Parrot 2.1.1 Released! | parrot.org/ | Tasks: PCC deprecations branch, HLL subclassing and MMD branch
Set by moderator on 19 February 2010.
Whiteknight Austin: you going to be upset if I use spaces instead of tabs? 00:02
Austin Not at all.
I program in a proportional font, so I expect the tabs are all over the place for you. 00:03
Whiteknight w00t, because those are my settings and I'll be damned if I'm changing them :)
Austin So long as you keep the 1 tab = 12 spaces equivalence....
:)
Whiteknight I was going to use 1 tab = (7.435 * line number) spaces, but an even number is slighly better 00:04
Austin: how to test it? 00:11
parrot-nqp setup.nqp test failed
Austin What, tab size?
Oh.
What fails? 00:12
purl see "doesn't work"
Austin (Because it works for me.)
Whiteknight Cannot find harness-nqp script. Did you set HARNESS_ROOT_DIR? at t/harness line 36.
Austin Hmm... chmod +x t/harness-nqp.sh ? 00:13
Whiteknight ah, that did it
Austin++
Austin Is there a way to set that in git, like svn:executable? 00:14
Whiteknight t/bootstrap/UnitTest/Testcase.nqp ..... 1/? Could not find non-existent sub assert_that
Austin: I think there is, I feel like I've done it at some point
Austin Ruh roh. 00:15
purl ruh roh is Raggy?
Austin That's a real error. (I remember - I fixed it today or y-day.)
Edit that file, remove the test.
Whiteknight Austin: I changed that file to +x and commited the change 00:16
I also just commited a test for nested super()
Austin Woot!
Does it pass?
Whiteknight++
Whiteknight test does indeed pass 00:17
I'm sorry I doubted
Austin ++
:)
I'm about waist deep in the mocking stuff right now. I think I've got a model that works... 00:19
It's just a question of building the horribly tangled thing...
dalek kapo: ed480bb | whiteknight++ | :
make harness-nqp.sh executable
kapo: 31e3b92 | whiteknight++ | t/Syntax.nqp:
test that super() nests to multiple levels correctly
 00:20
Austin sings, "The judge said 'Son, what is your alibi? If you were somewhere else then you won't have to die.'" 00:38
Whiteknight You use a proportional font? 00:43
Austin Yeah
Whiteknight cringes
Austin :)
Everybody talks about how code is read more than written... 00:44
Well, what if there were fonts that were optimized for reading?
Oh,wait! _There already are!_
Whiteknight I don't know, but it's definitely not how my mind works
Austin Wait 'til you get older... 00:45
The eyes start to go... suddenly, it matters.
dukeleto 'ello 00:46
Austin Hey, duke 00:47
00:47 patspam joined
dukeleto Austin: how goes it? did you see my msg about the diehard stats tests? 00:48
Austin Yeah, I'd seen those when I was googling around before, looking for how to test random. For a guy with a stats background, I'm ashamed to admit that my eyes glazed right over... 00:49
dukeleto Austin: i may be interested in implementing some of them. i also said that i would work on fixing the randomness of parrot RNG's and haven't done that yet either... 00:53
Austin: feel free to bug me about stats stuff. i had to take way too many stats classes for my math degrees 00:54
Austin More power to you, then. I know a lot of people have put a lot of thought into how to do good RNG, so it seems like it ought to be "free" from the system(s).
Yeah, me too. 00:55
Stats and economics.
Two great tastes that go great together. Like Lutefisk and ... some other horrible thing
dukeleto Austin: if parrot gets GSL binding (another thing I have promised and not delivered on) then we get like 60 different RNG algorithms 00:56
Austin: hah!
00:56 abqar joined
Austin What's a GSL? 00:56
purl a GSL is a pure C library. I want to access it's functions from Parrot, perhaps as PMCs, perhaps not. what is the best plan of action to start?
dukeleto Austin: search.cpan.org/~leto/Math-GSL-0.20...GSL/RNG.pm
Austin Apparently not the first time that has been asked.
dukeleto Austin: GNU Scientific Library
Austin: i wrote the Perl 5 bindings
Austin Ahh, okay. 00:57
purl, forget a GSL
purl Austin: I forgot gsl
Austin purl, GSL is the GNU Scientific Library. See www.gnu.org/software/gsl/ 00:58
purl OK, Austin.
01:12 ascent_ joined 01:22 kid51 joined 01:24 bacek joined 02:24 bacek joined
Whiteknight Austin: src/Exceptions.nqp is depressing 02:25
bacek hi again. 02:28
purl oh, you're back!
kid51 Uh-oh: Am once again getting this smolder submission error: 02:44
Could not upload report to Smolder at smolder.plusthree.com
HTTP CODE: 500 (read timeout)
Austin Heh. 03:04
Tene Where is this depression Exceptions.nqp ? 03:09
Austin: you parroting tonight? 03:10
Austin Tene: Yeah, I'm up.
