#parrot Parrot 2.2.0 Released! | parrot.org/ | Channel log: irclog.perlgeek.de/parrot/today | Tasks: Fix compact_pool shenanigans | Fix HLL bugs (TT #389, #1040) | Prioritize Rakudo Needs
Set by moderator on 16 March 2010.
lucian is going to sleep 00:06
davidfetter g'night, lucian
treed night 00:09
00:09 japhb joined
dalek kudo: 3a03034 | jonathan++ | src/Perl6/ (2 files):
Get the $bottle ~~ s[full] = 'empty' syntax working too.
 00:15
kudo: 05086a3 | jonathan++ | docs/ROADMAP:
Add 'basic s///' to the done section of ROADMAP. One priority #1 item down for Rakudo Star.
Whiteknight purl msg Austin in t/Program.nqp, test_global_at_start_fails and test_global_at_exit_fails don't hang anymore if I comment out the try/CATCH blocks. seems like something in there is throwing an exception and gtting into a loop 00:16
purl Message for austin stored.
Whiteknight purl msg Austin I get this exception when I comment out the try/catch blocks: Running test: test_global_at_start_fails not ok 3 - test_global_at_start_fails # A previously-registered Program instance has unprocessed marshalling queues 00:18
purl Message for austin stored.
nopaste "Whiteknight" at 68.46.29.192 pasted "example t/Program.nqp test failures for Austin++" (24 lines) at nopaste.snit.ch/19969 00:21
Whiteknight purl msg Austin nopaste.snit.ch/19969
purl Message for austin stored.
00:22 chromatic joined 00:23 theory joined 00:38 snarkyboojum joined 01:02 hiroyuki_y joined
Whiteknight I can't do any more debugging tonight. Lost. 01:02
dalek kudo: c053d61 | jonathan++ | src/pmc/p6invocation.pmc:
Fix .can.Bool.
 01:12
kudo: 1120182 | jonathan++ | src/ (2 files):
Re-enable infix:<Z> and infix:<X> with a tweak so Z- style meta-ops still parse correctly.
cotto HELO INTERNETS 01:26
dalek kudo: e3937f7 | (Solomon Foster)++ | build/PARROT_REVISION:
Bump Parrot to 2.2.0.
 01:52
kudo: 72046f9 | (Solomon Foster)++ | src/core/Any-list.pm:
Don't assume that a Code block will know how to .count.
dukeleto cover? 01:59
purl cover is not at link?
01:59 brooksbp joined 02:01 allison joined
dukeleto purl, cover is tapir2.ro.vutbr.cz/cover/cover-results/ 02:02
purl ...but cover is not at link?...
dukeleto purl, forget cover
purl dukeleto: I forgot cover
dukeleto purl, cover is tapir2.ro.vutbr.cz/cover/cover-results/
purl OK, dukeleto.
02:05 hudnix joined
Austin Whiteknight: The funky try/catch blocks are intended to drain the component marshalling queues, which is why you occasionally get those exceptions when you comment out the blocks. 02:13
cotto coverage? 02:18
purl coverage is cv.perl6.cz
02:18 snarkyboojum joined 02:38 JimmyZ joined
dalek rrot: r44977 | cotto++ | branches/ops_pct/compilers/opsc (6 files):
[opsc] revert a batch of pod fixes that don't apply to nqp, fix the opsfile test
 02:51
Austin chromatic: ping 03:03
bacek_at_work cotto, ping 03:19
03:22 hiroyu___ joined
dalek rrot: r44978 | coke++ | trunk/DEPRECATED.pod:
Add reference to TT #1427
 03:24
03:27 _2x2l joined
sorear hmm, was the 2.1.1 release announcement deleted? 03:30
Coke no, just not sticky, apparently. fixing. 03:31
there. 03:32
sorear I see it again
sorear upgrades 03:35
purl upgrades are good!
