation of the 'use' command is the problem in a nutshell. Installers, repositories etc are just supporting infrastructure that is allowed to be less efficient in execution that the 'use' command.
DanielC I wonder how quickly SQLite can go through 10,000 records. 13:49
mberends we'll build some test cases. performance testing is essential here. 13:50
it will help to visualize it at the lowest possible level: how many file I/O's must the perl6 process do for a 'use' command. CPU and memory are ample, filesystems are huge, but drives are slow. 13:55
DanielC yes 13:56
pmichaud fwiw, I should point out that there are three primary reasons why I've been saying that Rakudo wouldn't be attempting module versions until the August time frame 14:01
(1) in terms of "where to put modules" (in the filesystem), that really depends somewhat on being able to work with an installed Parrot, which thus far Rakudo hasn't been able to do. I expect that to change this week. 14:02
lichtkind mberends: but gerrally speaking macros are evaluated at compiletime and result is returned as an AST?
or has anybody else here clues about p6 macros? 14:03
pmichaud More to the point, Parrot imposes its own ideas of where loadable modules should go, and so we (on the Rakudo side) have to keep those considerations in mind. And Parrot hasn't fully defined its API for that yet.
lichtkind mberends: s/gerally/generally/
pmichaud (2) The specification itself needs more brainstorming about what is/is not required. That's been happening this week on #perl6 and the mailing list, which is good. 14:04
mberends lichtkind: I'm going to brush up my macro-fu in the Synopses to answer that ;)
lichtkind mberends: thanks 14:05
14:05 skids joined
mberends buubot: spack macro 14:06
buubot mberends: S02-bits.pod:16 S03-operators.pod:8 S04-control.pod:5 S05-regex.pod:8 S06-routines.pod:36 S09-data.pod:3 S11-modules.pod:6 S12-objects.pod:7 S29-functions.pod:2
pmichaud (3) I hadn't envisioned that managing versioned modules was going to be significant for anyone before August. I can understand that module packaging and distribution might be important, but being able to handle multiple versioned modules didn't seem like it was a likely high priority (Perl 5 lives without it, for example)
14:07 sri_kraih left
mberends pmichaud: we'd better get with the (1) program before bikeshedding then 14:08
pmichaud mberends: well, I'm fine if we start exploring things. 14:09
DanielC I'm surprised that Parrot puts restrictions on where a module can go.
pmichaud DanielC: I was surprised about that too.
DanielC: "restrictions" is perhaps too strong a word -- "expectations" might be better.
i.e., parrot expects languages to store their modules in a subdirectory of the parrot tree 14:10
(which I think probably won't be what ultimately happens)
I think I've planted the seed/idea that each individual compiler should be responsible for its module management, with a common API to handle cross-language module loading, but it's not official yet. 14:11
DanielC pmichaud: Any restrictions on the file name? Can the module Foo::Bar be stored in something like /parrot/perl6/1ffae822021bc
pmichaud DanielC: that's still to be determined, a bit 14:12
from a personal perspective, I find hash keys to be lousy filenames, though.
I'd still like to be able to figure out what a file is from looking at a directory listing.
DanielC I can't say I love them. I'm just thinking of the DB idea. 14:13
pmichaud a DB can still map to a human-understandable name.
DanielC true
mberends the middle road is something like how Windows maps long file names to 8.3
not everything from Redmond is bad 14:14
pmichaud anyway, my personal feeling is that there ought to be several achievable middle steps between "what we have today" and "full S11 support".
DanielC pmichaud: If the module name is Foo::動物.pm we an URL-encode the entire string, append metadata an make that the fie name.
$filename = urlencode("Foo::動物.pm") ~ "-$version-$auth" 14:15
pmichaud well, I'd hope there at least some level of directory hierarchy there
also probably need to url encode $version and $auth, too
and $auth is really long to be putting in a filename, not sure we need that 14:16
DanielC Can two modules have the name but different authority?
pmichaud of course.
But just because I don't like hashes as filenames doesn't mean they should be forbidden at all
14:17 _Chillance_ joined
mberends tokens for $version and $auth would be fine, database mapped 14:17
pmichaud using Web::Foo-1ffae822021bc would be fine for me
I can look at that name and say "oh, that's Web::Foo"
DanielC $filename = urlencode("$module-$version") ~ ShortHash($auth)
yeah
Web::Foo-v1.2.0-1ffae822021bc looks good. 14:18
pmichaud 11:19 <masak> DanielC: in my opinion, the discussion we're having about that is very useful, but it won't produce much code, because Rakudo doesn't implement modules as per S11 yet.
I disagree somewhat with masak++ here, fwiw.
I think we can start producing code before Rakudo has finished S11 14:19
indeed, I think the two have to be designed together
DanielC :-)
pmichaud or at least take both components into consideration
thus my comment about "achievable middle steps" -- that's what we should be aiming for. I'd be very happy to see us start with a completely filesystem-based module loader for a while, until some of the other parameters become known. 14:20
DanielC ok 14:21
mberends rakudo has a ver-auth-modules branch to try things out
pmichaud I saw that, excellent.