He's talking about Kakapo's Exceptions.nqp, which defines a bunch of exception classes. 03:11
Tene Ew, yes, Exceptions.nqp is unfortunate.
I was thinking of working on type-based EHs now. Lemme throw together an example for you to look at.
Hmm. The resume continuation isn't getting set in subclasses of Exception. That's strange. 03:22
nopaste "tene" at 76.27.121.193 pasted "Austin: similar to the existing types only/except list API" (63 lines) at nopaste.snit.ch/19819 03:26
Austin Tene, does that run? 03:29
Tene No, it doesn't.
Austin ok
Tene oh, crap, I said "handle_types".
Austin yeah, I wondered at that 03:30
Tene That's the method used right now to set filters based on the integer type id.
should be handle_classes oslt
I'm wondering if you have comments or strong opinions on the "handle_these" / "handle_except_these" API before I implemented it.
Austin As I see it, there's a couple of different use cases. 03:31
First, you can try to catch a particular exception. 03:32
That's like catching file-not-found or something.
Next, you can try to catch a whole family of exceptions: any unit test failure, any database error, etc.
Tene any IO error, which will presumably include file-not-found 03:33
Austin The sort-of converse to that would be Java-esque: catch anything except the horribly fatal ones. (Runtime Errors)
Tene Is there ever a use case for "ignore IO errors, but catch anything else"?
Austin They solve that problem by creating two separate class hierarchies implementing a common interface. 03:34
There's the control-exception scenario: catch any of these.
And the kind of obvious "anything but a control exception"
Which is what nqp currently does.
try { foo() CONTROL { say("control exception"); } CATCH { say("anything else"); } }; 03:35
03:35 janus joined
Austin Most of those scenarios reduce down to any subclass of X, or not-any subclass of X. 03:36
Tene Right, that was the original motivation for .handle_types_except
Austin The control-exceptions could be trivially recoded to catch the one type or its children
Tene I guess with a type hierarchy, there's less need for "catch any exception of a type in this list".
Austin The place where I see "any on this list" being used is when the coder rattles off a set of exceptions to catch. It may be harder to see in *perl, since the catch blocks look like if statements. 03:37
But other languages make you declare your types as part of the catch: catch (FileNotFoundException e) {...} catch (QuotaException q) {...} ... 03:38
So I can see them maybe being used, but I don't think it's the end of the world if the performance isn't great.
Anyway, I think that anything-but-these becomes less useful because at this moment we can just set up a class tree the right way. 03:40
Tene well, we definitely do need "anything-but-control", although I guess we could have a "not-control-exception" superclass for everything else.
Austin Right 03:41
I'd like to know how .net handles this stuff.
okay, they don't distinguish. 03:44
(and coders suffer from catching too many things)
Anyway, I guess the upshot of this is that the handle_foo approach is just fine, and will encompass every case I can think of, so long as it honors subclasses. 03:45
Tene Great, okay. I'll start implementing that as soon as I track down this issue with resume continuations in subclasses. :( 03:46
Austin (and handle_not_foo)
:)
Tene ex['message'] gets the message just fine. 03:47
ex['resume'] gives Null PMC
Austin heh
03:47 Andy joined
Austin One thing that seems important to me is resumable versus not. 03:48
Tene Yes, that's definitely important, but I'm not entirely certain how it should be implemented. 03:49
Something about roles should be considered.
seems like.
Austin Well, as I understand it, a resumable exception comes with a resume continuation.
Tene Right.
Austin Now, think about "finally" 03:50
If I write some code like: try { do_something() } ... finally { close( handle ); }
I think I need to catch every exception as it goes by, in order to make sure I send control over to the finally block. 03:51
But a resumable exception shouldn't get finalized.
Tene but if the handler chooses *not* to resume it, it should. 03:53
Austin yeah
If some "vice president of exceptions" decides to resume it, then you should not have finalized. But otherwise you should have. 03:54
:-$
This is why the handler is supposed to execute in the dynamic context of the thrower. 03:55
The part I don't understand is where does the change occur from context-of-the-thrower to back-10-levels-up-the-stack? 03:57
Tene okay, the issue is accessing it through ex['message'] 04:01
if I getattribute instead of using the keyed interface, resuming works fine with subclasses.
Austin heh. What's wrong with that picture?
Tene oh, setting through the keyed interface doesn't work either. 04:02
vtable override stuff?
I don't really understand that bit of parrot.
Austin Okay. So apparently marking the exception as "handled" is what turns off the resum-ability. 04:03
Tene Not right now it doesn't, although I could imagine an interface that does.
the default handlers generated by PCT check for the 'handled' attribute to decide whether to rethrow. 04:04
If it's handled, wouldn't that mean that you *do* want to resume it? 04:06
Austin Heh.
No.
If you want to resume it, you invoke the resume continuation. 04:07
Tene Well what we *want*, then, is that each handler either rethrows, resumes, or marks the exception handled. 04:08
Austin Right.