03:54 janus joined
Coke guybrush? 04:22
guybrush is <reply>I'm guybrush threepwood, mighty pirate. (TM) 04:23
pmichaud use.perl.org/~pmichaud/journal/40248 04:43
(for those who are wondering about my absence from Perl 6 and Parrot)
05:47 chromatic joined
chromatic Austin, pong. 05:53
cotto bacek_at_work, pong 06:20
. o O (I need to work on those ping times) 06:31
clock? 06:36
purl cotto: LAX: Tue 11:36pm PDT / CHI: Wed 1:36am CDT / NYC: Wed 2:36am EDT / LON: Wed 6:36am GMT / BER: Wed 7:36am CET / IND: Wed 12:06pm IST / TOK: Wed 3:36pm JST / SYD: Wed 5:36pm EST /
sorear Where is the list of functions that are kosher to call from a dynpmc? 06:59
Parrot functions, I mean 07:00
cotto PARROT_API is the marker we use (or should be using) iirc
sorear ah, I was under the impression that was for embedding and there was a separate list for extenders 07:01
cotto It could be. My track record for getting things right is pretty bad this week. 07:02
chromatic There are separate headers named embed.h and extend.h, but that's a historical artifact we've never yet fixed. 07:04
cotto did particle ever figure out what broke rakudo on win32? 07:07
cotto needs to get to bed. 07:09
chromatic I didn't hear anything other than the memory usage. 07:11
07:15 fperrad joined
dalek rrot: r44979 | cotto++ | branches/ops_pct/compilers/opsc/src/Ops/Compiler/Grammar.pm:
[opsc] tighten grammar to avoid capturing whitespace surrounding macros
 07:29
rrot: r44980 | cotto++ | branches/ops_pct/compilers/opsc (2 files):
[opsc] pass oplib to Ops::File new_str, fix emitter test
07:57 jan joined 08:07 bacek joined
bacek aloha 08:07
seen chromatic
purl chromatic was last seen on #parrot 56 minutes and 4 seconds ago, saying: I didn't hear anything other than the memory usage.
sorear hello
purl hello, sorear.
bacek chromatic, 2% slow down of pcc branch vs trunk related to "poor man polymorphism" in fill_params. We did a quite good job of optimising after previous round of PCC refactoring... 08:09
msg chromatic, 2% slow down of pcc branch vs trunk related to "poor man polymorphism" in fill_params. We did a quite good job of optimising after previous round of PCC refactoring... 08:10
purl Sorry, I've never seen chromatic, before.
bacek msg chromatic 2% slow down of pcc branch vs trunk related to "poor man polymorphism" in fill_params. We did a quite good job of optimising after previous round of PCC refactoring... 08:12
purl Message for chromatic stored.
chromatic Getting rid of the linked list in CallContext should take care of the rest. 08:14
bacek yeah. It will help a lot. 08:26
But it's task for different branch
. o O ( Is it possible to have normal branching in svn? )
chromatic You have git and I have git; there's no reason we have to do this in svn. 08:28
bacek we are still using svn as "exchange point" 08:29
let me try to branch of branch
chromatic Should work. 08:30
It worked in CVS, in so far as you could do it in CVS. 08:31
bacek bacek@icering:~/src/parrot$ git svn branch pcc_megrecells -m "Sub-branch to replace CallContext's cells linked list with array" 08:32
Copying svn.parrot.org/parrot/branches/pcc...hon_6Mar10 at r44974 to svn.parrot.org/parrot/branches/pcc...recells...
Looks like it will work
chromatic What's your plan? Or do you want my plan? 08:33
dalek rrot: r44981 | bacek++ | branches/pcc_megrecells:
Sub-branch to replace CallContext's cells linked list with array
 08:34
bacek Your first
08:35 iblechbot joined
chromatic I think we need a pointer to the cells and a pointer to the first free cell. 08:35
If we don't have a free cell, we double in size and copy. 08:36
bacek Erm...
chromatic Yeah, there's a flaw in that. How do we know when we've used all of the cells? 08:37
Okay, we track the number of positionals already. That's the number of used cells.
bacek Why don't have Cell*, cell_size, cell_treshold and resize it (same as R*A)? 08:38
chromatic We also track the number of allocated cells.
Yeah, same strategy.
bacek fill_params just use VTABLE_push_foo
everything else is encapsulated in CallContext 08:39
chromatic Works for me.
bacek We have freedom to use any strategy we want to
Deal :)
chromatic First, I sleep. 08:44
bacek It's a good habbit 08:46
08:51 uniejo joined 08:59 cosimo joined 09:02 AndyA joined
sorear the documentaion for the destroy() vtable function states that the called function must not make new references to the doomed value 09:05
what hook should I use, if I cannot control the results?