At the moment I'd like us to just come up with a consensus opinion about module location, and perhaps a simple module-to-filename mapping. 14:22
(it can be rakudo specific)
DanielC $filename = urlencode("$module-$version"); for (@all_other_metadata) { $filename = ShortHash($_) } <-- this is automatically extensible.
s/$filename = ShortHash($_)/$filename ~= ShortHash($_)/ 14:23
DanielC is assuming that ~= exists
pmichaud my feeling (and I could be wrong about this) is that the demarcation between "what all Perl 6-compliant implementations must do" and "what is reserved to the implementation" is a very fuzzy concept at the moment
as far as module management is concerned 14:24
mberends nod, because it's not specified anyway
pmichaud I think I'm saying that it's not specified because we don't want to commit ourselves too soon to a particular view 14:25
DanielC I like the " Web::Foo-1.2-1ffae21bc" idea. It is transparent, and it doesn't prevent us from adding a database layer later if that's faster than using the file system.
pmichaud Also, I noticed there was some discussion yesterday about how one might specify a module to a command-line installer. 14:26
I think there might be something to be said to using the same syntax that is part of a 'use' statement :-)
pochi could it be a hybrid system? look in db first, if not there then search the fs?
mberends pochi: you could argue equally for the reverse order
pmichaud could it be a hybrid system? Sure, I can imagine that. 14:27
DanielC pochi/mberends: That might miss the performance benefit of using a DB.
mberends in fact the DB will always know everything, but it may not win on performance
pmichaud "cpan install Module::Foo:ver<1.2.*>:auth<cpan:JRANDOM>" 14:28
pochi it would add the module it found on the fs to the db, so that the next time the module is use()'d it would find it in the db.
pmichaud pochi: that assumes write permission to the db
DanielC pmichaud: What would that do? Install *all/any* modules that match 1.2.* ?
pmichaud DanielC: I would think it would use the same criteria as a use statement. 14:29
"cpan install Module::Foo" # give me latest / most appropriate version of Foo
"cpan install Module::Foo:auth<cpan:JRANDOM" # give me latest version of Foo coming from cpan:JRANDOM
mberends DanielC: therefore the latest 1.2.* 14:30
DanielC Sounds good.
pmichaud in other words, we already have a syntax for identifying a module, we ought to (re)use it
DanielC Sounds very good, actually. Very useful.
"give me the latest version of the branch that I know works with my program" 14:31
pmichaud furthermore, that syntax is nicely extensible. :-)
DanielC yes
mberends I like the idea of initially coding what works without a database, and then adding the lightest possible database only when there's a proven need. 14:32
DanielC +1 14:33
14:33 Chillance left
pmichaud there was also some discussion about whether modules should re-use the existing CPAN infrastructure or make a new one 14:34
I don't think it should be an either-or choice
I think whatever is built should be able to interface nicely with CPAN, but not limit us to that.
(that statement, of course, is meant as a "in the long run" sort of thing, not necessarily what we do in the first steps) 14:36
14:36 iblechbot joined
mberends the idea that the existing CPAN is Perl's killer app is so important, it would be folly to ignore. And yes, decentralized repositories are the new archive network. 14:37
pmichaud and there's more than one way to decentralize :-) 14:39
afk for a bit 14:40
lichtkind mberends: true 14:41
mberends: did youz check s05?
mberends as obra_++ wrote on May-28, we are welcome to try to distribute our Perl 6 projects via CPAN, and download/install with our proto tool 14:42
DanielC mberends: I hear that Berkeley DB is faster than SQLite but it does not support SQL.
mberends lichtkind: I'm reading S06 first, it has more relevant info 14:43
skids Has anyone attempted to spec a robust replacement for -w? If it's done soon enough module authors might actually standardize on one warning flag set. And we could add a flag for "this is inefficient" so when people complain about performance, the fist question is "what does it say when you turn efficiency warnings on?"
mberends DanielC: that makes sense
DanielC RPM uses Berkeley DB... 14:44
14:45 finanalyst joined
DanielC Berkeley DB is now owned by Oracle, but it seems to have a compatible license (Sleepycat License). 14:46
skids ]\
mberends Perl 5 tied hashes also fitted Berkeley DB very nicely
14:46 fridim_ joined
DanielC BerKeley DB website: "Extreme performance—eliminates the overhead of SQL and interprocess communication" 14:47
mberends there are also versions with more open licensing 14:48
DanielC Ok. I didn't know that.
skids scolds cat
s1n @seen masak
lambdabot I saw masak leaving #perl6 2h 34m 6s ago, and .
pmurias having the storage backend pluggable could be a good idea 14:49
14:50 pmurias left, ZuLuuuuuu left 14:51 icwiener joined 14:54 LadyLuna1y joined 14:58 shinobi_cl joined 15:03 DemoFreak left, _Chillance_ left, Chillance joined 15:10 wolverian left, LadyLunacy left
pugs_svn r26977 | pmichaud++ | [t/spec]: Change rakudo fudge from 'todo' to 'skip'. 15:13
15:20 nihiliad joined
mberends lichtkind: S06 'Macros' distinguishes between compile time and run time calls to macros, without and with the &, but you probably found that anyway. 15:24
lichtkind yes 15:25
mberends: but i dont understand consequences
you mean runtime i get 7 comtiletime 3+4 ?
