www.parrot.org | Parrot 1.6.0 "half-pie" released: The JIT is dead! | Testing priorities: Exception and MultiSub
Set by moderator on 20 September 2009.
00:00 mokurai joined 00:01 mokurai left 00:05 mokurai joined
japhb Tene, it's a good start. I'm going to tweak a couple things (like moving the top of action_fetch() out to an init-time function), but I think the only real thing to think about is where the working directory should be. 00:05
00:05 quek joined
japhb I'll see if proto has a decent precedent here. 00:06
Tene japhb: IMO, we need a 'mktemp' function. 00:14
japhb: and then fetch to that directory 00:15
japhb Tene, but if configure/build/install fails, people want to examine the wreckage.
Tene japhb: Sure, which is why you only rm the directory if all steps succeed properly.
You only move on to the next step if the last step succeeded, and 'cleanup' is the last step 00:16
00:16 ash_ joined
japhb And it could be a significant savings to have a known directory where fetched subdirs go, so that we can later just update and rebuild, rather than doing complete fetch again. 00:16
But that may be a later change.
Tene Sure.
japhb: you want a pure parrot http-fetching library? 00:17
japhb: or libcurl wrapper?
japhb: or config-detected downloader, like wget/curl?
japhb I'd prefer pure Parrot. Doesn't have to be fancy yet.
Tene Or... what were you thinking? 00:18
kk, I'll do something basic this week. I was going to look at Socket refactors anyway.
That was my inclination too.
japhb we *can* do a config-detected downloader, if it's too painful, I suppose.
Ah good.
And that puts some real testing on the socket API.
Tene I don't think that windows even ships with anything like wget anyway... so we'd need a native parrot solution anyway 00:19
AFK dinner. 00:20
japhb k
We need some way to stop that. 00:21
Sorry, wrong window. 00:22
Trying to fetch Close, it pulls a pile of stuff and then Subversion wants authentication info for 'Authentication realm: <close.googlecode.com:443> Google Code Subversion Repository' and won't seem to proceed without it. We need some way to stop that.
(That was the message I was sending to Austin, but I just realized Tene probably cares too.) 00:23
Tene: incoming commits 00:24
00:29 rhr joined
Tene Hmm. Don't know enough about svn. :( 00:33
00:33 yanger joined 00:42 yanger left 00:48 Austin joined 00:52 particle joined
allison anyone around to answer a runcore question? 00:58
japhb use.perl.org/~geoffrey/journal/39682 # Parrot Plumage "Day" 6 01:05
And now, maybe I can start working on actual new code.
Or maybe eat dinner.
Austin japhb: I've changed the externals link on close to use http, which is the read-only approach to the repo. That should reflect what most people do. 01:14
japhb Austin, cool, let me try a fetch again 01:18
Austin, seemed to work like a charm. 01:19
Austin In fact, japhb, my inclination would be to specify no-externals as part of the fetch, since the link in this case is for documentation.
Are we still using the "I'll give you a command line to use" approach? 01:20
dalek ose: r167 | Austin++ | trunk:
Changed externals on wiki to read-only
japhb Austin, at the moment no -- Tene's commit to add fetch is building the command directly from the repository metadata. 01:21
s/metadata/section of the metadata/
Austin Okay.
So I need a setting to tell him to ignore the externals. 01:22
japhb Whirlpool development: the best kind of suck. :-)
Yep, was just thinking that.
Austin purl, msg tene Plumage needs a mechanism to tell the SVN fetcher to --ignore-externals
purl Message for tene stored.
japhb You know, I think the refactor that is needed is to change the instructions from having a single 'type' flag to having per-stage info. 01:24
Hmmm
Tene japhb: agreed. 01:39
japhb: require at least a type for each stage, and optionally additional type-specific metadata 01:40
like, say { type : svn, extra-args : --ignore-externals } 01:41
and, as a last-resort fallback, { type : custom, command : "$SVN co --ignore-externals $REPO $DEST" }
oslt
japhb (here for just a sec, cooking dinner): Tene, yeah, sounds like a plan. About time to merge my thoughts about the metadata spec back to the wiki anyway. 01:44
Oh, for metadata spec revision, I'm thinking just an incrementing number. So the next rev will be revision 1, then 2, and so on.
Any objections, let me know. 01:45
Tene japhb: sounds fine to me... I don't think we're at a stage to worry about trying to support old versions of the spec yet, though. ;)
japhb: any specific requests for tonight? I'm going offline to hack in not too long. 01:46
If nothing specific, I think I'll work on Socket.
cconstantine Tene: Hi :) I think I figured out a possible root-cause of my tailcall problems; the 'let' special form from stene is broken 01:52
Tene broken how?
cconstantine either broken or not acting as expected 01:53
Tene explain? 01:54
cconstantine given nopaste.snit.ch/18076
the call to func1 in func can't find func1
(I changed lambda to fn) 01:55
Tene I'm not too surprised. I don't remember considering that case.
Good catch.
cconstantine it appears to work for functions because it's somehow able to find it in the global scope
ie, the code posted works because the '$P0 = 'func1'($P1)' works, even though it should be pulling func1 from the lexical scope 01:56
Austin Seems like 'let' isn't adding the function to its block symtable 01:58
Or setting a lexical, if it happens at runtime. 01:59
cconstantine I think? 02:00
I added a bunch of sentinal 'say's in actions.pm, and it's detecting if the vars inside the val form before registering any vars 02:02
japhb Tene: No special requests. Socket work and HTTP fetch are excellent next tasks. 02:03
Tene okay, AFK
japhb ditto
Tene NO I WAS AFK FIRST
... oops. ;)
Austin So my bank's online banking just rolled out (to me) their new! improved! interface. And about 40 seconds later, I have found a Severity 1 bug. 02:15
No, you don't look like complete buffoons.
02:16 rhr joined 02:21 integral joined 02:22 TiMBuS joined, JimmyZ joined
cconstantine Austin: which bank is that again? ;) 02:35
Austin "TD bank North"
All your statements will be June 02. 02:36
Good luck on those taxes.
cconstantine hehe 02:38
So, you found the problem in my 'let' any idea how to get it to add the lexical var before evaluating the val portion? 02:39
Austin Where's the code? 02:40
purl i think the code is already out
Tene purl: msg WhiteKnight re "Add an improved file-like API to the Socket PMC" can you explicitly define what that API is? 02:41
purl Message for whiteknight stored.
cconstantine github.com/cconstantine/Reason
Austin Is this scheme?
Or lisp, or whatever?
cconstantine kinda
the lisp I want
02:42 janus joined
cconstantine will be mostly inspired by clojure 02:42
Austin Okay. You know that compiling is a bad idea, right?
cconstantine sure?
purl But are you sure you're sure?
02:42 rhr joined
cconstantine I started with Tene's stene 02:42
Tene Austin: What do you mean by that? I don't think I know what you're talking about.
Austin Ok, so you create an immediate block, add all the symbols to it, and then process statements. 02:45
You'll have to R-your-own-FM on this one, but Scheme R5 does not define the order of eval for let args. Let*, on the other hand, is defined left to right. 02:47
So let may be fine, but let* might need to insert each var=val pair as it is found, rather than after they are all found. 02:48
jrtayloriv has a lightbulb come on over his head, and begins to see why Austin told him to make the TOP block immediate ... 02:49
Tene cconstantine: process each let* item as it's seen, instead of after the fact. 02:50
cconstantine I'm not sure I know how to do that 02:52
02:53 rg joined
cconstantine does this line "my $val := $<val>.shift.ast;" cause the val evaluate itself? 02:54
Tene .ast
Austin Instead of doing {*} #= start, add a call to {*} #= item after each item
And of course, catch $key eq 'item' in the action 02:55
cconstantine item?
purl hmmm... item is in an =over, so the indent accumulates.
cconstantine ah, val-var pair 02:56
Austin Tene: the reason I said that compiling was a bad idea was that the compilation symbol table is unavailable at runtime. Writing lisp-alikes pretty much means doing all the work in the execution environment. 02:57
03:00 kyle_l5l_ joined
cconstantine so, move the #= begin to after the 'let', add a #= item after ')'? 03:00
Tene cconstantine: you probably want to keep the #= begin 03:01
cconstantine: then do some setup in #= begin, handle each item in #= item, then finish up at #= end
cconstantine right, move not remove 03:02
Austin cconstantiine: Yes. Use begin to create the block, use item to add individual items to the block,
cconstantine will this allow a '(let ( (fun (fn (x) (fun 1)))))' ?
Tene cconstantine: yes.
cconstantine awesome 03:03
Austin Doctor, will I be able to play the violin after the operation
Tene cconstantine: the test you probably want to use is: (let ((a 1)(b (+ a 5))) (say b)) 03:05
approximately
cconstantine my thing was so that I could do recursion in an anonymous function 03:06
without resorting to a special form
Tene Yeah, I know.
cconstantine that's a good test too 03:07
dalek ose: r168 | Austin++ | trunk/ (34 files):
checkpoint: Renamed close::Compiler -> Slam
 03:08
cconstantine ok, what is the '@?BLOCKS' in this function? 03:18
Austin global stack of PAST blocks 03:19
cconstantine unshift is a push?
jrtayloriv yes
Austin In this case, yes.
cconstantine shift is a pop? 03:20
Tene unshift is a push onto the *front*
Austin I guarantees that the top of the stack is always [0]
Tene shift is a pop from the front
Austin *I = it
cconstantine ok
so if I don't want to add or remove blocks I better shift and unshift the same number of times
Tene Yes. 03:21
03:31 yanger joined
yanger ping whitenight 03:31
purl I can't find whitenight in the DNS.
cconstantine I'm popping from an empty array... is there anyway find out where? all I get is src/gen_grammar.pir:1888 in the trace 03:32
Austin Go look in gen_grammar.pir. 03:35
Or read the backtrace
cconstantine neither are helpful 03:38
at least not to me
Austin NoPaste the backtrace 03:42
03:42 rhr joined
nopaste Someone at 71.98.77.93 pasted "backtrace" (11 lines) at nopaste.snit.ch/18078 03:52
cconstantine nopaste.snit.ch/18078
Austin First thing to know about a backtrace is that there are two sets of methods: Grammar and Grammar;Actions. If it's in the Grammar, that's from your .pg file. 03:53
(Those are rare, but they happen.) 03:54
If it's in Grammar;Actions, then it's from your .nqp file.
cconstantine right
Austin Unless you put your actions in the same class as your grammar, which would be bad. 03:55
:)
cconstantine obviously
nopaste "tene" at 97.113.17.205 pasted "hackish wget.pir for japhb" (103 lines) at nopaste.snit.ch/18079
Austin (It may not be obvious. But a Grammar is a class, and the rules are method names. And the actions are method names, and they are the same names as the rules.)
japhb Tene: looking ... 03:56
Tene japhb: I threw together a proof-of-concept wget.pir. It seems to work on every url I tried it on.
japhb Tene, excellent
reading it now
Tene There are (obviously) some rough edges... but I think I can clean it up into a decent library.