Tene marking it handled would invoke any 'final' blocks.
Austin rethrow is easy. and resume ought to be easy.
Tene Parrot noticing the lack of any additional handlers would invoke any 'final' blocks.
Austin (provided you get that problem sorted out)
Tene The problem is whether we can either: enforce that all handlers do one of these three, or detect when a handler doesn't do one of those. 04:09
Austin The p6 approach to finally is for the language to maintain a queue of them in some other data structure, and the handler unspools the queue when it decides to handle.
I can live with that, I guess.
Tene I could just deprecate the keyed interface to resume continuations, and make you getattribute instead, and give it a method that invokes the resume cont. 04:10
ex.'resume'()
Austin While I'm sure there's something wrong with deprecating the keyed interface, I'd love a method. 04:11
But I can write it, so no strain
Detecting the failure-to-act:
that has to be some kind of opcode condition. 04:12
Like inserting a dummy stack frame? 04:13
If you try to return from the handler without acting, ...
That's a little wiggy.
dalek rrot: r44552 | bacek++ | branches/ops_pct/ext/nqp-rx (4 files):
Add to-be-nqp-settings
 04:14
rrot: r44553 | bacek++ | branches/ops_pct/compilers/opsc (5 files):
Use nqp-settings. Cleanup builtins little bit
rrot: r44554 | bacek++ | branches/ops_pct/src/ops/core.ops:
Remove redundant 'goto NEXT'
Tene how much utility would there be in 'finally' blocks that run in *most* leave conditions, except not in those where the handler misbehaves?
any?
purl rumour has it any is being done as a particular language feature
Austin Little.
Depends on how you define 'misbehaves'?
But I think finally blocks can be left to the languineers. 04:16
Tene returns or goes elsewhere without doing one of those three.
Austin "goes elsewhere" makes me nervous. Does that mean sub-call, or more than 100 opcodes from the target? 04:18
Tene Oh! I know! Just *always* do the finally blocks, and then if the handler resumes, run the 'finally' block backwards.
eh, probably just 'returns'. I'll have to think about it more. 04:19
Austin Don't sink a lot of energy - there's a solution already. 04:20
Tene Explain?
Austin hll's can keep a finally queue themselves. 04:23
A handler that wants to not-resume can just suck on the queue until it gets to its own level or higher on the stack.
Tene Well, yes, if we assume "The HLL doesn't misbehave", that's fine. It would be nice if there was a way that we could offer something on the parrot level that would handle it, but it's outside the scope of what we're working on right now. 04:24
Austin while @queue { if @queue.peek.is_below( $current_context ) { @queue.pop.finally(); } }
Yeah.
Plus, we gotta trust somebody. 04:25
Tene Right.
Austin Might as well be the hll.
Tene Okay, I'm comfortable with that.
I'm *un*comfortable with this keyed interface misbehaving in subclasses, though.
Austin Can't trust the vm... 04:26
Tene I'll try to remember to bring it up during the day when others are around. 04:27
Austin What is the problem?
purl the problem is that WikiDoc doesn't use Pod formatting codes, and I don't like to have two sets of formatting codes.
Austin It doesn't get set when throwing a subclass?
Tene It does get set, but fetching it through get_pmc_keyed_str doesn't work, because it doesn't make it to the VTABLE in Exception.pmc 04:28
fetching it with getattribute works fine.
ex['resume'] # fails on subclasses
getattribute ex, 'resume' # works fine
Austin Ah. Can I see your code?
You might have a C3 problem.
"The first parent is the main parent" 04:29
nopaste "tene" at 76.27.121.193 pasted "Sure, here you go." (47 lines) at nopaste.snit.ch/19821
Austin Or it might be a difference between object.pmc and exception.pmc
Tene if you swap the # from line 29 to 28, it dies with Null PMC access in invoke() 04:31
Austin you've modified the get_..._keyed to support resume? 04:32
get_pmc_keyed, I guess, since resume isn't a string 04:33
Tene get_pmc_keyed just ends up fetching the named attribute.
and it works fine on Exceptions. 04:34
Nothing in that path has been modified.
Austin Huh?
okay. I got it 04:35
Tene Um, nevermind, I guess. I added a printf to Exception.pmc, and it *does* call get_pmc_keyed in Exception.pmc when fetching from the subclass. 04:36
Austin But there's no data? 04:37
Tene yeah, it returns a null pmc. tracking it down a bit more...
Austin Heh. Whiteknight's email was talking about overriding the get_pmc_keyed stuff to do method lookup. This stuff proxies through that same code. 04:38
Tene it calls GET_ATTR_resume
nopaste "tene" at 76.27.121.193 pasted "GET_ATTR_resume definition" (10 lines) at nopaste.snit.ch/19822 04:41
Austin You know the data is in there. 04:45
Because you can access it with getattribute.