15:26 ab5tract joined
DanielC m/[\. || \d]/ matches a dot or a digit. How do I match anything that is *NOT* a dot or a digit? In other words, what is the opposite of the preceding regex? 15:27
15:27 beggars joined
DanielC m/<-[\. || \d]>/ does not seem to do it. 15:30
pmichaud you don't need || inside of character enumerations 15:35
DanielC k
pmichaud Rakudo doesn't support \d inside of character enumerations yet 15:36
you can always do
DanielC 0..9 ok?
pmichaud <!digit> <-[.]>
slightly faster might be [ \D && <-[.]> ]
DanielC I don't understand the first one. To me it looks like a match for a non-digit followed by a match for a non-dot. 15:37
pmichaud the ! makes it zero-width
<!digit> means "don't match a digit at this point", but doesn't consume anything.
it's basically the same as <!before \d> 15:38
DanielC k
pmichaud std: say pi()
p6eval std 26977: OUTPUT«Undeclared routine:␤ pi used at 1 ␤ok 00:02 35m␤»
DanielC The firs one works, but /[\D && <-[.]>]/ gives an error: Statement not terminated properly 15:39
pmichaud maybe just one &
DanielC yeah
that works, thanks
lichtkind mberends: do you think that effect the return value of the macro if its called runtime oder compiletime?
mberends lichtkind: no. (still writing another reply) 15:40
lichtkind mberends: wont stress you take your time 15:41
15:41 shinobi_cl left
mberends just the generated code. calling a macro at compile time should inline the definition at the point of call. preceding the macro name with & must cause the compiler to generate a runtime call to the code defined in the macro, thus treating it as a subroutine. 15:43
DanielC my $filename = urlencode($module); 15:44
for %metadata.sort.kv -> $k,$v {
$v = $v ~~ m/[\D & <-[.]>]/ ?? $v !! md5($v).substr(0,5);;
$filename ~= "-$k:$v";
}
mberends lichtkind: s/ just the generated code. //
DanielC Extensible naming scheme for modules. 15:45
Ex: Foo::Bar-auth:92de1-ver:1.2.0
lichtkind mberends: so macro s {3+4} say 2 * s(); expands to say 2 * {3+4 } 15:47
mberends DanielC: not bad
DanielC :)
pmichaud ...assuming that all Perl 6 implementations have a md5() builtin :-| 15:48
and that 5 chars of the md5 hash is sufficient to avoid a collision 15:49
mberends lichtkind: it may expand to 2 * ( 3+4 ), because the { } are just macro constructing
DanielC pmichaud: I thought the naming scheme only had to be implementation-specific. As long as *Rakudo* has any reasonable hash function...
mberends DanielC: we may choose a more compact hash, and yes, implementations do not have to be interoperable. 15:51
DanielC pmichaud: 5 hex chars => You can expect a collision after ~1e15 distinct values.
pmichaud DanielC: good counterpoint, thanks.
lichtkind mberends: true but to be hygeninical clean which syn states macros are by default it has to be in its own namespace ie block
pmichaud DanielC: (~1e15) where did you come up with that figure? 15:52
DanielC sqrt(16^5) => birthday paradox
IIRC
pmichaud that assumes md5 keys are evenly distributed in the first 5 hex digits, though.
DanielC Yes, and they should be, to a good approximation. 15:53
pmichaud why is that?
DanielC because md5 was designed to be a cryptographically secure hash function.
Any lack of randomness would be considered a break.
pmichaud in the entire hash, yes.
not necessarily in the first five digits.
DanielC If there any subset of the hash is not random, the hash is not random. 15:54
If any subset of the hash is not evenly distributed as in a random function, that automatically constitutes an attack on the hash. 15:55
s/attack/break/
pmichaud anyway.
I'm not sure I entirely agree (and md5 has been broken), but I'll grant your point for now. 15:56
15:56 alester joined
DanielC md5 is broken => someone found a way to find collisions that is better than the birthday attack 15:56
15:57 |MoC| left
DanielC The attack is complex. Are we worried about an elaborate attack to get two modules with the same set of hashes? 15:57
pmichaud I'm worried about two modules ending up with the same key. 15:58
DanielC At most, it would have to be an intentional attack.
It will not happen by chance, other than through the birthday paradox (ie. sqrt(16^digits)) 15:59
pmichaud Again, I'm not sure I agree to the first 5 characters of precision.
our keys aren't randomly distributed.
sorry, I miswrote.
our module names aren't randomly distributed.
DanielC that has nothing to do with hashes.
a random function will produce random output on non-random input. 16:00
otherwise what's the point?
pmichaud hashing isn't a "random function"
16:00 M_o_C joined
DanielC yes, it is 16:00
That's the most central property of hashing
well... pesudo-random
pseudo-random = statistically equivalent to random 16:01
lichtkind mberends: so macro s { '3+4' } say 2 * s(); expands to say 2 * '3+4' ; which still is 14 ?
pmichaud the keys aren't "random", they just don't have any particular probability distribution
mberends we only have to worry about the risk of collisions behind a specific module name.
pmichaud and that can be true for the key as a whole without having to be true for all substrings of the key
mberends lichtkind: yes, that's what I believe :) 16:02
pmichaud anyway, I'll drop it here.
DanielC If it makes you feel better, we can use sha256, but I tell you that the change is pointless unless you are worried about intentional attacks.
pmichaud The purpose isn't cryptographic security, it's avoiding collisions. 16:03
mberends even md5 is overkill, and most hash indexing schemes have a fallback for collisions
DanielC intentional attacks = someone intentionally wanting to submit two modules with different (say) authorities that hash to the same file name.
pmichaud that's my point, I'd like to see a fallback for collisions.
DanielC pmichaud: avoiding collision is one of the central properties of a cryptographic hash.
mberends the fallback, seldom invoked, can use the contents of the files 16:04
DanielC pmichaud: The three properties of a cryptographic hash are (1) collision-resistance, (2) second pre-image resistance and (3) pre-image resistance. 16:05
pmichaud (I said I'm dropping it, so that's what I'm doing.)