Austin CConstantine: Is there an updated grammar online? Or can you nopaste that part of it with let?
japhb What is the ibm link in the DESCRIPTION there for?
Tene japhb: I took it from examples/io/http.pir 03:57
cconstantine Austin: my work-in-progress is in github
Tene japhb: Updating the header now 03:58
japhb Tene, nod
cconstantine Austin: I kinda knew that. the gramar is compiled to code, and te actions get called based on the gramar targets
nopaste "tene" at 97.113.17.205 pasted "Slightly updated wget.pir" (99 lines) at nopaste.snit.ch/18080 03:59
Austin Are you sure this is the same pg that generated the code?
cconstantine yup
Tene japhb: I suddenly realize that I could probably have written this in nqp... :P
japhb Tene, yeah, I was thinking the same thing. 04:00
Austin I'm wondering why the grammar identifies itself as "Reason::Grammar" but the backtrace says reason;Reason;Grammar
japhb Probably a tad less painful.
May need another builtin or two.
cconstantine is it bad that I think the pir is less scary then nqp?
Austin no.
But you'll change your mind.
cconstantine I live in asm at work
Austin You have my sympathy. 04:01
japhb I'm thinking an 'rx()' or 'regex()' builtin that just compiles a string representing a regex into the regex sub, so you can invoke it on the URI to get matches.
Austin First thing to try is make clean; make.
I really don't understand that namespace thing.
japhb Come to think of it, has someone written a URI language yet?
Tene japhb: although, there's no infrastructure in Parrot to use NQP libraries yet...
So, wouldn't be as useful. 04:02
japhb Tene: What do you mean by that?
(I mean, I think there's nothing in particular that would stop you from doing it. It just may require a little glue PIR.) 04:03
nopaste "cconstantine" at 71.98.77.93 pasted "sample code" (3 lines) at nopaste.snit.ch/18081
cconstantine Austin: that's now causing a 'invoke() not implemented in class 'String''
Austin That's better.
Backtrace?
purl Backtrace is not sufficient?
nopaste "cconstantine" at 71.98.77.93 pasted "backtrace" (8 lines) at nopaste.snit.ch/18082 04:04
japhb Tene, you have clearly spent too much time writing PASM ... unless those bits were cribbed from the old example, I suppose. 04:05
Austin This one isn't even in your grammar. You win.
Tene japhb: which bits?
purl which bits are they?
Austin Looks like you're running code you've produced from the script, no? 04:06
cconstantine cheers!
japhb Tene, the contents of the MORE loop, for example. Using 'ne' and 'eq' as ops, instead of the sugared if statements.
Tene japhb: I mean, if I want to write a general-purpose HTTP library, and I write it as an nqp file, there's no infrastructure to put it in the parrot repo so that people can use it.
japhb: No, I wrote that. I can never remember the sugared syntax.
Austin Try running your compiler with --target=pir
Tene "does it go before, or after, or in between, or ..."
I can't go, like, use HTTP:from<nqp>; 04:07
japhb 'if A OP B goto LABEL'
nopaste "cconstantine" at 71.98.77.93 pasted "make && ./reason --target=pir in-2.rl" (47 lines) at nopaste.snit.ch/18083
Tene japhb: Sure, and I could just look it up... I just can never remember.
japhb I guess my years writing early BASIC variants got me completely used to that syntax. ;-) 04:08
Austin So you're calling the "+" function. Where is it?
cconstantine one of my builtins
Austin is it called "+", or "inline:+" 04:09
sorry, "infix:+"
cconstantine I can call it normally like (+ 1 1)
japhb Tene, it took a minute, but I think I see what you meant about using NQP libraries. It's almost there, but not quite. 04:10
Austin So why is it not a tailcall?
cconstantine it could be 04:11
I invented (^func ...) as a tailcall
explicit tailcal
Austin Yeah, and the example source you gave used ^+ 04:12
So why does the pir not have a tailcall?
japhb Tene, what's your next step with the http GET code, out of curiosity? 04:13
cconstantine Austin: I may have changed the source from what I posted
Austin I see.
nopaste "cconstantine" at 71.98.77.93 pasted "make && ./reason --target=pir in-2.rl" (51 lines) at nopaste.snit.ch/18084
Tene japhb: come up with a silly name and make a library for it; add it to runtime/library
japhb Tene, :-)
Austin Same error? 04:14
cconstantine yeah
Tene MECHA-HTTPDownloader 04:15
Austin Does it print "Finding: +" ?
Tene HTTPDownloaderZILLA
cconstantine and it's definately the '+'.... odd
Austin ^^ 04:16
nopaste "cconstantine" at 71.98.77.93 pasted "new infile (works)" (3 lines) at nopaste.snit.ch/18085
Austin Oh. 04:19
There's your problem.
actions.pm:242
You're converting from a VAR to a String, and pushing it into the tailcall 04:20
Probably because the Past CALL op lets you do that with function names. 04:21
$op.name("+"), etc.
cconstantine right
so just push the $cmd?
Austin So, does it work if you don't do ^+, but just do + ? 04:22
cconstantine with the $cmd = $cmd.name() it doesn't work either time 04:23
wait
scratch that
Austin Yeah, I see that in 'simple' now too.
cconstantine it works for + but not for ^+ 04:24
it came from 'simple'
mikehh all tests PASS (pre/post-config, smoke (#28221), fulltest) at r41540 - Ubuntu 9.04 amd64
Austin Right.
Because simple generates result = "+"(x, 1)
where tailcall generates $P0 = new String("+") ; $P0(x, 1) 04:25
or whatever
So the PIR compiler is resolving the "+" for you
cconstantine yeah
Austin So if you eliminate that conditional block at :242, and just push the function name into the node, what happens? 04:26
cconstantine push $cmd.name() ?
Austin Eliminate that. 04:27
In tailcall, just always do push($cmd)
cconstantine it appears to work :) 04:28
Austin w00t 04:31
Austin is da man.
Go me.
cconstantine w00ts!
I could have sworn I tried that 04:32
Austin You tried it with nobody looking.
cconstantine obviously
I'm reminded of the first story here www.ccil.org/jargon/jargon_47.html#SEC54 04:33
Austin jrtayloriv ping 04:35
jrtayloriv pong
Austin You doing C or Phemer tonight?
jrtayloriv Trying to do phemer. I got it to eval integer constants properly at least. Pretty pitiful state though. Still blocking on the symbol deal ... I must be missing something really obvious :| 04:36
Austin It'll come to you. Can I give you a homework assignment on symbols? 04:37
nopaste "cconstantine" at 71.98.77.93 pasted "recursive lambda" (6 lines) at nopaste.snit.ch/18086
cconstantine so, an anonymous function referencing it'self from the let it's defined in isn't exactly working Is that even possible?
jrtayloriv Austin, Please.
Austin cconstantine: Yes, it's possible. But you have to do more magic. 04:38
cconstantine I create the var before I cal .ast on the val.. is that not enough?
Austin jrtayloriv: In the tcl code, and possibly other languages, coke changes from using LexPad to DynLexpads. I think it's called HLL type mapping. I don't know much about it beyond that it exists. But you need it. 04:39
cconstantine: You keep using the word 'package' in your code. You need to start using the word 'lexical' instead.
jrtayloriv: You need it because you need to be able to insert lexicals into a callers lexpad. 04:40
cconstantine if I claim all symbols are lexical will it fall back on package if it isn't in the lexical stack?
jrtayloriv Austin, Where would I need to do that? 04:41
Austin cconstantine: If you set the symbol_defaults() to have :scope('package')
cconstantine: But you would be better off paying close attention to what I'm telling JT4 04:42
Tene cconstantine: you just need to put the function in a block's symbol table right as it's defined.
Austin Jesse, I have no idea where it's done, or how it's done. I do know that Coke is doing it, because he was talking about it the other night.
Tene Nothing more than that.
cconstantine Tene: could this be an 'immediate' vs. 'declaration' thing? 04:43
Tene cconstantine: No.
jrtayloriv Austin, I am starting to look through the partcl code, as we speak. I was just wondering where I would need to insert a lexical into a *caller*'s pad?
Austin define. 04:44
From the conversation last night, define is basically tcl's upvar with level 1.
jrtayloriv Oh, because define function will include something it top level environment block (which it is getting called from) pad, correct? 04:45
Austin right
cconstantine Tene: I thought github.com/cconstantine/Reason/blob...ons.pm#L93 did that for me
jrtayloriv Austin, Thank you, thank you, and thank you. I'll study partcl tonight and tomorrow (at least) and see what I can glean from the upvar/lexpad code. Whatever I'm trying now is obviously a dead end ... 04:46
Austin I looked at the upvar code. Stay away from that.
He's doing it differently, because he tracks the call stack separately from Parrot. 04:47
jrtayloriv OK -- I still want to look at it, so I can understand what you mean by "is basically upvar with level 1".
Tene cconstantine: yes, but you're doing that *AFTER* the function is already parsed.
jrtayloriv But I'll make sure to treat it like something I should only poke at with a long stick :)
Tene cconstantine: github.com/cconstantine/Reason/blob...mar.pg#L66 is the line where the function is being parsed, and the line *after* it is currently where it would get put in the symbol table.
Austin That's tcl syntax. Find a tcl book. But basically, uplevel and upvar perform actions in the lexical scope of their n-levels removed caller.
cconstantine Tene: so I need another {*} after <var=symbol> ? 04:48
jrtayloriv Austin, OK, so you would recommend then, looking more at how he deals with lexpad/dynlexpad? 04:49
Austin So Tcl's [upvar 1 foo] defines or fetches (I can't recall which) 'foo' from the 1-time removed caller's scope.