But if get_attr_str is looking in the wrong place... 04:46
Tene op getattribute is just: $1 = VTABLE_get_attr_str(interp, $2, $3) 04:47
Austin But which vtable? 04:48
purl which vtable are we referring to at the moment?
Austin I'm wondering if there's some shenanigans going on with pmcproxy
Tene my printf shows that it does end up in Exception.pmc's get_attr_str 04:49
Austin Right
But what set_attr did it go to?
throw calls vtable-set-attr-str directly
And when you call get-attr-str directly, you get the data back
But when you call get-pmc-keyed, which locally-calls get_attr_str, you get bupkis. 04:50
Maybe try calling the vtable version of getattr-str from get_pmc_keyed?
Tene no change. 04:52
Austin bummer
Is this code on a branch somewhere? 04:53
Tene Yes.
exceptions_refactor
oh, if I add a 'resume' method, I'll have to deal with the C/PIR boundary, I think. 04:56
So, that's less than ideal. 04:57
Well, I'll just stick with using getattribute for now, and try to ignore it. 04:58
05:09 davidfetter joined
dukeleto sweet. parrot compiles with the latest clang 05:13
the test suite passes too! 05:19
dukeleto does a little dance
dalek rrot: r44555 | tene++ | branches/exceptions_refactor/src (2 files):
Allow rethrown subclassed exceptions to find the next handler.
 05:20
Austin Suppertime! 05:25
dalek rrot: r44556 | bacek++ | branches/ops_pct/compilers/opsc (13 files):
Brake everything. Restructure files and directories. Strip out old stuff
 05:53
rrot: r44557 | bacek++ | branches/ops_pct/compilers/opsc/opsc.pir:
Fix opsc.pir
rrot: r44558 | bacek++ | branches/ops_pct/compilers/opsc/Rules.mak:
Remove removed opsc_core from Rules
rrot: r44559 | bacek++ | branches/ops_pct/compilers/opsc/src/Ops/Op.pm:
Borrow Ops::Op from ops2c
 06:09
rrot: r44560 | bacek++ | branches/ops_pct/compilers/opsc (2 files):
Add build of Ops::Op
rrot: r44561 | bacek++ | branches/ops_pct/compilers/opsc/t/04-op.t:
Add tests for Ops::Op
06:33 eternaleye joined 06:52 uniejo joined 07:13 eternaleye joined 07:56 fperrad joined 08:10 iblechbot joined 08:36 bacek joined, riffraff joined 08:53 mikehh joined 08:57 payload joined 09:02 slavorg joined 09:08 mikehh joined
dalek rrot: r44562 | mikehh++ | trunk/runtime/parrot/library/Test/Builder/Test.pir:
fix codetest failure - trailing spaces
 09:26
09:34 mikehh joined 09:36 AndyA joined 09:40 bacek joined
dalek rrot: r44563 | bacek++ | branches/ops_pct/compilers/opsc (3 files):
Made Ops::Op inherited from PAST::Block and construct them during compilation phase
 09:42
09:43 eternaleye joined 09:53 JimmyZ joined 09:54 JimmyZ joined
mikehh All tests PASS (pre/post-config, make corevm/make coretest, smoke (#32446), fulltest) at r44562 - Ubuntu 9.10 i386 (g++ with --optimize) 09:55
dalek rrot: r44564 | bacek++ | branches/ops_pct/compilers/opsc (3 files):
Add stub for Ops::File
 09:58
rrot: r44565 | bacek++ | branches/ops_pct/ext/nqp-rx/src/gen/settings.pm:
Update nqp settings
rrot: r44566 | bacek++ | branches/ops_pct/compilers/opsc (2 files):
Fix parsing almost empty ops
rrot: r44567 | bacek++ | branches/ops_pct/compilers/opsc (2 files):
Implement Ops::File.read_ops
mikehh heading back to amd64 - bbiab 10:07
10:18 mikehh joined 10:33 gaz joined 11:20 kid51 joined 11:23 smash joined 12:05 bkuhn joined 12:12 jimk joined 12:30 tetragon joined 12:37 plobsing joined
dalek rrot: r44568 | bacek++ | branches/ops_pct/compilers/opsc/t/03-past.t:
Rewrite test in NQP
 12:42
rrot: r44569 | bacek++ | branches/ops_pct/compilers/opsc/t/03-past.t:
Fix test
rrot: r44570 | bacek++ | branches/ops_pct/compilers/opsc/src/builtins.pir:
Add lc
rrot: r44571 | bacek++ | branches/ops_pct/compilers/opsc (5 files):
Add normalization of op args as in ops2c
rrot: r44572 | bacek++ | branches/ops_pct/compilers/opsc (2 files):
Stole expand_args from ops2c
rrot: r44573 | bacek++ | branches/ops_pct/compilers/opsc (3 files):
Move expand_args into Actions
rrot: r44574 | bacek++ | branches/ops_pct/compilers/opsc (3 files):
Create op variants during parsing
12:45 ruoso joined 12:49 cognominal joined 12:50 cognominal joined 12:57 whiteknight joined 13:01 payload joined
whiteknight good morrow, kind #parrot 13:01
13:19 iblechbot joined 13:30 FullMetalHarlot joined
whiteknight ENETSPLIT 13:49