DanielC sorry
mberends DanielC: imagine a very compact hash, resulting in collisions 1% of the time. We can still design a fallback system around that and have only a 1% average inefficiency. 16:07
pmichaud okay, I'll bring it up just to clarify. 16:08
I don't mind if we use md5. I don't mind if it doesn't have a fallback to begin with. But at present I don't have a good md5 implementation in Parrot/Rakudo, so something else might be easier.
more importantly, it may need to be fast-to-compute until there's an indexing system behind it. 16:09
DanielC CRC is fast to compute... (not a cryptographic hash though) 16:10
mberends good enough though, even CRC-16
DanielC Are we worried about intentional attacks or just about accidental collisions?
mberends the latter
pmichaud I'm only worried about accidental collisions.
DanielC Any hash should work then.
mberends yes
pmichaud there are other (better) mechanisms we can use for detecting/avoiding intentional attacks. 16:11
i.e., involving attacks on the file itself, not just its name.
DanielC I first wrote MD5 because it is the simplest hash that is popular and is second-pre-image resistant.
pmichaud sure, I would've chosen md5 first also, except that I know that md5 isn't immediately available in Rakudo.
DanielC But CRC is cool too. 16:12
I understand.
pmichaud although I _think_ that parrot may be using md5 internally these days for some things, in which case we might be able to easily expose it to Rakudo.
16:13 xinming_ is now known as xinming
pmichaud std: undef() 16:14
p6eval std 26977: OUTPUT«Undeclared routine:␤ undef used at 1 ␤ok 00:02 35m␤»
16:16 mikehh joined
DanielC If we are going for fast, the Adler-32 checksum is even faster than CRC, and is easy to implement. en.wikipedia.org/wiki/Adler-32 16:19
pmichaud I tend to vote for simple first.
i.e., it's good not to over-design before we know what we need.
DanielC At first blush, Adler-32 seems like the simplest. 16:20
pmichaud sounds good to me.
DanielC It is a 32-bit sum. Assuming that the output is evenly distributed (and that's definitely an assumption now) it would take ~65,000 authorities to find a collision in the authority field. 16:22
pmichaud: Do you think it'd be hard to change the algorithm later on? 16:25
pmichaud no.
the other question is: "how often do we think that the :auth will be the distinguishing characteristic for module names"? In these early days, I'm thinking "not often" 16:26
DanielC I just read that Adler-32 is especially week for short messages. So collisions will happen much sooner than ~65,000 modules.
pmichaud we should know how we'll handle it when we get there, but I can envision baby steps where we don't worry about it.
DanielC So Adler-32 may be a good choice "for now" because it's easy, but for deployment it might be better to pick something else. 16:27
lichtkind mberends: but my most important issue is what is in the AST a macro returns the whole routine body or the return value? 16:32
mberends DanielC: you're only hashing the authority, not the module name, so that would be 64k authorities per module. I would vote for Fletcher-16, Adler's predecessor. I used it once in C, long ago.
DanielC mberends: It would be a lot less than 64k, but yes, it would be per module. In fact, it would be per module + version. 16:33
mberends lichtkind: I think that's the distinction between compile time macro calling and run time macro calling. The compile time substitutes the return value, the run time substitutes the macro body. 16:34
16:35 ZuLuuuuuu joined
mberends DanielC: indeed, and we need to distinguish only a handful of authorities 16:35
DanielC mberends: Yes.
lichtkind mberends: now i get it big thanks
mberends lichtkind: thanks to you too, I didn't know that much when you began asking :) 16:36
lichtkind haha
DanielC Fletcher-16 is faster but less reliable than Adler-32, right?
mberends collides a bit more often
ISTR it's about 13 effective bits out of 16 16:37
DanielC ok
mberends 8k authorities per module + version
TimToady that's not how macros work
DanielC assuming that's randomly distributed, that gives us 90.
TimToady nothing ever substitutes the macro body 16:38
mberends lichtkind: heads up!
DanielC 90 authorities before we expect a collision via the birthday paradox.
TimToady a macro returns only a string or an AST
a string is inlined (in its own scope) and reparsed
lichtkind mberends: what does mean that? i hope you particpate also in the p6 tut in which will lead my search (still german only) 16:39
TimToady if you have { 3+4 } it would be stringified to '3+4' and reparsed
an AST is just inserted
16:39 alester_ joined
DanielC mberends: 13 random bits => 90 authorities before we get a collision (birthday paradox). 16:40
lichtkind TimToady: so it doesnt matter if i write in the macro 3+4 or '3+4' but the difference if the macro is parsed or is reparsed?
TimToady run time doesn't change this, but assumes a sub {} around the AST (which, for a string return, would have to be compiled)
16:40 alester left
TimToady there are only two possibilities, AST and everything else 16:40
16:40 FurnaceBoy joined
mberends DanielC: yes 16:40
TimToady and everything else is assumed to be stringified
mberends DanielC: that would be 1 collision per 90 distinct (module+version) instances that have multiple authorities. Still unimaginably remote. 16:42
16:43 cotto left
DanielC mberends: It does sound unlikely in practice. Generally there is only 1 authority per module... 16:43
mberends nod
16:43 Psyche^ joined
DanielC mberends: On the other hand, do we really need the extra speed of a 16-bit sum? Is there any real harm in using a 32-bit sum? (Fletcher-32, Adler-32,CRC-32). 16:45
16:45 justatheory joined
mberends it results in shorter names for the filesystem 16:46
DanielC 32-bits => 5 characters base 64. 16-bit => 3 characters base 64 (I think). 16:48
mberends each base64 char is 6 bits 16:51
DanielC So that means 6 characters vs 3 characters. 16:52
Foo::Bar-auth:92de10-ver:1.2.0 16:53
Foo::Bar-auth:a12-ver:1.2.0
lichtkind TimToady: the is reparsed trait triggers the decision that a string is returned?