Tene cconstantine: you need to register the symbol in the symbol table before you try to refer to do it. That's your goal.
Austin Yes. See if you can figure out how to do HLL mapping to get DynLexPads used instead of LexPads.
Tene cconstantine: So, yes, that's one way to do it.
cconstantine hehe, you say that like theres more than one way ;) 04:50
I'll give that a shot
Tene Sure there's more than one way. 04:51
Just keep in mind the order everything happens in, and make sure that it's *registered* before it's *used*, and you're fine.
cconstantine does the $block.symbol(...) register it? 04:53
Tene Yes. 04:54
cconstantine awesome
mikehh rakudo (0331d60) builds on parrot r41540 - make test / make spectest_smolder (up to 28454 -> #28223) PASS - Ubuntu 9.04 amd64 04:55
partcl r745 builds on parrot r41540 - make test PASS (smolder #28224) - ubuntu 9.04 amd64 04:56
cconstantine you guys are so much help 04:58
how would I go about keeping $var around for val to bind to it? 05:00
Tene cconstantine: just pull it out of the AST again.
cconstantine doing $<var>.shift.ast causes a shift off an empty array 05:01
Austin Yeah, you're destroying that as you read it.
Try using $<var>[0] in the first rule, then shift it off in the second. 05:02
cconstantine *gasp*
I think it might work! :)
dalek TT #1070 created by mikehh++: building parrot with the make -j option fails after r41514 05:12
cconstantine each 'block' adds to the stack doesn't it 05:15
Tene cconstantine: what? 05:23
cconstantine I'm getting stack overflows at around 1k tail calls 05:24
nopaste "cconstantine" at 71.98.77.93 pasted "attempt at 'last'" (11 lines) at nopaste.snit.ch/18087
Tene cconstantine: write a small PIR program that reproduces the problem.
nopaste "tene" at 12.190.38.210 pasted "tailcalls seem to work fine for me..." (10 lines) at nopaste.snit.ch/18088 05:25
Tene cconstantine: I was able to run that program up past 100k without problem. 05:26
over 1M calls
cconstantine right, I doubt tailcall is broken... and I'm allocating stack where I think I'm not
Tene cconstantine: do you have a small scheme program that reproduces the problem? 05:27
and can you give me a link to your repo again?
cconstantine github.com/cconstantine/Reason
Tene I'll look into it. I've got a bit of time before I need to sleep.
cconstantine ok, I've mostly given up for the night 05:28
Tene cconstantine: how can I reproduce the overflow?
cconstantine can you compile 'reason'?
Tene yes 05:29
cconstantine I do a 'make && ./reason in.rl' 05:31
with the stuff in pastebin as the contents of 'in.rl'
Tene I don't see the paste you're talking about.
cconstantine nopaste.snit.ch/18087
nopaste "tene" at 12.190.38.210 pasted "This runs over 1M for me without problem" (3 lines) at nopaste.snit.ch/18089 05:32
Tene I will look at your problem now.
cconstantine so it oes 05:34
so my 'last' is busted
or 'seq'
Tene Yes, every block will add to the stack.
japhb Tene: Starting work on changing 'instructions' section of metadata -- does it look like I'm on the right track? trac.parrot.org/parrot/wiki/Module...taProposal
dalek tracwiki: v8 | japhb++ | ModuleEcosystem 05:35
tracwiki: WIP: Updating metadata instructions section to new design
tracwiki: trac.parrot.org/parrot/wiki/Module...ction=diff
cconstantine Tene: ah, then let creates a block; therefor it adds to the stack on each recurssion
Tene cconstantine: yes.
cconstantine ok, I'll work on figuring out how to fix that later. 05:36
05:36 mokurai joined
cconstantine thanks for all the help :) 05:36
Tene cconstantine: you'll have to find a way to make PCT emit a tailcall for (let), or make the last statement in your blocks something that you can explicitly tailcall. 05:37
cconstantine: I suspect it might be worthwhile to investigate adding this as an optimization to PCT.
cconstantine: I think the place to look is going to be compilers/pct/src/POST/Compiler.pir 05:38
cconstantine oh wow
Tene hm?
cconstantine that's in parrot
Tene Yes
cconstantine 's deep dark heart
Tene japhb: I think there might need to be an optional stage after fetch... unpack. 05:39
japhb Tene: Oooh, good thinking.
Tene japhb: if the fetch stage has a link to a tarball... the unpack stage should be 'tar'
cconstantine ok, I"ll sleep on it. if you feel like working on it don't let me stop you :)
cconstantine is an eastern tmz denizen 05:40
Tene cconstantine: I just might. :)
cconstantine :)
Tene Okay, I'm going to sleep now.
cconstantine 'night
Tene japhb: feel free to leave me a specific task or goal when you finish for the night.
japhb Tene: 05:41
Did you already get HTTP wired into plumage?
Tene japhb: No.
I'll do that, then.
japhb There's your goal. ;-)
Tene 'night
japhb Thanks for all your help,.
good night
Tene np
dukeleto 'ello 05:52
jrtayloriv hoody hoo 05:57
mikehh hi
dalek tracwiki: v9 | japhb++ | ModuleEcosystem 06:03
tracwiki: WIP: Updating metadata instructions section to new design, part 2
tracwiki: trac.parrot.org/parrot/wiki/Module...ction=diff
tracwiki: v10 | japhb++ | ModuleEcosystem 06:13
tracwiki: WIP: Updating metadata instructions section to new design, part 3
tracwiki: trac.parrot.org/parrot/wiki/Module...ction=diff
06:15 uniejo joined
Tene japhb: still up? 06:16
purl Your face, Tene. That's what.
japhb Tene, yes indeed
See most recent version of metadata spec. :-) 06:17
Tene japhb: I'm going over all of the options on were to put the HTTP module, and the others that we need, and I'm trying to remember the issues...
We could:
1) Keep the http downloader modules in the plumage repo
2) add it to parrot as runtime/library/HTTP.pir 06:18
3) add it to parrot as runtime/library/PlumageUtilitiesNamespaceBlah/HTTP.pir
Others?
purl Others are going to be more guidance than specification
Tene japhb: will plumage be shipping with Parrot, eventually? 06:19
japhb 3 is a no-go, I think. When Plumage is slightly more empowered, people will start using it, and it will install itself (and its modules) into Parrot's install tree, presumably. Eventually, the Plumage code base will merge into the Parrot tree wholesale.
I'd say, for now just leave the hackish one in the Plumage repo (option 1). 06:20
A more powerful/complete version would be something that Plumage would be used to install. ;-)
Tene So this *will* end up in the Parrot repo eventually, but not quite yet... 06:21
I feel a little hesitant to grab a large/potentially contested namespace like HTTP and Downloader... 06:22
japhb Allison said that she wants it to cook faster than Parrot deprecation policy would allow.
Tene, so don't. If it sits in the Plumage repo, we can easily rename it before we do the merge. 06:23
Tene I keep thinking we need some kind of prefix... CoreUtils, PlumageUtils, PUT (ambiguous on PArrot/Plumage)...
eh, whatever, okay.
hardly important.
japhb Plumage:: ? We could move the Glue module within that too, I suppose.
japhb shrugs 06:24
Tene Once it's merged, is this HTTP stuff intended just to support Plumage, or is it supposed to be available for other secret plans?
japhb As I said, we needn't solve the naming now. If we only affect the Plumage repo, no conflicts can occur until we start installing Plumage into the Parrot install tree. 06:25
Tene Yeah, okay. I'll drop it. Having trouble sleeping, and this kept coming up.
I'll go review your changes ot metadata.
japhb My thought is that we should assume a hackish, minimalist HTTP client for Plumage's use that will be shipped with it. A full HTTP client, minus the hackishness, would be a project that Plumage would install (and perhaps could then use, if it did nifty stuff) 06:26
np
Insomnia is Teh Suck.
06:27 fperrad joined
Tene Yeah, okay, that sounds good. 06:27
It can' tbe *too* hackish, though... we'll need proxy support, redirects, etc...
japhb: entertained by considering making 'test' required too. 06:29
japhb Tene: yes, agreed (re: proxy and redirects)
Tene, *chuckle*
jrtayloriv Austin, ping 06:30
Austin Jesse? 06:31
purl Jesse is obra
Tene to go without tests, use: { type : "DontTrustMySoftware" }
jrtayloriv I can't find anywhere that Coke is using DynLexPad, although from reading what docs and websites, and the IRC conversation you and coke had, it does seem like DynLexPad is what I want.
But the docs are pretty skimpy on it. Any languages you know of that use it, that I could look at? 06:32
Austin Sadly, I am ignorant of it. I know he was talking about it here on irc, but that's all. 06:33
jrtayloriv ahhh nm -- I need to start looking at tests as documentation from now on ... 06:34
Austin code.google.com/p/partcl/source/bro...&r=558 06:38
In a sub called 'mappings', he maps. It's an interpreter call.
But I can't find any reference to dynlexpad. 06:41
:(
jrtayloriv Austin, OK. So basically, my goal is to determine how to use interp.'hll_map'() to make it create dynlexpad's instead lexpad's, correct? 06:42
Austin No.
jrtayloriv grrrr
Austin Your goal is to determine how to use dynlexpads. Hll mapping may or may not figure in to it.
Tene japhb: this wget implementation is *slow*. 06:43
japhb ?
Tene japhb: try running it on, say, www.parrot.org/files/parrotify_logo.png
japhb I like your non-testing metadata. :-)
Tene, will do in a couple. Head down in some plumage code
jrtayloriv Austin, OK. I am reading the docs for hll_map ... had never heard of it until tonight. Also reading the tests for dynlexpad. Thanks for the pointer -- seems like this is the right area to be looking in for my solution, at least. 06:44
Austin :)
jrtayloriv I could not find DynLexPad anywhere in Coke's code either though. I'll have to ask him later what is going on with that. 06:45
Austin Maybe he hasn't committed yet. 06:46
Tene japhb: if I add a counter to the recv loop, it runs through 13 times real quick, and then stalls on the last iteration.