13:51 KatrinaTheLamia joined, mikehh joined, bkuhn joined, riffraff joined, dalek joined, NotFound joined, Austin joined, PacoLinux joined, preflex joined, wagle joined, TonyC joined, szabgab joined, pmichaud joined, Coke joined, dngor joined, eirik joined, cotto_work joined, ttbot joined, confound joined, sri joined, bacek_at_work joined, Tene joined, ingy joined, solarion joined, rhr joined, workbench joined, slavorgn joined, cxreg joined, Infinoid joined, jjore joined, elmex joined, baest joined, frodwith joined, tewk joined, treed joined, athomaso1 joined, eiro joined 13:52 mikehh joined 13:58 Andy joined 14:00 payload1 joined 14:01 mikehh joined 14:07 mikehh joined 14:13 davidfetter joined 14:34 cognominal joined 14:36 JimmyZ joined 15:06 ruoso joined 15:07 mikehh joined 15:09 zostay joined 15:12 bubaflub joined 15:14 slavorgn joined
Tene anyone willing to express an opinion on how handle_types, handle_types_except, handle_classes, and handle_classes_except should interact with each other? 15:33
dalek kudo: 9efead6 | (Solomon Foster)++ | src/core/Bool.pm:
Add Bool.perl.
kudo: bbe9441 | (Solomon Foster)++ | t/spectest.data:
Turn on S02-literals/char-by-number.t and S02-names_and_variables/perl.t.
kudo: af0424a | (Solomon Foster)++ | src/core/Signature.pm:
"Bool::True" instead of "1", as per jnthn.
Tene Hey! Those are commit messages, not opinions about exception handlers!
Austin Fail - you can only handle one or the other.
Tene orly? Hmm. That sounds reasonable. 15:35
Austin Since classes is new, types vs. classes can be exclusive. 15:36
15:36 lucian joined
Austin classes and classes-except is the first time that handle_yes and handle_no might be meaningful in the same scenario. 15:36
I'd say look-for-no, and if not found, look-for-yes. 15:37
That lets you do "handle subclasses of A, except for any subclasses of A::X
Tene I'm tempted to just have a sense bit for include/exclude, and if you really want anything more-complicated, use a subclass of ExceptionHandler (NYI) 15:38
Austin That also works.
Tene or, handle A, and rethrow if A::X
So, let's do that.
Austin I was wondering if typed versus class-ed handlers weren't different pmcs
Tene but your inclincation is that typed filters and classed filters wouldn't make sense on the same handler? 15:39
right?
purl ooh, the system. right, right.
Austin No.
Tene purl, forget right.
purl Tene: I forgot right
Tene Is that "No, you're wrong about my inclination." or "No, both on the same handler wouldn't make sense."? 15:40
Tene english fail
Austin I don't see them both on the same handler.
Tene Okay. I should be able to reduce this to a single array and two sense bits, then.
Austin And I kind of see deprecating typed exceptions, too. 15:41
But that's for tomorrow.
15:41 Psyche^ joined
Austin Here's a weird thought: does an exception handler have anything in common with a multisub, in this model? 15:43
We're basically saying "catch these, and send them here, here, or here." 15:44
16:03 whiteknight joined 16:04 theory joined, riffraff joined, Andy joined 16:06 betterworld joined 16:10 lucian joined
Austin sings, "So here we a-are, in the Tijuana jail; ain't got no friends to go our bail!" 16:11
Tene Austin: Quite possibly.
Tene committed, now afk working.
dalek rrot: r44575 | tene++ | branches/exceptions_refactor/src/pmc/exceptionhandler.pmc:
Refactor ExceptionHandler to use one array instead of four.
 16:16
16:18 davidfetter joined
mikehh All tests PASS (pre/post-config, make corevm/make coretest, smoke (#32454), fulltest) at r44574 - Ubuntu 9.10 amd64 (g++ with --optimize) 16:22
davidfetter oh hai 16:24
16:28 whiteknight joined 16:44 patspam joined 16:53 payload1 left
dukeleto 'ello 16:58
17:02 smash joined
Tene Hi! 17:03
17:03 shockwave joined
shockwave Hello. 17:04
To overload the == operator, there is 'is_equal' vtable method. I don't see a vtable method for '!='. Is it not possible to override != ?
Austin There's the 'ne' opcode. 17:09
== maps to eq, or iseq. != maps to ne, or isne 17:10
(ne returns !vtable_is_equal)
(See $_OPS/cmp.ops) 17:11
17:11 kurahaupo joined
shockwave Austin. They way I find out about vtable methods is by looking at src/vtable.tbl. There is a is_equal method there. 17:14
Austin sure
:)
But somebody has to call those methods.
shockwave You mentioned that != maps to ne, or isne. Is there a way to make the connection of != to ne?