DanielC I don't think the extra 3 chars make a big difference. 16:54
TimToady has nothing to do with whether a string is returned, but what the supplied rule is expected to parse 16:55
16:55 Patterner left, Psyche^ is now known as Patterner
lichtkind TimToady: thanks that was the crucial bit to me, i understood it yesterday the other way around 17:00
17:00 rewt joined
lichtkind TimToady: so macro s is reparsed { '3+4' } say 2 * s(); result would be 10 ? 17:03
17:09 frew|work joined
mberends DanielC: base64 could also give you Foo::Bar-auth:AnYtXt-ver:1.2.0 17:10
DanielC mberends: yes
And usually, it will.
mberends :)
17:12 alester_ left
DanielC The Fletcher algorithm is really easy: ($A,$B) = 0; for @blocks -> $D { $A += $D; $B += $A } 17:13
dalek kudo: 50ec44e | pmichaud++ | (3 files):
Use iso-8859-1 (fixed-width) instead of utf8 for parsing when we can.
kudo: 0b9c9a3 | tene++ | (2 files):
Track changes in HLL library loading API.
mberends DanielC: that may be why I once used it :) 17:14
DanielC Fletcher16 and Fletcher32 are the same algorithm, with different block size.
:)
mberends it was probably on PC-DOS, so 16 bit :) 17:18
DanielC buubot: split
hm...
DanielC wants to find the simplest way to split a string into characters.
mberends buubot: spack split 17:19
buubot mberends: S02-bits.pod:5 S03-operators.pod:2 S05-regex.pod:6 S06-routines.pod:2 S19-commandline.pod:9 S29-functions.pod:13
DanielC thanks
17:20 kane_ joined
DanielC sub Fletcher16($str) { 17:21
my ($A,$B) = (0,0);
loop split //, $str -> $chr {
$A = ($A + ord $chr) % 256;
$B = ($B + $A) % 256;
}
return (chr $A) ~ (chr $B);
}
DanielC thinks that's about right.
s/loop/for/ 17:22
17:23 bloonix left
DanielC Hmm... still need to convert the output to hex or base 64. 17:23
mberends part of the debate was about Unicode chars in names, so ord $chr will probably overflow 256. 17:24
DanielC hm
indeed
mberends but nevermind, it's close and a good first shot :)
DanielC :-) 17:25
17:26 bloonix joined
DanielC s/256/255/ 17:29
17:29 ejs joined 17:30 Traveler8 joined 17:35 kane__ left
DanielC rakudo: say ord "動" 17:39
p6eval rakudo 764684: OUTPUT«21205␤»
DanielC On my computer it says 229. Why? 17:40
sjohnson kanji for "move" 17:42
DanielC Any idea why my local Rakudo's ord function gives 229? 17:43
pochi rakudo: say "動".chars
p6eval rakudo 764684: OUTPUT«1␤»
pochi it probably doesn't think of it as unicode? 17:44
DanielC rakudo: say "動".bytes
p6eval rakudo 764684: OUTPUT«ResizableStringArray: Can't shift from an empty array!␤in method Any::bytes (src/gen_setting.pm:242)␤called from Main (/tmp/HfNytq7uAt:2)␤»
DanielC Did I call the wrong method?
mberends DanielC: there may be differences between text from a script file and from your console, ie outside of rakudo 17:45
DanielC mberends: Indeed, you are right. 17:46
mberends .bytes is fairly new and possibly untested for some inputs.
DanielC I just put it in a file and now it gives the right answer.
pmichaud Parrot doesn't always know how to read things as utf8 from the command line or from the console input 17:47
17:50 [particle] joined 17:53 sri_kraih joined
DanielC Does Perl 6 have an integer division operator? 17:59
18:00 wknight8111 joined
DanielC int $a/$b will do. 18:00
mberends DanielC: are you working on a Unicode hash now? 18:03
DanielC yeah 18:04
sub Fletcher16($str) {
my ($A,$B) = (0,0);
for split //, $str -> $chr {
my $val = ord $chr;
$A = ($A + $val % 255) % 255;
$B = ($B + $A) % 255;
if $val > 255 {
$A = ($A + int $val/255) % 255;
$B = ($B + $A) % 255;
}
}
return int2hex($A*256 + $B);
}
But I think my "split" line is broken.
rakudo: say split //,"hello" 18:05
p6eval rakudo 0b9c9a: OUTPUT«Statement not terminated properly at line 2, near "//,\"hello\""␤in Main (src/gen_setting.pm:0)␤»
pmichaud // isn't a valid pattern 18:06
DanielC :-(
pmichaud try ''
DanielC rakudo: say split '',"hello"
p6eval rakudo 0b9c9a: OUTPUT«hello␤»
DanielC rakudo: say (split '',"hello").perl
pmichaud rakudo: say (split '', 'hello').perl
p6eval rakudo 0b9c9a: OUTPUT«["h", "e", "l", "l", "o"]␤»
DanielC excellent.