e.g. Perl5 DESTROY semantics, runs arbitrary code 09:06
09:24 bacek_ joined
bacek_ sorear, in the nutshell - you should't put pointer to "doomed" pmc into some other live object. But you can run any other arbitrary code 09:24
sorear, for example check FileHandle.destroy 09:25
sorear yes
that much is clear to me
but I want to run code which I have no control over
hmm 09:27
on second thought
does "arbitrary" code extend to "calling back into the Parrot runloop and allocating memory while traversing PMC structures"?
how does the garbage collector take to recursion?
bacek_ it should be fine afaiu. If it's not it's a bug. 09:28
09:33 he_ joined 09:47 payload joined
dalek kudo: fc02ce6 | moritz++ | t/spectest.data:
run more test files
 09:53
09:57 payload joined 09:59 payload1 joined 10:09 riffraff joined 10:16 payload joined 10:21 smash joined
smash hello everyone 10:21
10:23 payload1 joined
lucian treed: hi 10:54
purl hola, lucian.
lucian purl: shut up
purl goes on and on about how much shutting up she's doing
he_ Tried today's build, worked better, but t/compilers/pge/03-optable.t test fails, ref. smolder.plusthree.com/app/projects/...ails/32720 11:02
11:49 contingencyplan joined 12:01 whiteknight joined
Austin Good morning, Whiteknight. 12:01
whiteknight good morning Austin, get all my messages last night? 12:02
Austin yeah
whiteknight I was feeding the kid and only had one hand free, so I figured it would be easier to write you a bunch of long explanatory messages than it would be to just fix the problems
Austin The one is directly correlated to the other - those funky try/catch blocks are there to drain the marshalling queues, so removing them will cause problems.
whiteknight on the bright side, I've gotten much better at typing with one hand
Austin Boy, there's a joke there... 12:03
whiteknight blah blah blah, internet pornography, blah blah blah 12:04
...and then we all chuckle uneasily
Austin blah blah blah ParrotQuotes blah blah blah 12:05
whiteknight purl msg Coke: Your patch in TT #1427, any idea how that is going to affect performance?
purl Message for coke stored.
Austin So I've got nqp generating trampolines to assemble multisubs. 12:06
And I've come across what I think is a bug. 12:07
(Whooo, what a surprise..)
whiteknight I'm not sure I'm entirely clear about what a trampoline is
I had read about them a while ago, and I'm not sure I took the time to really get it then
Austin It's internal, so don't worry. But it's a sub, whose purpose is to bounce you over to another sub.
dalek tracwiki: v52 | Austin_Hastings++ | ParrotQuotes 12:08
tracwiki: trac.parrot.org/parrot/wiki/ParrotQ...ction=diff
Austin You know how websites have landing pages?
You come to one because of google, or because they put it in a tv ad
And then you get redirected to where they "really" want you to go?
whiteknight yeah 12:09
Austin Well, that's a trampoline.
You hit it, it bounces you over to where you're really supposed to go, maybe after fixing your args or something.
whiteknight okay
The wikipedia article on the topic is turning out to be mostly unhelpful 12:10
en.wikipedia.org/wiki/Trampoline_%28computers%29
Austin In my case, multisubs, there's no way to dynamically attach multi signatures to a live sub.
(Wikipedia: I think it's one of those things where you have to know the answer before you ask the question.)
whiteknight Ah, I see what you're saying: Put together a thunk with a multisig that is a closure calling the sub you want 12:11
why can't you attach a multisignature to an existing Sub?
you should be able to add the Sub to a MultSub with the signature
Austin Yeah, but the mmd stuff queries the sub about what it wants to compute the score, I think 12:12
Umm, "not quite" on the whole thunk/closure thing. 12:13
There's no data.
It's just a redirector.
Anyway, let me tell you a little story, and you tell me if this is a bug. 12:14
I've got a multisub with three subs in it.
whiteknight yay! story time
Austin One of them isn't relevant, so I'll ignore it.