Austin Anyway, in $PARROT/t/dynpmc/dynlexpad.t I think is everything you need.
Test #2
japhb Trying to read the full number of bytes requested in the recv? 06:47
Or maybe need to see if the socket is ready for read before trying to recv?
Tene Hmm. It looks like it's actually stalling in the 'recv' method. 06:48
looks like it's blocking in the 'recv' in libpthread.so 06:50
It's calling string_make many times. 06:51
06:52 theory joined
Tene Wait... maybe. 06:52
japhb Need to make socket non-blocking?
Tene japhb: no, blocking is the right behavior here.
purl okay, Tene.
Tene purl: forget blocking 06:53
purl Tene: I forgot blocking
Tene japhb: I'll work it out... I'm just talking out loud here. Ignore me.
japhb Tene, sorry if silly suggestions, this window is only getting 5% of my brain at the moment. :-)
Tene japhb: nah, don't worry about it. 06:54
06:54 iblechbot joined
Tene Hmm. It works instantly connecting to my local webserver, even on large files... 06:55
So maybe it's network errors.
Eh, I'll troubleshoot it later.
jrtayloriv Austin, It looks like I actually do want to use hll_map to use dynlexpad instead of lexpad, like so: interp.'hll_map'(lexpad, dynlexpad) ... that should make it where when I do store_/find_lex it uses the DynLexPad vtables instead.
Tene jrtayloriv: make sure to do that at :immediate time, iirc. 06:56
Austin Right. There's another caveat, too. All your subs that might get lex-ified by define, or anything else that makes up its own names at runtime, has to be declared with :lex
For now, that's just the $TOP block 06:57
jrtayloriv Austin, Right, saw that (about :lex) too.
Tene, My TOP method creates an PAST::Block.new(:blocktype('immediate') ...) is that what you mean?
Tene jrtayloriv: Yes. 06:58
jrtayloriv ok
Austin Tene: the tests show it running in a block marked :init
Tene Austin: Oh, okay. nm then. 06:59
Austin: I must have been thinking of hll_mapping subs or something else.
Austin shrugs.
You know more about it than I do. (Since everything I know has come in the last 10 minutes...)
jrtayloriv: This is one of those compiling/interpreting moments. If your compiler is an interpreter, you'll want to make sure to add this to the startup of the compier, too. 07:00
*compiler
07:04 chromatic joined
jrtayloriv Austin, What is the "this" that I need to add to startup of the compiler? 07:05
Austin hll mapping the lexpads
If you're running it as an interpreter, then you might encounter a (define) in your interpreting. 07:06
jrtayloriv doesn't grok ... is going to see if another cup of coffee helps ... brb 07:12
japhb Tene, plumage.nqp and metadata/* now updated to spec rev 1. Also, plumage now checks spec uri and version before working with the metadata. 07:19
Austin, see above WRT Close.json
Tene japhb: I also need to make a metadata file for parrot-mysql repo 07:20
japhb Tene, nodnod. Should be cleaner with the new rev.
Austin, actually, you probably just want to update your copy from gitorious.org/parrot-plumage/parrot...Close.json 07:21
Austin Works for me. 07:22
japhb (This copying won't be necessary some time from now, but we're not there yet. :-)
Tene japhb: pretty sure I have the tree-merging algo done in my head, just need to write it down.
japhb Tene, cool beans 07:23
Tene japhb: also, this http work will make it possible to grab the referred-to url, for purposes of merging.
dalek ose: r169 | Austin++ | trunk/plumage/Close.json:
Incorporated japhb's changes for plumage.
Austin Committed. Thank you. #169
japhb Austin, np 07:24
07:24 baest joined
japhb (accelerating progress)++ 07:24
Tene japhb: maybe I need to go make a curses-based app to build metadata files... ;) 07:25
japhb Tene, heh. We will certainly have tools for generating them some time down the road, but you're welcome to play around whenever. 07:26
Once we can actually run through a full build/install, I'll probably write something small to import the proto projects ....
Or you can, if you like.
Tene, by the way, don't hesitate to add tasks like that to TASKS, if you're not planning to do it immediately. 07:27
Tene japhb: I meant it as a joke. :P 07:30
japhb Tene, I tend to interpret such jokes in the same way Parrot was. ;-)
OK, I think it's really, truly, finally time for me to sleep. G'night all. 07:42
Tene I'm gonna go try sleeping again too. 07:43
'night
08:04 gaz joined
dalek rrot: r41541 | dukeleto++ | trunk (2 files):
[t][TT #763] Fix is_deeply() on hashes with undefs and add tests
 08:31
jrtayloriv night #parrot 08:35
jrtayloriv zzzzzzzzzzzzzzzzz
ttbot Parrot trunk/ r41541 MSWin32-x86-multi-thread make error tt.ro.vutbr.cz/file/cmdout/108126.txt ( tt.ro.vutbr.cz//buildstatus/pr-Parrot/rp-trunk/ ) 08:37
dalek rrot: r41542 | dukeleto++ | trunk/t/library/test_more.t:
[t] Add another test for is_deeply() on hashes with undefs
 08:40
TT #763 closed by dukeleto++: Test/More.pir's is_deeply() needs work with hash corner cases 08:44
ttbot Parrot trunk/ r41542 MSWin32-x86-multi-thread make error tt.ro.vutbr.cz/file/cmdout/108167.txt ( tt.ro.vutbr.cz//buildstatus/pr-Parrot/rp-trunk/ ) 08:47
08:54 Austin_Hastings joined 09:00 masak joined
mikehh I am getting an intermittent failure with t/pmc/os.t test 9 - it failed yesterday while running fulltest in testr and just now while running smoke (#28233) - but on retesting it passes 09:37
All other tests PASS (pre/post-config, smoke, fulltest) at r41542 - Ubuntu 9.04 amd64 09:47
10:05 slavorgn joined
dalek TT #1071 created by mikehh++: Intermittent failure with t/pmc/os.t test 9 10:35
mikehh re-ran the my test run: 10:51
all tests PASS (pre/post-config, smoke (#28234), fulltest) at r41542 - Ubuntu 9.04 amd64 10:52
dalek ose: r170 | Austin++ | trunk/src/close/Compiler/ (19 files):
checkpoint: "Class"-ifying node, type, symbol, message
 10:55
11:00 einstein joined
einstein I have got a question, certain hll register there own type of Integer instead of the default parrot integer.pmc type. 11:07
my question are there any hll which also store extra information/metadata onto their own Integer type?
or does these hll only add extra "functionalities" to there own implementation of integer? 11:09
Austin einstein: I just had a look at tcl, and it just adds extra functionality 11:11
Patterner My "make" is broken. I have to run it 3 times before it doesn't compile anything... 11:13
einstein Austin: ok that could to know, so if class which an integer attribute can store the integer value instead of the, and can recreate the registered hll integer type if it is the integer attribute is retrieved 11:14
moritz Patterner: parallel build is horribly broken atm, try without any -j options (if you have any, that is)
Patterner just "make"... what is the "-j" default? 11:15
einstein Austin: ok that could to know, so if I have a class with an integer attribute then can I store the integer value instead class object of the registered hll integer type, and can recreate the registered hll integer type if it is the integer attribute is retrieved
Patterner (and yes, my Build script does "make distclean" first :) 11:16
moritz -j defaults to 1
Austin einstein: correct. In fact, you could just return a parrot:int to an int register, if it seems better to do that. 11:17
einstein austin: thanks for the information 11:18
11:33 jsut joined 11:42 bacek joined 12:02 whiteknight joined
whiteknight good morrow, parrot 12:04
purl msg Tene basically add similar methods to Socket like what is in FileHandle. "Readline" is the one I hear most requested, but "read", "put", "printf", etc would all be useful. 12:05
purl Message for tene stored.
Austin Howdy, Whiteknight 12:11
whiteknight good morning Austin 12:12
12:13 ruoso joined
whiteknight just took a call from a customer who was having "problems with the website" in their browser: IE6 12:13
"sir, your browser IS the problem"
"please take your computer outside and light it on fire"
Austin heh heh heh 12:15
jonathan ewww IE6
Austin NS4 was good enough for grandpa...
jonathan There comes a point where the cost to support it is more than you'll ever reap in profit from those using it...
Austin Consider Opera: the zero point is now! 12:16
jonathan Austin: That's why I don't test sites I build in Opera.
slxix everyone should view their website in w3m. 12:17
12:20 cconstantine joined
whiteknight the problem is that our customers almost exclusively come from the freight rail industry, who aren't exactly known to be "on the cutting edge 12:25
" or "technology literate"
and everything is so javascripty in modern websites, it's amazing that IE6 is even still functional for everything 12:26
dalek a: 62fbfec | fperrad++ | (2 files):
refactor some test targets
 12:29
a: 3ef4945 | fperrad++ | :
Merge branch 'master' of git@github.com:fperrad/lua
12:34 cconstantine left, cconstantine joined 13:03 bluescreen joined 13:04 shorten joined
fperrad ping pmichaud 13:27
purl I can't find pmichaud in the DNS.
pmichaud fperrad: pong 13:48
fperrad pmichaud, have you seen my message about parrot_nqp ? 13:49
pmichaud fperrad: not yet -- just got on the computer this morning
14:02 bluescreen joined, PacoLinux joined
Patterner I have a problem with libpcre... the test fails. 14:04
the test gets "/usr" from "pcre-config --prefix", but /usr/lib/libpcre.so is a gentoo "fake" dynamic library (a GNU ld script, pointing to /lib64/libpcre.so.0). Any idea how to handle that? 14:06
14:10 payload joined
NotFound Patterner: some time ago I searched some forums about that problem. Looks like the official answer from gentoo is that linker scripts doesn't harm anyone. 14:19
Patterner sigh
Patterner feels harmed :)
14:21 einstein joined
cognominal moritz++ # Tie::Hash::Indexed as a way to learn ties and XS 14:24
14:29 Psyche^ joined 14:34 iblechbot joined 14:40 theory joined 14:49 AndyA joined 15:01 Andy joined
moritz hm? 15:19
not mine
moritz not mhx
dukeleto 'ello 15:42
15:47 zerhash joined 15:50 allison joined 16:00 theory joined
cotto_working hi dukeleto 16:13
16:45 kjeldahl_ joined
einstein can it be that the vtable function instantiate_str can be deprecated, it is only used by imcc/pbc.c ? 17:05
japhb whiteknight, You may already have heard this, but several large web sites are delisting IE 6. Most are starting with a message suggesting upgrade, then not supporting functionality other than the most basic on IE 6, then finally dropping it altogether. 17:07
And by big I mean Google, Yahoo!, Digg, ... 17:08
moritz it's about time
cotto_work einstein, that doesn't sound like a bad idea. Please file a ticket about it. 17:09
japhb The only reason Microsoft is still supporting it is ... get this ... the extended support for Windows XP caused by the Vista backlash.