sjohnson rakudo: say split '', "happy house".join " "
p6eval rakudo 0b9c9a: OUTPUT«Statement not terminated properly at line 2, near "\" \""␤in Main (src/gen_setting.pm:0)␤»
pmichaud rakudo: say 'hello'.split('').perl 18:07
p6eval rakudo 0b9c9a: OUTPUT«["h", "e", "l", "l", "o"]␤»
sjohnson rakudo: say (split '', 'hello').join "-"
p6eval rakudo 0b9c9a: OUTPUT«Statement not terminated properly at line 2, near "\"-\""␤in Main (src/gen_setting.pm:0)␤»
pmichaud sjohnson: method calls required a : or ( ) for params
(the .join)
sjohnson rakudo: say (split '', 'hello').join("-")
p6eval rakudo 0b9c9a: OUTPUT«h-e-l-l-o␤»
sjohnson >_> 18:08
<_<
DanielC rakudo: say ("hello".split('').map {.ord}).perl
18:08 [particle]1 left
p6eval rakudo 0b9c9a: OUTPUT«Statement not terminated properly at line 2, near "{.ord}).pe"␤in Main (src/gen_setting.pm:0)␤» 18:08
pmichaud std. say 'hello'.join '-'
std: say 'hello'.join'-'
p6eval std 26977: OUTPUT«##### PARSE FAILED #####␤Syntax error (two terms in a row?) at /tmp/wNVLmQIjDh line 1:␤------> say 'hello'.join'-'␤ expecting any of:␤ POST␤ infix or meta-infix␤ infix stopper␤ method arguments␤ postfix␤ postfix_prefix_meta_operator␤
..standard stopper␤ statement mod…
pmichaud std: say 'hello'.join '-'
p6eval std 26977: OUTPUT«##### PARSE FAILED #####␤Syntax error (two terms in a row?) at /tmp/xyC66rM6M3 line 1:␤------> say 'hello'.join '-'␤ expecting any of:␤ infix or meta-infix␤ infix stopper␤ standard stopper␤ statement modifier loop␤ terminator␤FAILED 00:02 35m␤»
DanielC rakudo: map { ord $^a } "hello".split('') 18:09
p6eval rakudo 0b9c9a: OUTPUT«Statement not terminated properly at line 2, near "\"hello\".sp"␤in Main (src/gen_setting.pm:0)␤»
DanielC :-( 18:10
pmichaud need a comma between params
DanielC rakudo: map { ord $^a }, "hello".split('')
p6eval rakudo 0b9c9a: ( no output )
DanielC :-)
pmichaud rakudo: say map { .ord }, "hello".split('')
p6eval rakudo 0b9c9a: OUTPUT«104101108108111␤»
18:11 frew|work left
DanielC On my computer it gives : too few arguments passed (1) - 4 params expected 18:11
rakudo: say map { .ord }, "hello".split('') 18:12
p6eval rakudo 0b9c9a: OUTPUT«104101108108111␤»
DanielC ah, I know...
stupid shell.
mberends quotes, huh?
DanielC yeah 18:13
mberends afk & # birthday beers @neighbours 18:15
DanielC pmichaud: Does Rakudo support #(( ... )) multi-line comments? 18:16
pmichaud rakudo: say #(( not this )) "yes, it does"; 18:17
p6eval rakudo 0b9c9a: OUTPUT«yes, it does␤»
DanielC Mine says: Can't use embedded columns in column 1 at line 3, near "((\n\tmy $fi
pmichaud oh, that should say "embedded comments"
DanielC But mine is a multi-line comments.
pmichaud multi-line comments aren't allowed to begin in column 1
18:18 jdv79 joined, jdv79 left, abra left
DanielC ? 18:18
pmichaud S02
18:18 jdv79 joined
DanielC So I need to add a space before #? 18:18
lichtkind mberends: still there?
DanielC pmichaud: I added a space before # and now it works.
mberends yes 18:19
pmichaud S02: "An embedded comment is not allowed as the first thing on the line."
jdv79 anyone know if there is any online version of the talk entitled "The Art of Ballistic Programming" by L. Wall? 18:20
pmichaud should be
mberends lichtkind: pong
pmichaud it was broadcast live when he gave it at MIT
lichtkind mberends: so if i understood larry correctly 18:21
mberends: so macro s is reparsed { '3+4' } say 2 * s(); result would be 10 ?
mberends lichtkind: I'm confused, because I would expect 14. but I'm really going now. Prost! 18:22
lichtkind prost
mberends: 14 would be in parsed mode
skids pmichaud: Apparently he wasn't mic'd well so only the slides came across. 18:26
18:26 jdv79 left, jdv79 joined
jdv79 if anyone answered me could you answer again - my dsl line just died:( 18:27
DanielC mberends / pmichaud: Unicode implementation of Fletcher-16 located in perl6-examples/misc/Fletcher.pl 18:28
I haven't tested it much. 18:29
18:30 sri_kraih left
lichtkind anybody here with clue about p6 macros? 18:33
18:39 sri_kraih joined
skids jdv79: From what I heard, they tried to tape that talk, but the audio didn't come out. 18:41
18:47 finanalyst left 18:48 justatheory left, Traveler8 left 19:06 DemoFreak joined 19:07 cls_bsd left 19:09 LadyLuna1y left, LadyLunacy joined, cls_bsd joined 19:10 LadyLunacy left
DanielC rakudo: "he_" ~~ /<-alpha -\_>/ 19:11
p6eval rakudo 0b9c9a: OUTPUT«perl6regex parse error: Error parsing enumerated character class at offset 29, found '\'␤in Main (src/gen_setting.pm:0)␤»
DanielC Grr... I'm really struggling with the new regexes. 19:12
I want to match anything that is not alphanumeric and not an underscore.