Austin Probably no.
The 'ne' opcode is a test-and-branch op, while isne returns a result as an int expression. 17:15
shockwave Sorry, that's not what I mean.
Austin ok
What *do* you mean? :) 17:16
shockwave What I meant is, if I don't see something in vtable.tbl, for mapping to !=, how can I draw a conclusion as to what it would be based on what's in cmp.ops
Austin Well, cmp.ops contains this: $1 = !VTABLE_is_equal(interp, $2, $3); 17:17
Tene if you want to know what an op does, just read src/ops/foo.ops 17:18
Austin So I'd recommend you read the pod first (perldoc cmp.ops, or check the docs/ directory), and then follow up by reading the source.
Just be warned that the docs frequently lie. 17:19
:<
shockwave I'm about to read the doco you guys mentioned, but it's looking like there is no != vtable override. 17:20
I base that assumption on the fact that ne is defined as !VTABLE_is_equal.
One could draw the conclusion that vtable method names are the same as the function names, minus the VTABLE_ part 17:21
Thus, there is no VTABLE_isnt_equal.
Hopefully I'm wrong.
Thanks, Austin, Tene.
Austin You're welcome. You're also right.
(Why do you want there to be a vtable op for != ?
Tene shockwave: you're correct. There isn't a separate vtable for != 17:22
!= is *defined* as "the opposite of ==", so that's how it's implemented.
shockwave Austin, so that I can map both != and == to an override in my language.
Austin Oh. No.
shockwave Operator overload. 17:23
Tene >< ew
No, that's not supported. 17:24
shockwave Some people like it, some people dont. But, in either case, I need the functionality.
It really helps to override operator for things like 3D matrices and vectors.
Tene nods
shockwave Otherwise you find yourself doing things like Vector.plus(Vector.times(Vector)) 17:25
Tene when dealing with matrices and vectors, etc, is it not just "the opposite of =="?
shockwave Tene, yes.
Tene "Yes, it is just the opposite of ==" or "yes, it is not just the opposite of =="?
Apparently I fail at english negation today. 17:26
kurahaupo question would go away if it were VTABLE_equal(pmc1,pmc2,equalresult,neresult)
question would go away if it were VTABLE_equal(target,pmc1,pmc2,equalresult,neresult)
shockwave Tene, somehow, I need to be able to do this in PIR: $P0 != $P1 # Both would be vector objects.
and I need to call a custom made function to deal with that. 17:27
vtable is_equal maps nicely to ==, but to deal with !=, I'm not sure what kind of mess I'm gonna need to create.
tewk $P2 = $P0.'custom_made_function'($P1) 17:28
Tene shockwave: is the answer returned going to be different from: { $I0 = $P0 == $P1 \\n not $I0 }
shockwave Tene. Yes.
Tene shockwave: if you define VTABLE_equal, then both == and != will work
Okay, in that case, I wouldn't say that the operation is !=
Austin I think that != does not have the same definition as not( == ) 17:29
shockwave BUT, the user is free to overload != however they want.
Austin Okay
Tene at least not as parrot thinks of it.
shockwave Unfortunately.
particle smells like a multimethod to me
kurahaupo User can overload either eq or ne but not both 17:30
plobsing shockwave: can't you modify tene's suggestion to use a PMC { $P2 = $P0 == $P1 \\n $P3 = not P2 } ?
Tene plobsing: he wants to support == and != not being related at all.
Austin Yeah, that's a multimethod. Tene is right.
Tene that is, it's possible for both == and != to be true, or both to be false.
kurahaupo Yuck. 17:31
Austin shockwave: Build a :multi sub. Then when a user overrides !=, extend the multisub by adding another arguments-permutation
Tene shockwave: vtables aren't the place to be doing that. use methods, and have the compiler for your language compile operations to function calls.
s/methods/functions/whatever
shockwave Tene. As Parrot is defined, I must do that. It's just that == mapped very nicely to is_equal, and Parrot took care of the details. 17:33
I'll do it with subs.
Thanks
kurahaupo Or silently convert overloading of != into overloading of == with a wrapper that inverts the result.
Tene and one of those details is that the != op is defined as the opposite of the == op. You can still use ==, and just have your default != op use parrot's != 17:34
kurahaupo: that won't work for defining both == and != independently.
"== op" overloading goes in vtable equals. default != function just calls parrot != on the arguments. 17:35
kurahaupo That's intentional. throw error if they try.
Tene kurahaupo: he *want* to support that. that's a requirement of his language.
shockwave For now, I will restrict != to not be overloadable and do the inversion thing.
17:36 payload joined
shockwave So that if one overloads == in the language, it will behave as Parrot (Automatically supply a !=, which does the inversion). 17:36
I can live with that.