einstein cotto_work: ok i will do that 17:10
cotto_work einstein++
it's great to get rid of old cruft
cconstantine has a friend who is stuck with IE6 at work. Hopefully the stuff japhb is talking about forces the hands of IT departments that still require the use of IE6. 17:18
dalek TT #1072 created by jessevdam++: deprecate vtable function instantiate_str
NotFound japhb: I still have a system with a win98 SE partition X-) 17:21
moritz I installed Win98 so often that I knew the registration key by heart at one point 17:22
jrtayloriv Basically, when I do interp.'hll_map'(lexpad, dynlexpad) it will make it where when I explicitly do .lex 'foo', it will use the LexPad vtables, but if I do store_lex 'baz', without declaring 'baz' as .lex, then it uses DynLexPad vtables for store_/find_lex. Is this correct? 17:30
Or does it just always use DynLexPad vtables in either case?
cotto_work cconstantine, I doubt it. There are a lot of poorly-written legacy apps that depend on ie6.
jrtayloriv actually got a "you have to use Internet Explorer to view this site" message the other day, at a customer support site ... jaw dropping 17:32
cconstantine cotto_work: a newb can dream? 17:35
japhb NotFound, moritz: ouch.
cotto_work and wake up 17:36
whiteknight japhb: yeah, it's something we're going to do here eventually too: disallow IE6 17:37
because our website is so JS-heavy, older browsers just can't do it
japhb whiteknight, for that problem, I just let jQuery handle the pain.
whiteknight we don't use JQuery, even though I keep suggesting it :( 17:38
japhb Mind you, IE 6 is by far the slowest browser they still support, but every release they somehow find a way to make it faster anyway, which boggles me.
whiteknight, :-(
17:39 uniejo joined 17:40 mokurai joined
jrtayloriv <-- jQuery makes this guy's life way easier. 17:41
17:41 particle joined 17:43 Austin joined
NotFound "a sales assistant at supermarket Tesco told the reporter that netbooks got their name because 'a Japanese man on a plane fell asleep with a laptop on his thighs and was horribly burned, so the industry has dropped the name laptop" 17:46
17:51 joeri joined 18:01 chromatic joined 18:02 iblechbot joined 18:07 darbelo joined, japhb joined 18:15 kyle_l5l joined
nopaste "Austin" at 98.235.55.43 pasted "Help! Why does $P0 end up null?" (43 lines) at nopaste.snit.ch/18093 18:19
Austin Can someone with more PIR savvy than me take a look at the nopaste (nopaste.snit.ch/18093) and help me understand why $P0 winds up null? 18:20
pmichaud looking
whiteknight ditto
pmichaud rebuilding, too 18:21
whiteknight Austin: it's null at the say?
Austin Yeah, the one I commented out gets a NULL PMC access
pmichaud it shouldn't.
whiteknight you arne't setting $P0 to anything
pmichaud it's a lexical, the "store_lex" should be changing it 18:22
whiteknight store_lex overwrites the PMC pointer, doesn't copy to $P0
pmichaud lexicals are mapped to context registers
the PMC pointer *is* $P0
see, for example, $P3 later in the same code snippet 18:23
Austin Note: This was intended as a test to see what happened when a lexical was re-declared. Is IMCC doing something?
pmichaud oh
right
the problem is that you're redeclaring 'foo'
that's a no-no
redeclaring foo causes it to lose its original mapping 18:24
imcc should probably complain when that happens
chromatic I can work on that. 18:25
Austin Okay. I modified the test to print set P6 (which I .lex foo later) early, and then to print it after the .lex foo P0, and sure enough it gets the foo values. 18:26
So IMCC is the one responsible for the .lex mapping, and it quietly remaps the binding. 18:27
pmichaud correct. 18:28
Austin So there can't be two same-named lexicals in the same scope, unlike package's
*package's = package symbols
whiteknight pmichaud: as a general inquiry into lexicals, why does the current system use LexInfo/LexPad pairs, and why does it require lexicals to be declared at compile time? 18:30
pmichaud whiteknight: I'm not the designer, I just repaired the installation (more) :-)
jrtayloriv I was wondering the same thing. Why not always use DynLexPad? 18:31
pmichaud s/installation/implementation/
Austin Speaking of package symbols, does anyone know what happens if I have a sub 'foo', and I bring in another declaration of 'foo' via load_* or eval?
pmichaud Austin: it overwrites the original, usually
whiteknight Austin: locusts
pmichaud whiteknight: I think the point of the current implementation was exactly the register mapping 18:32
chromatic LexInfo stores the lexical information for a Sub: which lexicals does the sub have.
pmichaud for example, if I have .lex 'foo', $P0
chromatic LexPad stores the lexicals for an *invocation* of the Sub.
Think recursion.
pmichaud and then I call a sub that does store_lex 'foo', $P2
whiteknight chromatic: yes, I think I know *what* the implementation does, just not *why* it was designed this way
pmichaud then $P0 is changed
without having to do a re-fetch
in order for that to happen, we have to have a structure that maps lexical names to registers 18:33
and that's what LexInfo does
I don't think it's quite possible to do that with dynlexpad
at least, not as a declaration. We'd have to have some sort of opcode to associate a lexical symbol with a register
and getting that mapping correct is likely to be tricky 18:34
whiteknight okay, that makes a bit more sense then
pmichaud jonathan++ and I have also been playing with the notion that subs should also have a "default LexPad" that is used as the template for initializing its lexicals when the sub is invoked
Austin Pmichaud: So declaring foo twice in the same file binds the name to the first one, and loading it in later overwrites?
pmichaud Austin: rebinds, I think. 18:35
Austin Sure.
pmichaud I think the first one is still in memory, unless/until GC'd
Austin Is this one of those cases where "get_*_global 'foo'" will find the newly-loaded foo, but anybody that calles 'foo'() will still point to the original? 18:36
whiteknight that's actually a good question
pmichaud I don't think so.
It used to be that way, yes.
but I think that mis-optimization got removed.
Austin It still is, if you leave out the loading.
whiteknight would definitely like to see a test to verify operation
pmichaud test coming up
Austin At least at 41311
pmichaud you're correct, Austin -- the imcc misoptimization is still in place 18:37
nopaste coming 18:38
nopaste "Austin" at 98.235.55.43 pasted "Name-switching code" (66 lines) at nopaste.snit.ch/18094
"pmichaud" at 72.181.176.220 pasted "IMCC still optimizes 'foo' incorrectly" (23 lines) at nopaste.snit.ch/18095
Austin In the nopaste above, I found that I had to call Test::foo instead of foo, or the swapping would be ignored. 18:39
pmichaud correct
IMCC translates the call to 'foo' into a direct call
instead of doing the symbol lookup
2nd nopaste coming
Austin What does that mean?
purl That boy needs therapy.
Austin "direct call" ?
pmichaud 2nd nopaste coming
Austin Sure.
whiteknight at compile time, IMCC translates the call to 'foo'() to a direct link with the sub and doesn't check that the runtime bindings are the same 18:40
nopaste "pmichaud" at 72.181.176.220 pasted "IMCC still optimizes 'foo' incorrectly" (46 lines) at nopaste.snit.ch/18096
Austin I get that, whiteknight, because it matches the data. But I don't understand how it's possible.
pmichaud look at the trace output, and the differences between the calls to 'foo' and 'bar' 18:41
whiteknight the literal sub 'foo' is stored in the constants table in PBC
pmichaud er, 'baz'
whiteknight and the call to 'foo'() calls the item from the constants table
pmichaud the call to 'baz' takes place by doing "find_sub_not_null" (which looks it up lexically and in the package)
the call to 'foo' has been "optimized" by IMCC, which says "oh, I know what 'foo' is, so I'll just jump to it directly"
the problem is that IMCC _doesn't_ know what 'foo' is at runtime, and so it's calling the wrong sub. 18:42
Rakudo gets hit with this problem a lot.
whiteknight do we have a ticket?
chromatic doesn't think it's a bug. 18:43
whiteknight chromatic: so a direct call 'foo'() should be a compile-time binding?
chromatic Yep.
whiteknight (I don't disagree, just want clarification)
okay, so a dynamic call should always be $P0()?
pmichaud chromatic: even if that sub is replaced later?
chromatic If your language supports replacing those bindings, you get to do some extra work. Seems fair to me. 18:44
whiteknight pmichaud: technically only the entry in the namespace has been replaced, the literal sub hasn't been
pmichaud whiteknight: sure, but the default meaning of 'bar'() is to do a name lookup
which also looks in the lexical scopes as well
Austin I've been hedging around this issue with JRTaylorIV, and to a lesser extent cconstantine.
whiteknight pmichaud: I neither agree nor disagree. What we expect that syntax to mean is dependant on the user
pmichaud (looking in the lexical scope was fixed by chromatic++, btw) 18:45
chromatic I think I deleted two lines of code to make that work.
whiteknight so long as both ways are possible, one is the default, and everything is documented, that's what we have
pmichaud if the "direct call to 'foo'" is likely to remain, then I think I'll switch PCT so that it always does a name lookup 18:46
Austin Pmichaud: Don't do that.
Please?
pmichaud Austin: why not?
whiteknight a flag to specify behavior might be easier
Austin There's a valid way to configure either behavior, and one is definitely faster than the other.
chromatic Much, much faster.
whiteknight there is plenty of room to opimize both codepaths though 18:47
Austin Leaving the 'foo' as a var node under call produces the lookup, while moving the name into call.name() produces the other.