rakudo: "he_" ~~ /<-alpha -[_]>/ 19:14
p6eval rakudo 0b9c9a: ( no output )
DanielC rakudo: "he_" ~~ /<-alpha -[_:-]>/
p6eval rakudo 0b9c9a: OUTPUT«perl6regex parse error: Unescaped '-' in charlist (use '..' or '\-') at offset 32, found '-'␤in Main (src/gen_setting.pm:0)␤»
DanielC rakudo: "he_" ~~ /<-alpha -[_:]>/
p6eval rakudo 0b9c9a: ( no output )
DanielC rakudo: "he_" ~~ /<-alpha -[-_:]>/ 19:15
p6eval rakudo 0b9c9a: OUTPUT«perl6regex parse error: Unescaped '-' in charlist (use '..' or '\-') at offset 30, found '-'␤in Main (src/gen_setting.pm:0)␤»
DanielC rakudo: "he_" ~~ /<-alpha -[\-_:]>/
p6eval rakudo 0b9c9a: ( no output )
DanielC rakudo: if "ab_" ~~ /<-alpha -[\-_:]>/ { say "y" } else { say "n" } 19:16
p6eval rakudo 0b9c9a: OUTPUT«n␤»
19:17 ejs1 joined 19:29 Maghnus left 19:30 Maghnus joined 19:31 eMaX joined 19:42 Fuad joined
Fuad Hello 19:42
Matt-W hello 19:59
jnthn hi Fuad :-) 20:00
20:02 viklund joined
Fuad howdy,jnthn:) 20:03
20:07 Eevee joined
DanielC rakudo: my @friends = ( {name=>"Joe"}, {name=>"Mike"} ); say @friends[0]{name}; 20:07
p6eval rakudo 0b9c9a: OUTPUT«Could not find non-existent sub name␤»
pmichaud <name> 20:08
undeclared barewords are treated as subroutine calls in p6
DanielC Am I supposed to put quotes around it?
rakudo: my @friends = ( {name=>"Joe"}, {name=>"Mike"} ); say @friends[0]{'name'};
p6eval rakudo 0b9c9a: OUTPUT«Joe␤»
DanielC ah
pmichaud you can either do @friends[0]{'name'} or @friends[0]<name>
DanielC thanks
<name> is shorter.
pmichaud That's one of the reasons it's there :-)
DanielC Why does <name> work? I thought <> was a list constructor? 20:09
pmichaud not precisely
<name> is the same as ('name')
which isn't a list
jnthn <name age> gets a hash slice though. :-)
pmichaud <abc def> is the same as ('abc', 'def')
which is a list
jnthn %h<name age> that is
iirc
DanielC ok...
pmichaud parens and angles don't construct lists 20:10
just like (3) is not a list
neither is <3>
DanielC ok...
pmichaud (caveat: (3) might be a list now.)
(there's some spec changes taking place) 20:11
(regardless, %hash<name> is equivalent to %hash{'name'}
and %hash<name age> is equivalent to %hash{'name', 'age'}
jdv79 skids: thanks 20:13
20:14 ejs1 left 20:25 wknight8111 left 20:26 eMaX left 20:27 cotto joined 20:32 eternaleye left 20:37 ejs left
DanielC @karma 20:38
lambdabot You have a karma of 5
DanielC Hmm... I don't get karma points for committing to perl6-examples. Oh well. 20:39
pmichaud: Is a module always one file? 20:40
20:42 eternaleye joined 20:44 Fuad left, M_o_C left
pmichaud I think Infinoid was going to switch dalek to report perl6-examples changes 20:45
DanielC I thought he had already done it... 20:46
Anyways, is a module always one file?
I'm just wondering if a module can contain many files. 20:47
21:05 hanekomu joined 21:08 fridim_ left 21:10 hanekomu_ joined 21:26 hanekomu left 21:51 frew|work joined 21:53 volalba joined
volalba who studies CS here 21:54
21:57 iblechbot left 21:58 volalba left 22:12 jferrero joined 22:21 explorer__ left 22:27 tulcod left 22:41 volalba joined
DanielC This channel is quiet today. 22:44
22:45 volalba left
jnthn Sunday. 22:48
DanielC I guess.
pochi they're all celebrating whit sunday 22:49
DanielC I'm looking at the Rakudo source code. It's hard to find anything written in Perl. It's all Parrot. 22:50
literal check the setting
src/setting
DanielC ah, yes
Lots of .pm files there. What is this directory for? Why is it called settings?
literal not settings, but "the setting" 22:51
aka. "the prelude"
DanielC ok
22:51 hanekomu_ left
literal basically, the initial Perl 6 environment 22:52
see S32 on perlcabal.org/syn/
DanielC Ok, I see where it defines 'class Object'
I see the basic classes here.
pmichaud the other perl stuff is in src/parser/ 22:53
src/parser/actions.pm is the bulk of the compiler
literal I believe the idea is that this is the standard library which Perl 6 will provide by default, but in some environments people will swap it out for something else
pmichaud src/parser/grammar.pg is the grammar
DanielC pmichaud: What is a .pg file? 22:54
pochi perl grammar?