The other two have multi-signatures (they're methods) of (_, _) and (_, String). 12:15
The _ is, as I understand it (and I may *very* well be wrong here) sugar for "Any PMC type"
whiteknight it's any PMC type, but matching with the least closeness
Austin I call the multisub using 7 $foo.append("a") 12:16
The NQP generated preserves the literal-string in the code, so the pir looks like $Pxx.'append'("a")
Which of the multisubs should get called? 12:17
whiteknight I think (_, _)
Austin Heh.
whiteknight I don't think the invocant counts towards the multisig
Austin Actually, he does.
whiteknight maybe neither gets called, since the arity of the call is 1, but the arity of the multisigs are both 2 12:18
Austin Because a multisub is a multisub. You can write add(Float, Int) and add(Int, Float) differently, blah blah dispatch
Making it a method just means that the invocant is the first param.
whiteknight are you sure about that? 12:19
nopaste "Austin" at 68.39.12.202 pasted "t/pmc/multidispatch.t" (17 lines) at nopaste.snit.ch/19970 12:21
whiteknight ah, okay then
Austin So down in the bowels of src/multidispatch.c, there's this function called mmd_distance 12:22
whiteknight more like the colon 12:23
Austin And there's a loop at about line 762 that iterates over all the args/multisig
Larry gets the colon.
he_ Hm, why does "./parrot -o runtime/parrot/library/PGE.pbc compilers/pge/PGE.pir" result in a parrot bytecode file with bc_major = 5 and bc_minor = 3, while the current values are 6 and 5 respectively?
whiteknight he_: Do you have an older installed version of Parrot?
he_ At least that's what tools/dev/pbc_header.pl tells me (after I fixed it to work via a minimal patch) 12:24
Austin And that loop updates a variable called "dist"
whiteknight ok
Austin Which purports to be the mmd_distance between the args as passed and the sig as-wanted
(Obviously, this sub gets called once for each entry in the multisub) 12:25
he_ whiteknight: no older parrot installed on this machine as far as I can see.
Austin he_: Are you on unix/linux? 12:26
whiteknight he_: I don't know then. Try make realclean, reconfigure, rebuild
nopaste "he" at 158.38.152.80 pasted "patch for tools/dev/pbc_header.pl to make it work again" (18 lines) at nopaste.snit.ch/19971
he_ Austin: I see this problem on NetBSD/macppc and NetBSD/sparc64, but not on NetBSD/i386. 12:28
Austin run 7 od -t x1 FILENAME | head -1 12:29
look at the last two bytes on the first line, what are they?
(Where FILENAME is your .pbc file 12:31
he_ Says "0000000 fe 50 42 43 0d 0a 1a 0a 04 01 00 02 02 00 05 03" which agrees with my patched tools/dev/pbc_header.pl
Austin Okay, so your file really is 5/3.
how about ./parrot -V 12:32
Does that say 2.2.0?
12:34 ruoso joined
Austin Anyway, whiteknight, the mmd_distance function checks explicitly for string/String (and int/Int and...) promotion, and if the difference between the arg and the param is an autobox, that gets a dist++; continue. 12:36
So the mmd distance between (_, string) and (_, String) is 1.
Otherwise, there is an explicit check for _: if type_sig==PMC, dist++;continue 12:37
he_ Trying the realclean route, did "distclean" earlier.
Austin Which to me says that the distance between _ and any "lowercase type" (int, string, float, short, etc.) is 1. 12:38
he_ parrot -V says 2.2.0-devel 12:39
Austin So the dispatcher is presented with (_,_) which has a distance 1, and (_, String) which also has a distance 1.
he_: That's bad.
he_: Create a x.pir file with two lines: .sub foo ; .end and compile that with ./parrot -o x.pbc x.pir 12:40
he_: Then run od -t x1 x.pbc | head -1 and see if the bytecode revision is correct. 12:41
Coke msg whiteknight: I have no idea how it will affect performance; you should ask bacek, who opened the ticket.
purl Message for whiteknight stored.
he_ On sparc64 I see 6/0 instead of 6/5, and parrot version 2.0(!) 12:42
Austin Lucky for you, Coke will be finished reading the backscroll shortly.
he_ And compilers/pge/PGE.pbc is oldish, Jan 20. Not re-generated, not cleaned, apparently.
whiteknight Austin: Okay,I see what you are saying
he_ Missing dependency / cleanup? 12:43
Austin he_: Sounds that way.
whiteknight so a string matches (_,_) just as well as matches (_,String)
12:43 tetragon joined
Austin whiteknight: Yes, and with the same mmd distance. 12:44
(1)
It seems that there are two separate rules - the switch and the if statement below it - that both are trying to say "to box costs 1"
But in this case, it's not a question of boxing vs. not boxing, but a question of boxing with an explicit rule (string->String) an boxing with an implicit rule (string->_) 12:45
I'm not sure if this is a bug, or not.