The only problem is with tailcall.
pmichaud the find_name opcode is that slow...?
whiteknight I would like to see something like a :binding("static"|"dynamic") flag there
Austin I have no idea. But I'm sure two opcodes are slower than one.
whiteknight give PCT either as a default, and let the user decide if they want something else 18:48
pmichaud I can't tell you how many hours jonathan and I have lost because of this particular IMCC misoptimization
Austin How so?
chromatic The find_name opcode at runtime is much slower than the find_name opcode at runtime for every call.
Wait, let me rephrase.
Austin chromatic: :)
pmichaud just because we define a method called 'list', all of our calls to a "list" function end up being mis-optimized to call that method.
chromatic The find_name opcode at runtime for every call is much slower than the find_name opcode once at compile time.
pmichaud, that sounds the like "Don't store methods in namespaces" bug. 18:49
whiteknight pmichaud: that problem should be resolved by the :nsentry stuff, no?
pmichaud should be, isn't.
Austin WhiteKnight: Currently, (call name=foo (arg1) (arg2)) produces the static binding, while (call (foo) (arg1) (arg2)) produces the dynamic.
pmichaud I'll hold off on a pct switch for now then, yes.
whiteknight Austin: so long as there is a way for the compiler-designer to specify, I don't care about syntax or which is the default 18:50
pmichaud however, consider the following Perl 6 code:
sub foo () { say 'foo'; }
chromatic Remember that this isn't callee-side binding; it's callee-side binding.
pmichaud sub bar () { say 'bar'; }
&foo := &bar;
foo();
If IMCC optimizes the call to 'foo', then we end up with wrong results. 18:51
whiteknight chromatic: rephrase?
chromatic Don't put the "I might rebind this!" flag on the sub that you might rebind.
Put it on the sub where you call something that might be rebound.
pmichaud I don't follow.
whiteknight I think chromatic is stuck in an infinite loop today! :)
"Don't do X, do X!" 18:52
PerlJam whiteknight: that's not what he said :)
pmichaud I don't quite understand what he said. :)
whiteknight don't put the flag on the sub you might rebind, put the flag on the sub that might be rebound 18:53
Austin What does 'find-sub-not-null" do?
chromatic In pmichaud's example, don't put a flag on foo() or bar(). Put the flag on whatever *calls* foo().
pmichaud what "flag"?!?
I totally don't understand.
"put a flag on whatever calls foo" simply means "put a flag on every call"
chromatic If you add a flag to say "Never allow static lookups."
pmichaud because every call in Perl 6 could be rebound
Austin: looks up a symbol by name in the lexical, package, and global scopes, throws an exception if not found (i.e., if null) 18:54
if there's something that is "a flag set on every instance", then having the flag set should be the default :) 18:55
Austin Okay.
pmichaud thus I think PCT should default to "the name lookup behavior is the default, and you put a flag if you want the optimized behavior" 18:56
Austin So if I have a lexical variable masking a global sub, it's going to try to call the variable?
pmichaud Austin: yes.
PerlJam pmichaud: that sounds backwards too
pmichaud PerlJam: why?
PerlJam (from an optimization standpoint)
pmichaud we're supporting _dynamic_ languages 18:57
in a dynamic language, rebinding is the norm
chromatic Rebinding *function calls* is the norm?
pmichaud when I say "foo()" in Perl, I'm calling the function associated with the current &foo symbol
whiteknight so long as there is a flag provided to keep the current behavior, I see no issue with this 18:58
pmichaud I'm not calling the function that happened to be called foo at the time compilation took place
Austin chromatic: putting a function into a lexical variable seems pretty common for the pure-functional guys
chromatic That binding is statically bound in Perl 5 though.
Austin "Parrot is designed to be the optimal virtual machine for implementing Perl 5" 18:59
-- pmichaud @ yapc
TimToady it's statically bound in Perl 5 only up to the point you say *foo = sub {...} 19:00
chromatic Sure, but consider \\&foo *before the declaration of* foo.
PerlJam TimToady: heh ... I was just typing an that in an example :)
Austin So it seems like an obvious question is, Is there a language that allows functions and variables with the same name? 19:01
chromatic Perl 5.
Austin C
PerlJam chromatic: I thought I understood you until just now.
chromatic my $x = \\&foo; 19:02
*foo = sub { ... };
Austin Chromatic: I don't think p5 allows that. The sigil separates them, no? 19:03
chromatic Assume no 'sub foo { ... }' in scope.
Taking a reference to a sub that does not exist binds to the sub that will exist, if it does exist at the point of dereferencing the sub ref.
Austin My head just exploded.
TimToady trying to make any sane point via Perl 5 examples seems...odd
pmichaud heh 19:04
chromatic My point is this: "always static" and "always dynamic" are both wrong.
whiteknight chromatic: we don't want IMCC or PCT to flip a coin
Austin whiteknight: That's what it's doing now.
chromatic I'm not suggesting random behavior either.
whiteknight chromatic: are you saying what I'm saying then? Use a flag? 19:05
Austin "If I see foo in scope, heads. Otherwise, tails."
chromatic I don't know how you support that Perl 5 behavior with a flag.
Austin What perl5 behavior?
pmichaud if I want to always bind to a specific sub, then use .const 'Sub' $P0 = 'foo'; $P0()
PerlJam Austin: the example he just gave
Austin PerlJam: I saw it, but I don't understand how it applies. 19:06
pmichaud _which_ PCT already supports, I might add. :)
19:06 cotto_w0rk joined 19:07 japhb joined, mberends joined
whiteknight pmichaud: how? 19:08
chromatic If our static bindings were to a *storage location*, we could probably support everything appropriately.
pmichaud whiteknight: I don't remember the exact steps -- essentially one takes a reference to the PAST::Block structure and invokes it 19:09
whiteknight You mean like a p5 typeglob?
chromatic Only by rough analogy, yes.
pmichaud essentially one creates a PAST::Val that refers to the block
at least, ISTR it's something like that. I know that Rakudo uses it.
Austin If IMCC optimized the 'foo' into a reference to the namespace entry, rather than the const table, I think.
pmichaud that's better, yes. And if we can prove that there's no 'foo' lexical intervening. 19:10
Austin Umm, no.
The lexical bit is what gives me the willies.
pmichaud why?
chromatic It's less scary if lexicals have sigils. 19:11
Austin :)
19:11 theory joined
Austin If I'm calling "foo"(), I expect a sub named foo. Not a reference from a variable named foo. That's where I'd put the onus on the compiler. 19:12
pmichaud what does "sub named 'foo'" have to do with namespaces, though? 19:13
i.e., why are namespaces more magical than lexpads as far as subs and references are concerned?
Austin I'm in a namespace. I call foo. I get the foo in this namespace. 19:14
pmichaud you're also in a lexical scope, though.
why shouldn't you get the foo that is in the current lexical scope?
chromatic His language doesn't look up functions in lexical scopes.
pmichaud sure 19:15
Austin No, it's a more legitimate question than that.
pmichaud but saying that functions can only exist in package scopes sounds... limiting
Austin I think a large part of it is "that's the way it currently is".
PerlJam Austin: where I live, there are several law enforcement agencies that have overlapping jurisdiction: city police, county sherrifs, constables, and state trooper. Just think of lexicals and packages in the same way :)
Austin PerlJam: Sure, and when you have an issue involving real property (real estate), it's always the Sheriff. 19:16
pmichaud Austin: would you also expect to look up the global foo as well?
Austin Pmichaud: global? 19:17
purl global is mankz.com/code/GlobalCheck.htm
pmichaud sure. right now 'foo'() looks in the lexical scope, package scope, and global namespaces for a function called "foo"
Austin By global, do you mean hll root namespace? 19:18
pmichaud that's why one can call "_dumper($P0)" from any sub and have it work
I'm not sure exactly where the global namespace lives. I just know that there is one.
and that it's separate from the current namespace
Austin Man, talk about learning something new every day.
pmichaud and (I think) that it's not hll-dependent.
Austin So how do I put things in the global namespace? 19:19
pmichaud not sure.
Austin :)
pmichaud To my knowledge, I've never done it, because that would be Evil.
Austin If *you* don't know how to get things in there, then I don't mind if it gets searched, because there's nothing in it. :)
Except _dumper, apparently. 19:20
19:21 theory_ joined
whiteknight So maybe that's just the answer: we have too many overlapping jurisdictions, and we must always manually specify 19:21
pmichaud I don't mind that too much for imcc. PCT will have a default, though. 19:22
whiteknight the invoke PAST node would have a :binding flag that takes "static", "namespace", or "callchain" or "lexical"
pmichaud too complex
currently function calls are PAST::Op with :pasttype('call')
Austin Also not compatible with current design 19:23
pmichaud there would be a :pasttype('callstatic') which performs the static version.
and perhaps generates an exception if there's no sub of the given name in the current compilation unit 19:24
Austin logs?
purl rumour has it logs is irclog.perlgeek.de/parrot/
19:24 preflex joined
pmichaud keep in mind that the behavior becomes somewhat different if 'foo' is a :multi 19:25
Austin Does it? 19:26
If IMCC optimizes the call, what happens to multi-ism? 19:27
Also, if I define 3 multis 'foo', and then create a lexical foo, what happens?
Can I create a lexical multi?
(If so, what happens if I define a package sub foo non-multi, and then define a lexical multi foo?) 19:28
whiteknight too many questions!
IMCC handles multies correctly, I'm pretty sure about that 19:31
and no idea about the multies and the lexical. Sounds like we need a test about that
alhough I suspect all the multis disappear
pmichaud a lexical would hide the namespace multies 19:32
(that's arguably the correct behavior)
Austin Whiteknight: If IMCC bind a particular multi, that'll be wrong if I load more at runtime.
pmichaud :multi subs are held as a single MultiSub PMC 19:33
Austin Okay, so the load will jam the new ones in, or overwrite pmc?