DanielC It looks like Perl code.
pmichaud yeah, something like that.
it's Perl code, at any rate, with a little bit of embedded PIR
DanielC One of these days I'll learn about the grammar features of Perl 6. 22:55
pmichaud early on one of the Parrot architects wanted to distinguish grammars from other files, so .pg got chosen and we never really switched.
DanielC I have never learned a grammar programming language, or whatever they're called.
pmichaud in Perl 6 it's just regular expressions
(very advanced regular expressions, but just regular expressions) 22:56
DanielC That's interesting.
DanielC makes a mental note to read all of S05
S05 is... long.
pmichaud yes, there's a lot there. 22:57
DanielC This 'rule' feature looks very nifty.
You can make hierarchical regular expressions... 22:58
skids the enraging part is "i know i read something that did exactly what i want in S05, but can't find it now" :-)
22:58 kate21de left
DanielC Can rules be recursive? For example, in a Lisp grammar you might define an element to be a number, a string, or a list of elements. 22:59
skids DanielC: more than hierarchal, regexes are subs and grammars classes
pmichaud DanielC: yes, rules can be recursive. 23:00
DanielC cool
So you can define a natural number to be either 0 or 1 + a natural number :-)
skids DanielC: if you really want to fry your brain, read STD.pm
DanielC where can I find STD.pm ?
skids in the pugs repository 23:01
DanielC Ok. Maybe later.
literal under src/perl6/
skids yes, read s05 first :-)
DanielC :-)
literal svn.pugscode.org/pugs/src/perl6/STD.pm
DanielC I've scanned up and down S05 a lot, but I haven't read everything.
*click* 23:02
skids he just couldn't resist, poor devil.
DanielC Mental note: figure out what a 'token' is in Perl. 23:03
pmichaud it's a regex with backtracking turned off by default
DanielC What is a twigil?
pmichaud (a token is a regex with backtracking turned off by default) 23:04
literal the * in $*VARIABLE
skids Second character after a sigil
DanielC What is backtracking?
literal en.wikipedia.org/wiki/Backtracking
DanielC Sorry if I'm asking too many stupid questions.
*click*
skids You'll know when we start linking with lmgtfy.com :-) 23:05
DanielC :-)
In STD.pm I see twigils like . ! ^ : * + ? Does that mean that I might find $+VAR, $!VAR, $:VAR, etc ? 23:06
skids yep.
literal well, $+VAR is deprecated, I think
DanielC It's amazing that Perl would have that many variables. 23:07
skids Twigils are for special purposes.
literal it's not really different variables, but different scoping usually
skids e.g. $!var is always an instance attribute.
23:07 ZuLuuuuuu left
DanielC ok 23:08
Is $* always a file handle?
$*FOO
literal nah
DanielC Seemed like a reasonable guess ($*IN)
skids It's scoping. 23:09
literal it's a global/contextual variable
$*IN is global to your process, for example
pmichaud well, not always :-)
DanielC So $*FOO is visible in all files, whereas a regular global is only visible in the current file? Is that right? 23:10
pmichaud $*FOO refers to the innermost dynamic scope that declared $FOO as a context var
if no such scope, then it looks in the process environment and some other places
DanielC How is a context var different from other vars? 23:11
pmichaud it's one that can be searched via the calling stack
it's like an environment variable in the shell
DanielC ok
pmichaud my $FOO is context; some_sub(); # inside of some_sub, $*FOO would refer to the $FOO I just declared 23:12
my $IN is context = open('file.txt', :w); say 'hello'; # outpus 'hello' to file.txt
*outputs
23:13 beggars left
DanielC In the $FOO example, why wouldn't you just use a global variable? 23:13
pmichaud It's another way of controlling scope and reducing global variable conflicts. 23:14
Basically $*FOO means "the contextual $FOO defined by one of my callers
DanielC Is it mainly to remind you that the thing you are looking at is a global variable? 23:15
rakudo: my $a = 2; sub foo() { say $a }; foo()
pmichaud more than a reminder
p6eval rakudo 0b9c9a: OUTPUT«2␤»
pmichaud oh, I see what you're saying. 23:16
it's more like this:
sub foo() { say $*a }; { my $a is context = 2; foo() }
DanielC ah
pmichaud that's how print() can refer to a $*IN that isn't in print's lexical scope 23:17
it's in print's callers' lexical scope(s)
DanielC rakudo: sub foo() { say $*a }; { my $a = 2; foo() } # should error
p6eval rakudo 0b9c9a: OUTPUT«Use of uninitialized value␤␤»
DanielC yeah 23:18
That's very neat.
pmichaud ("is context" isn't implemented in rakudo yet)
and rakudo still treats $*a as globals, per an earlier version of the spec.
DanielC ok
It's a very neat feature. So you can avoid making more truly-global variables. 23:19
23:21 viklund left
DanielC sub count() { $*cnt++ }; sub bar() { my $cnt is context = 10; count(); count() count() } 23:21
DanielC missed one ; 23:22
pmichaud in that case it would need to be "is context<rw>"
DanielC sub count() { $*cnt++ }; sub bar() { my $cnt is context rw = 10; count(); count(); count() }
That's an interesting feature. I've never seen anything like that. 23:23
23:28 justatheory joined 23:44 DanielC left 23:51 snarkyboojum joined 23:52 justatheory left 23:56 icwiener left 23:57 snarkyboojum 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!