Thanks for the sudjestion, guys. 17:37
Tene np
kurahaupo IMHO if == & != can have same value the those are the wrong names for whatever is being expressed. 17:38
shockwave kurahaupo, I agree. 17:39
17:40 payload joined 17:53 szabgabx joined 17:54 PerlJam joined 18:16 cosimo joined 18:23 kurahaupo1 joined
dukeleto if i give --cc=clang to Configure.pl, but don't set --link or --ld, is it using gcc for those? 18:33
in case anybody is wondering, i found the free non-commercial download page for icc: software.intel.com/en-us/articles/n...-download/ 18:39
whiteknight++
18:42 ruoso joined 18:43 gaz joined 19:02 cosimo joined
Coke skips review. 19:07
dukeleto: if you want to know what it's using, look at config_lib.pasm, or use parrot_config
(or the Makefile) 19:08
whiteknight dukeleto: (re: --cc=clang) I think it uses gcc in thosec ases, yes. I always do the complete "--cc=clang --ld=clang --link=clang" incantation 19:12
Tene So, there's a problem I ran into last night. Exception.pmc stores stuff in attributes. using getattribute on subclasses of Exception works perfectly. The standard interface, though, is get_pmc_keyed, and Exception.pmc's get_pmc_keyed just calls SELF.get_attr_str, which calls GET_ATTR_foo, and that just returns pmcnull for subclasses. 19:13
whiteknight of course, I have a huge obnoxious library of aliases and functions that do i for me
Tene: sounds delicious. Why does GET_ATTR_foo return PMCNULL?
Tene No idea. 19:14
whiteknight okay
Coke ISTR there is setup you have to do in the PMC to allow attr's to work in subclasses.
ISTR this broke tcl for after the initial switchover.
(but tcl doesn't subclass exception...)
Tene Coke: getattribute on the subclass instance works fine.
Coke it didn't say getattribute. =-) 19:16
Tene nods.
Coke s/it/I/
I think the ATTR's were special.
I think cotto might have been the guy that worked on that, but I can't recall for sure.
Tene okay, looks like it depends on how it's set. 19:21
cotto_work hi
Tene set with keyed, can only fetch with keyed. set with setattribute, can only fetch with getattribute
19:21 hercynium joined
Austin Heh. 19:21
nopaste "tene" at 76.27.121.193 pasted "Example of weirdness with subclasses" (45 lines) at nopaste.snit.ch/19829
Austin Data is fine, just two separate bits of data.
Tene Yeah.
if nobody has any ideas, I'll take it to the list. 19:23
Austin Hey, cotto_work. Do you know anything about how to get access to ATTRs in a subclass of a pmc?
cotto_work in pir? 19:25
in C it's the same as with the parent class
Austin No, C. This is for Tene.
There seems to be different data being accessed by the get/set_attr_str macros and the VTABLE functions when a C Exception.pmc is subclassed in pir. 19:27
Tene's recent nopaste highlights the difference using get/set_pmc_keyed, which in Exception just calls SELF.get/set_attr_str
cotto_work I have to take off for a bit but I can look at it this afternoon.
Tene cotto_work: look for mail on the list. 19:28
it'll be there shortly.
Austin What is the "PObj_is_object_TEST" ?
So if the subclass is an object, it calls the vtable method. Which looks for the attrib-index in an attributes table? 19:32
Tene #define PObj_is_object_TEST(o) PObj_flag_TEST(is_object, o) 19:34
sent.
I don't remember if this is the list I have issue posting to or not. 19:35
Coke I see no post nor admin request. 19:38
dalek kudo: 64e4d36 | (Solomon Foster)++ | t/spectest.data:
Turn on bit.t, equality.t, increment.t, and inplace.t, all in S03-operators.
 19:52
NotFound Tene: last time I tried, to access the attributes of the PMC in derived classes you need to add_attribute to the class. That doesn't mean doing that works well, though. 19:55
And when it works, someone broke it in some PMC by directing accessing the attributes for speed, or something. 19:56
In short: the full attributes thing doesn't work well with class derived from PMC 19:57
19:59 joeri joined
dalek rrot: r44576 | coke++ | trunk/DEPRECATED.pod:
The only place a .namespace is declared in the repo with an embedded :: is examples/
 20:02
20:03 bubaflub left 20:07 riffraff joined
Tene Hmm. My mail still hasn't arrived at the list. 20:07
20:14 kjeldahl_ joined
cotto_work I see it. 20:23
Tene ah, yes, I see it now too. 20:30
tewk Tene, I saw a similar problem because of PMCProxy 20:36
Basically an object had a pointer to a PMCProxy, but the PMCProxy didn't know it was an object so it used PMC_data(pmc) instead of VTABLE_get_attr_str. 20:37
I don't remember exactly but check it out. 20:38
20:45 ash_ joined 21:18 tjc joined 21:22 snarkyboojum joined 21:36 eternaleye joined
dalek kudo: 5e5969d | (Solomon Foster)++ | t/spectest.data:
Turn on S04-statements/ given.t and if.t.