*overwrite the pmc
pmichaud a load will take any additional :multi's and add them to what exists
I'm not sure what happens if the load contains non-:multi s of the same name. My guess is that it would re-bind in that case, or throw an exception. 19:34
Austin :)
pmichaud looks like it reinds 19:35
*rebinds
Austin So if lexical hiding multi is the right behavior, what's the usage model for lexicals? It can't be the perl example, since perl supports multi's, right?
nopaste "pmichaud" at 72.181.176.220 pasted "loading a non-multi rebinds" (46 lines) at nopaste.snit.ch/18097 19:36
pmichaud wait, that was a miskey
renopasting
Austin <whew> I was desperately looking for multi-ism anywhere... 19:37
nopaste "pmichaud" at 72.181.176.220 pasted "loading a non-multi rebinds" (18 lines) at nopaste.snit.ch/18098
pmichaud in Rakudo's case, we won't end up having any package calls anyway
all sub calls are lexical
(unless explicitly packaged scoped)
i.e., foo() always ends up doing a lexical foo() 19:38
adding a multi creates a new MultiSub object that includes the existing multi candidates
Austin So how do multi's work? Are you creating a lexical pmc multisub foo?
pmichaud sure 19:39
we already create lexical subs when someone does "my sub foo() { ... }"
Austin Why not just make package foo tailcall lexical foo?
whiteknight I think the best practices rule needs to be that loadable libraries should generally define their own namespaces, and not just dump into an existing namespace 19:40
Austin It seems like it would solve your list method problem.
pmichaud so, instead of a find_name opcode, you're suggesting we replace it with an extra sub/tailcall? 19:43
that'd be worse.
*much* worse.
whiteknight I still hold out hope that the PCC refactors are going to produce lots of opportunities for optimizations 19:44
pmichaud sure, but I'll claim that as a general rule invocation of a single opcode will always be more efficient than a PCC call 19:45
chromatic I agree with pmichaud.
whiteknight Oh, i don't disagree with that, in general
although our ops aren't exactly atomic, and some of them are very complex indeed 19:46
depends which sub and which PCC call
whiteknight nitpicks
pmichaud so, in the final analysis, while I disagree with imcc "optimizing" calls to package subs the way it does now, I'm not going to argue strongly for a change. But PCT might change so that it expresses the behavior I think should be the default, while providing an option to get back to the imcc behavior. 19:48
and to completely describe the existing imcc behavior, it's... 19:49
(1) use any lexically available sub
(2) use any statically available sub in the current package and compilation unit
(3) use a dynamically looked up sub in the current package (but not in the compilation unit) 19:50
(4) look in the global namespace
(5) throw a "sub not found exception"
I'm going back to my regex engine updates :) 19:52
19:56 bacek joined
whiteknight Want to hear something funny? 20:00
Austin A horse walks into a bar...
the bartender says...
whiteknight I don't know a lick about SQL or databases, and I'm my companies lead DBA now apparently
Austin "Why the long face?"
20:00 kjeldahl_ joined
Austin Congratulations! 20:00
purl Congratulations! are you joel?
whiteknight Austin: not a congratulatory thing 20:01
I've officially risen to my level of incompetence
Austin Why not? DBA's get a boatload of money.
whiteknight not this DBA
pmichaud I think "Congratulations" might be in the same sense of "funny" :)
Austin There's your problem, right there.
pmichaud++
And thank you, patrick, for that explanation of IMCC's behavior. I had no idea about the lexicals.
whiteknight exactly: no more money, lots of additional responsibility, and an increased probability that they're going to discover that I'm a bozo and fire my ass 20:02
pmichaud it just does whatever "find_name" does.
whiteknight: in my limited experience, there's an incompetency tipping point that once you reach it is followed by rapid increases in pay and status for little extra knowledge and achievement. You may be reaching that tipping point.
whiteknight and to think I originally responded to a job advertisement for an "embedded C programmer" 20:03
Austin :)
chromatic S/Andrew/Peter/
pmichaud of course, it all comes crashing down in the end, so invest wisely while on the upward slope.
Austin Andrew, what do you want to be doing?
NotFound whiteknight: They embed you in something?
whiteknight Austin: "embedded C programmer"
chromatic In Afghanistan, NotFound.
whiteknight :)
Austin NotFound: Testing the linear accelerator... Put a C programmer in this end, press the button, and ...
whiteknight and here I am developing websites in C# and doing database/SQL nonsense
Austin Would you like to go on a job interview? 20:04
NotFound The Large Hired programmer Collider, I suppose.
whiteknight Austin: entirely possibly. Have one handy?
Austin Maybe. I have some friends in PennDel doing C/C++ embedded. 20:05
whiteknight Penndel? My wife would cry with Joy
love you so much she might accidentally become your wife 20:06
Austin It's not bit-twiddling - it's embedded linux - but there's some DSP and gnarly stuff.
whiteknight so watch out
Austin You know, I'm laughing and I haven't even written out this next sentence:
It's for the technology arm of an MLM company. 20:07
Austin laughs hysterically.
whiteknight that is pretty funny
Austin You'd be safe: they make phones.
whiteknight Anything would be a step up from what I'm being paid now 20:08
I didn't know receiving bits of string each month satisfied minimum-wage laws 20:09
Austin Sure. How do you think the government keeps all those pigeons on payroll.
darbelo Pigeons? 20:11
purl Pigeons on the grass, alas.
Austin Okay. What was I doing when all this started? 20:13
Oh yes, name collisions.
whiteknight Austin's train of thought: 30 minutes late
Austin Still better than flying. 20:14
whiteknight and on that note, I'm leaving. Later
Austin purl, msg jrtayloriv There was an interesting discussion on name lookup today at about 18:38 UTC. irclog.perlgeek.de/parrot/2009-09-28#i_1548640 - there's an imcc bug that stands in the way, but function call's are supposed to look in lexical vars and then package. This should make the library issue easier to deal with. 20:17
purl Message for jrtayloriv stored.
darbelo Are the "GDB functions" in src/debug.c used by anyone?
bacek Good morning 20:21
purl Here I am, brain the size of a planet, and all they say is 'Good Morning'
darbelo Hmm. Is src/debug.c used by anyone? 20:22
NotFound darbelo: by parrot_debugger, mainly 20:23
darbelo The second question was retorical :) 20:24
jrtayloriv Austin, I'm here. And I've been trying to follow it. Just didn't want to interrupt the conversation and kill the momentum by having people explain simple things to me ... I'm slowing going through each thing that was said and trying to grok.
darbelo It was meant to be a commentary on the state of our debugging facilities :) 20:25
Austin Ok. I wanted to make sure you were aware of it.
darbelo rt?
purl rt is just RT (bestpractical.com/rt) or (:rt3) or (: rt bugs) or Obra's trouble ticketing system or the first IBM RISC workstation (www.contrib.andrew.cmu.edu/~shadow/ibmrt.html) or the bombsquad or the Right Thing or very very capable and open-source or an application framework that bundles a ticketing system or obra's baby or SOOOO slow :-S or email mailto:perlbug-owner@perl.org for access
jrtayloriv Yes, thank you.
NotFound darbelo: I located several problems using parrot_debugger
bacek seen fperrad
purl fperrad was last seen on #parrot 6 hours, 36 minutes and 36 seconds ago, saying: pmichaud, have you seen my message about parrot_nqp ?
Austin Did you make any progress on the lexpad thing? 20:26
bacek msg fperrad I can't build latest Lua. make: *** No rule to make target `t/lua-TestMore/src/Test/More.lua', needed by `Test/More.pir'. Stop.
purl Message for fperrad stored.
jrtayloriv Austin, I've been doing a lot of reading, trying to make sure I understand how LexPads and DynLexPads work. So actual code written, no, but the reading is starting to at least help me understand the problem better. 20:27
Austin ok 20:28
pmichaud I'm not sure that DynLexPads have ever been really "working"
afk for a bit
jrtayloriv Austin, When you said "the library issue" in your message, you are referring to the library of builtins (like 'define', 'set!', etc) that I'm trying to include at startup correct? 20:30
einstein how can i add "include path" for load_bytecode via the console cmd? 20:31
NotFound einstein: load_bytecode uses lib path, not incldue path. 20:32
Austin jrtayloriv: exactly.
jrtayloriv ok, just wanted to make sure you were talking about something else I had missed :) 20:34
fperrad ping bacek
purl I can't find bacek in the DNS.
bacek fperrad: pong
einstein hmm oops i gave it a faulty path :) 20:35
fperrad bacek, t/lua-TestMore is a git submodule 20:36
take a look on doc/running.pod, there are some instructions
bacek fperrad: thanks, looking
fperrad: (tt#1067) it's failing in t/pmc/userdata for you, isn't it? 20:38
fperrad bacek, failures on thread.t, thread_hll.t & userdata.t 20:40
20:40 ash_ joined
bacek fperrad: hmm... PMC_metadata contains garbage... 20:41
kyle_l5l bacek, freeze/thaw of PMCs with metadata gives me trouble 20:46
darbelo The code handling freeze/thaw of PMCs needs the application of fire. 20:48
bacek passing flamethrower to darbelo 20:50
darbelo kyle_l5l: what kind of trouble? 20:52
bacek fperrad: LuaFunction.init looks suspicious. attrs->env stays uninitialised. 20:53
fperrad: but id doesn't matter. Still failing... 20:54
kyle_l5l trouble when cloning libraries and objects. I think that if a PMC has metadata, and something in the metadata has metadata, that causes trouble.
darbelo kyle_l5l: do you have an easily-reproducable example? 20:56
20:57 bluescreen joined
kyle_l5l darbelo, it's reproducible, but it's pretty messy. I'll see what I can do. 20:58
NotFound In LuaFunction getfenv method if (!retval) must not be if (PMC_IS_NULL(retval)) ?
20:59 theory joined
darbelo Nopaste it. I can deal with messy. 21:00
I think.
kyle_l5l darbelo, sure. but first I must deal with more pressing issues: pizza.
ash_ would it be working considering writing a pmcc2cc (C++ version of PMC) ever? just curious 21:04
bacek ash_: probably no. 21:06
21:07 joeri left
ash_ don't think it would be useful? not even if you ever wanted to link to a C++ library? 21:07
NotFound ash_: What problem are you trying to solve? 21:09
ash_ thinking about linking to LLVM, there are C bindings but they are not at all as easy to use or understand
darbelo ash_: You want c++ in the VTABLE and METHOD bodies. Right? 21:11
NotFound VTABLE functions are static, they haven't decoration or linkage problems. 21:12
darbelo NotFound: He's working with a pure c++ api. no extern "C". 21:13
He wants to call c++ from inside the vtable. 21:14
NotFound Bad luck, then.