 21:39
21:41 tjc left 21:42 NotFound joined 21:43 bacek joined
cotto_work hi bacek++ 21:44
bacek good morning cotto :)
Why ++ today? 21:45
cotto_work opsc work over the weekend
Coke bacek-- # too many questions.
bacek Coke, Yay! :)
Coke is harsh.
cotto, you boson.
anticoke don't get too close 21:46
It won't end well.
cotto_work I should have some tuits this week. I'm looking forward to diving into that code again. 21:48
What's the current plan?
purl i guess the current plan is to get schwern and fimmy an SDSL line and put a couple boxen there. They're going to trade me power, space, and hardware needs for bandwidth.
Coke purl, plot?
purl Plot. What a funny word. Plot. Plot plot plot. Plot plot plotplotPLOT. PlotplotplotPLOTPLOT! Plot plot plotplotplotplot.
bacek cotto_work, I'm going to finish end-to-end skeleton for generating .c/.h files. 21:52
Next big thing - ops bodies parsing. 21:53
cotto_work Are you planning on writing a C parser?
21:55 tac-tics joined
bacek cotto, I'll probably borrow Close :) 21:55
cotto, and we need pmichaud++ help for speed up parsing. It takes about 35 seconds to parse all ops... 21:57
cotto_work quite likely 22:00
I'll see if I can find anything with the profiling runcore. 22:01
bacek I suspect a lot of backtracking. But :: isn't implemented in nqp-rx yet. 22:02
cotto_work Let me know if there's anything that'd parallelize well or that you don't feel like working on. 22:05
bacek cotto, ops bodies parsing I think. 22:06
cotto_work via a C parser or via cheating? 22:07
Coke for parsing ops bodies, do you need all of C, or just the special bits we've layered on top?
cotto_work (I vote for cheating since the compiler will catch any C errors)
Coke (or, just the bits thatwe're using)
bacek almost all of C... 22:08
cotto_work yeah. We'd need to parse anything that can appear in a function. I imagine we use a very significant subset.
bacek LTM doesn't work well, so something like [ goto NEXT () | .*? ] doesn't work as expected. 22:09
Close?
purl hmmm... Close is close file (or pipe or socket) handle or bad on sockets, use shutdown() instead or horseshoes or grenades or nuclear weapons or code.google.com/p/close/
cotto_work that presents some difficulties to the "flagrant cheating" approach 22:10
why not do the bare minimum parsing necessary to capture an op body (matched curly braces, string parsing, etc) and do simple search and replace on the op body? 22:14
I'll have to look at that approach when I get home tonight. 22:18
It'd be nice to avoid the complexity of a near-complete C parser if possible.
bacek, to clarify, you plan is to generate the different runcore files except without any changes to the op function bodies? 22:32
s/you/your/
bacek cotto, yes.
cotto_work ok' 22:33
22:51 kthakore joined 23:16 pootek joined 23:17 Whiteknight joined 23:21 snarkyboojum joined 23:27 lucian_ joined
cotto_work clock? 23:31
purl cotto_work: LAX: Mon 3:31pm PST / CHI: Mon 5:31pm CST / NYC: Mon 6:31pm EST / LON: Mon 11:31pm GMT / BER: Tue 12:31am CET / IND: Tue 5:01am IST / TOK: Tue 8:31am JST / SYD: Tue 10:31am EST /
Whiteknight clock? 23:43
purl Whiteknight: LAX: Mon 3:43pm PST / CHI: Mon 5:43pm CST / NYC: Mon 6:43pm EST / LON: Mon 11:43pm GMT / BER: Tue 12:43am CET / IND: Tue 5:13am IST / TOK: Tue 8:43am JST / SYD: Tue 10:43am EST /
cotto_work time marches on 23:44
Whiteknight how are you doing tonight, cotto_work? 23:45
cotto_work brain tired 23:46
want to do fun things
Whiteknight Ah, I know that feeling
Whiteknight resists the urge to complain about his job or his incompetent manager in a publically-logged forum 23:47
cotto_work fortunately my desire to get paid overrules my desire to go home early and hack on Parrot
Whiteknight ditto
ash_ Whiteknight: i don't know if you remember, but i mentioned to you before i am working on a sort of alternative nqp, based on the llvm, i just was going to let you know, i have a bison based parser that can parse most of the test suite for nqp, now i am working on building the AST from the parser 23:52
Whiteknight ash_: yeah, I remember. I mentioned your project in my blog a few days back
23:55 patspam joined
ash_ cool, thanks for mentioning it 23:55
Whiteknight: i am writing about my implementation on my website greaterthaninfinity.com/ if your interested, i am trying to keep that up to date with my work, but i am not much of a blogger, or wordsmith so it might be painful to read at times 23:59

Powered by Cro and the Raku® Programming Language.

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

Thank you!