I think the policy since long time had been to not require C++ 21:15
darbelo I would write the code that should go into the VTABLE in a separate file, compile it with c++ (but with extern "C") and make the VTABLE call that code. 21:16
jrtayloriv Anyone here who knows scheme and Perl 6? These two are pretty much equivalent, right? pastebin.com/d6af651c9 21:17
chromatic I'd change "eval '$a'" to "return $a", but they should otherwise be equivalent. 21:19
21:19 bluescreen joined
jrtayloriv ok, thanks. 21:19
21:20 hercynium joined
bacek NotFound: src/gc/api.c lines 260-261 looks suspicious. How we can have metadata without is_special set? 21:23
NotFound Don't know, but if there are no metadata, it doesn't harm. 21:27
chromatic Hm, mark_special() marks the metadata unilaterally, even if there's no (defined) pointer there.
src/gc/mark_sweep.c:420
pt_shared_fixup() in src/thread.c can add metadata without setting the special flag. 21:28
make_prop_hash() in the default PMC doesn't set the special flag.
darbelo src/gc/mark_sweep.c:420 was me. That used to be done in a separate pass over the next_for_GC linked list. 21:29
chromatic the Undef PMC's set_pmc() undoubtedly has a bug where it doesn't propagate PMC metadata (line 51). 21:30
NotFound Parrot_gc_mark_PMC_alive already check for nullness 21:31
21:34 Whiteknight joined
darbelo There's no need for the check in src/gc/api.c then 21:35
chromatic Right.
We can simplify that control flow.
pmichaud is it just me, or has parrot's backtrace become suddenly much less useful? 21:37
Whiteknight pmichaud: why do you say that?
purl Because ey thinks it is true. Duh.
pmichaud the backtrace lines seem to always identify the line number of the start of the sub, instead of the line that was being executed 21:38
nopaste coming
nopaste "pmichaud" at 72.181.176.220 pasted "backtrace example" (13 lines) at nopaste.snit.ch/18101
chromatic I'm sure there are bugs. IMCC's line counting has a really bad heuristic.
pmichaud it's not just bugs... it's radically different
NotFound Is fast the default runcore now? 21:39
pmichaud for example, in the nopaste, the first "called from" line says Compiler-Regex.pir:241
chromatic Yes.
pmichaud that's the beginning of the "quant" sub itself
previously Parrot would give me the line number _within_ the quant sub
same for the other line numbers given in the backtrace 21:40
they all point to the beginning of the sub, instead of the line that sub was executing at the time the exception is raised
NotFound I think is the runcore
pmichaud ah, so fast core only tells me the line number of the sub?
chromatic I don't think it's the runcore; the context's pc should always be accurate. 21:41
pmichaud let me see if I can come up with a short example 21:42
chromatic The only difference between the fast and slow runcores is that the slow runcore updates the interp's pc for every op dispatch.
pmichaud hmmm 21:44
I'm not having the issue for a short example
I'll try another 21:45
NotFound It may be only when entering new runloops from vtable calls.
pmichaud I'm not doing anything funny like that; this is pretty straightforward PIR 21:46
weird... I don't seem to be able to duplicate it with a simple example. I'll keep an eye out for it, though. 21:47
chromatic I plan to fix line number reporting soon anyway. 21:48
NotFound pmichaud: ops use vtable calls, and these enter new runloops for vtable overrides 21:49
Whiteknight how does parrot store line numbers anyway? 21:50
darbelo badly :)
chromatic Ugly.
Whiteknight do tell
pmichaud NotFound: I know that. I'm not using any ops that rely on vtable calls. 21:51
chromatic src/sub.c lines 242 - 266
(we can fix that entire counting mess if we have a counter for "number of ops executed in the current segment")
NotFound chromatic: that part is not speed critical 21:56
Whiteknight oi! that's horrible
chromatic I wouldn't make that argument on speed alone. 21:57
NotFound Whiteknight: is better that try to guess backwards where the previous opcode starts.
Whiteknight NotFound: it might be better then that, but it's still not good
NotFound Whiteknight: I know, I try to find a better way months ago but failed. 21:58
chromatic We could use a binary heap instead of an array. 22:00
Or even a binary tree implemented as an array.
We don't modify segments after generation. They're effectively read-only. 22:01
Store the annotation information by offset from pc to base.data.
It uses as much storage as currently, but we don't do weird math and we don't have O(n) lookup. 22:02
What's that, O(log n)?
nopaste "darbelo" at 200.49.154.173 pasted "unconditionally call Parrot_gc_mark_PMC_alive on metadata. All tests pass." (67 lines) at nopaste.snit.ch/18103 22:04
NotFound I don't see a need to speedup that thing. Is used only for reporting errors, isn't it?
chromatic It's not about speed, though. (But it does slow down the profiling core a bit.)
NotFound Uh, forgot the profiling core 22:05
chromatic For me it's all about the "Wow, that's ugly code -- surely there's a better way!" 22:06
Whiteknight I definitely don't like how it shadows pc 22:17
22:19 cotto_work joined 22:22 bacek joined
chromatic Sometimes I wonder what happened when my super technical musings completely derail all conversation in the channel. 22:27
darbelo bacek: nopaste.snit.ch/18103 22:32
less suspicious or more? 22:34
chromatic A little less, but....
... I'd rather flip the "is special" flag whenever we attach metadata.
Most PMCs don't have metadata.
We always have to check that flag, which is fine.
Only in that codepath should we check if there's metadata and then mark it.
darbelo So, a non-special PMC should never have metadata? 22:35
chromatic Any PMC with metadata should have the special flag set.
22:36 ruoso joined
darbelo The calls outside of mark_special were workarounds for mismanagement of the special flag, right? 22:37
chromatic I don't know. 22:38
I'm still not sure about the intent of the special flag.
Whiteknight it was used to determine the existance of PMC_EXT for a while 22:39
and maybe _sync too
NotFound Will not be easier to drop the special flag, and check all relevant flags in the is_special_PMC_TEST macro? 22:42
Whiteknight I think it will be, yes 22:43
chromatic Works for me.
bacek msg fperrad I don't quite understand curr_func in LuaUserdata.pmc. Looks like it trying to fetch "env" field from ordinary Parrot's Sub.
purl Message for fperrad stored.
chromatic We could add a has_metadata flag too. 22:44
NotFound Will be significant difference between checking the flag and checking the pointer? 22:45
bacek msg fperrad Indeed. It returns some garbage and GC epicly failing during mark. 22:47
purl Message for fperrad stored.
nopaste "darbelo" at 200.49.154.173 pasted "Amusingly, this passes all tests too." (22 lines) at nopaste.snit.ch/18104 22:48
22:51 cconstantine joined
bacek ok, $dayjob time 22:51
see you soon
22:53 patspam joined
NotFound darbelo: that probably only means that in the test suite the metadata is always in special PMC. So we need more tests. 23:00
darbelo NotFound: yeah, I would expect partcl to blow up horribly with that patch. But it shows how much the test-passiness of my previous attempt was worth. 23:03
23:04 rhr joined 23:07 payload joined 23:08 payload1 joined
darbelo NotFound: The question here is, if there's metadata on that PMC why isn't it special? 23:12
chromatic I looked at all instances of attaching metadata and none of them toggle the special flag. 23:13
NotFound My question is: did we really need that lot of macros for PObj flags? Looks like there is a lot of redundancy. 23:14
darbelo Hmm. We could detach the 'bears metadata' meaning from the special flag and fo "Look! Metadata!" and mark it. 23:15
We're halfway there already. It'd be a pretty minimal change. 23:17
NotFound The current code looks like: if is special, look for metadata. If not, look for metadata.
chromatic Yeah, the current code is wrong in a lot of ways.
cconstantine Tene: can you describe what 'library', 'export', 'import' and 'hll' do in stene? Or maybe throw me at some docs (I couldn't find any)?
darbelo NotFound: Exactly, we could just say 'look for metadata' and leave the special flag out of it. 23:18
NotFound chromatic: maybe we must centralize all that instances in a function. 23:19
chromatic Remember though that I'd like to make the marking code as straightforward and fast as possible. 23:20
That means checking as few flags as possible and branching as few times as possible.
NotFound chromatic: getting the flags one time and switchIf on it? 23:21
switch or if, I mean
chromatic Seems sensible to me. 23:22
darbelo was wrong. partcl passed make test with the last nopaste.
Whiteknight checking a has_metadata patch doesn't save us anything over just checking whether ->metadata is PMCNULL 23:29
or NULL
s/has_metadata patch/has_metadata flag/
cotto_work did someone break the build on src/byteorder.c? 23:32
darbelo not that I've noticed. 23:33
cotto_work It's perfectly possible that it's just my system that's goofy. The win32 build process isn't quite as fun as on linux.
NotFound $Id: byteorder.c 37848 23:34
cotto_work so it's been a while
darbelo msvc or strawberry perl?
cotto_work strawberry 23:35
purl strawberry is a Perl for Windows that works just like Perl everywhere else. See win32.perl.org/wiki/index.php?title...berry_Perl
cotto_work strawberry is also well how about RAWBERRY
purl okay, cotto_work.
cotto_work although it also broke when I tried using Strawberry + the msvc setup script from the wiki 23:36
Whiteknight any git heads around?
darbelo I'm guessing you used -j on make.
cotto_work nope 23:37
well, yes then no after -j failed
darbelo Then I have no idea why you are borked. 23:38
Have you tried rebooting? 23:43
;)
dalek rrot: r41543 | darbelo++ | trunk/src/gc/gc_private.h:
Remove duplicated ASSERT_ARGS definitions.
 23:44
ttbot Parrot trunk/ r41543 MSWin32-x86-multi-thread make error tt.ro.vutbr.cz/file/cmdout/108271.txt ( tt.ro.vutbr.cz//buildstatus/pr-Parrot/rp-trunk/ ) 23:54
Whiteknight FAIL
darbelo Was that me? 23:55
WTF?
Whiteknight well, it wasn't not you 23:56
darbelo Acording to the link it's been broken since 41515. 23:57
Same error. So, I'm blaming bacek.
Whiteknight bacek is TEH FAILZ 23:58

Powered by Cro and the Raku® Programming Language.

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

Thank you!