(Except of course that any failure to DWIM is automatically a bug...) 12:47
Coke I don't think anything should be generating that file atm. 12:50
(the PGE one.)
he_ ok, that made NetBSD/macppc succeed. 12:51
Coke check the make rule for compilers/pge and see if it generates the file in that directory or if it puts it elsewhere - that might just be a remnanat of an old build.
*remnant
whiteknight Austin: it seems to me it should hit both cases: First boxing string->String has a distance of one, and then the implicit match is another distance of one 12:52
Austin So I should expect what, a junction of behaviors? 12:54
dalek rrot: r44982 | gerd++ | trunk/docs/pdds/pdd30_install.pod:
Add a missing slash
whiteknight basically, yes. A string should match a String more closely than it matches a _
Austin Because this is a manhattan distance computation, there is always the possibility of conflict. That is, of foo(b, x) having a distance of 3 from signature (a, z) and having a distance 3 from signature (d, y) 12:57
whiteknight right
Austin So I can understand a sort of "well, suck it up" approach.
whiteknight I don't know what it does in those cases
Austin But in this case (it's *ME* and I'm *special*) I'm trying to use _ as "anything I haven't mentioned explicitly"
So I want to say that String is closer than _ 12:58
and call it a bug.
whiteknight Austin: I'll tell you what. If you put together a pure-PIR test case to exercise what you want, I'll try to makeit happen
I think that String should definitely be closer than _
Austin I'll put in a ticket, but this is (a) low priority (I've got a work-around - Kakapo ROCKS!); and (b) subject to counter-argument from chromatic, allison, etc. since there might be some aspect I haven't considered. 12:59
whiteknight speaking of Kakapo rocking, are the test failures fixed? I haven't had a chance to test it yet this morning 13:00
nopaste "Austin" at 68.39.12.202 pasted "Kakapo rocks\\!" (6 lines) at nopaste.snit.ch/19975
Austin No. I didn't say "Austin rocks!" 13:01
I was chasing this bug, instead of yours.
13:03 parthm joined
Austin Deja freaking vu. 13:05
whiteknight: You know, some dude found this same bug and opened a ticket (TT#1144) 5 months ago.
Heh. 13:06
Complete with pure pir example.
whiteknight saves you some effort then
Austin No, just pisses me off. It's my ticket. 13:07
he_ OK, I see a little more what's going on. compilers/pge/PGE.pbc got created at some point, but was not deleted after a subsequent update. 13:09
t/compilers/pge/03-optable.t does load_bytecode 'compilers/pge/PGE.pbc'
13:09 atrodo joined
he_ Absent the PGE.pbc file, PGE.pir will be used. 13:10
Currently there doesn't appear to be code to recreate the PGE.pbc file.
It appears that "realclean" won't remove the PGE.pbc file either, it must be manually removed (ugh)
Austin he_++ 13:11
find . -name '*.pbc' 13:12
he_ There's quite a few oldish files lying around. 13:13
whiteknight he_: you may need a fresh checkout. There have been lots of makefile changes lately and it sounds like your working copy is in an inconsistant state 13:16
he_ Well, I've continually updated with "svn update". 13:18
Typically at least once per release.
t/native_pbc/integer_5.pbc is in SVN, created with parrot 1.0, bytecode file format version 4.0 13:19
"that's not going to work", I guess.
plobsing he_: make svnclobber (make sure you don't have any local changes worth saving first) 13:21
Coke yes. svn update, realclean, configure, build, is not guaranteed to put you in a usable state.
he_ oh, that violates the principle of least astonishment...
Coke he_: when you're changing the build system, it's hard to guarantee. =- 13:22
)
a file may become no longer used nor removed, so if you do the build first and regen the makefile, it'll never be removed from that point on, leaving a dead file.
he_ but if I always do "make distclean" before perl ./Configure.pl (which I do), it *should* have cleaned up. 13:27
nopaste "he" at 158.38.152.80 pasted "leftover pbc files in the svn repository" (22 lines) at nopaste.snit.ch/19980
he_ Of these, t/tools/install/testlib/runtime/parrot/library/TGE.pbc doesn't really appear to be a pbc file, and the others are created by various old versions of parrot, and have not been updated. 13:28
Coke he_: no. distclean cleans things it /knows/ about. 13:31
if we change the list of known files between svn revisions, the old ones are no longer known, and therefore no longer cleaned. 13:32
whoops.
Austin find . '(' -type d -name '.svn' -prune ')' -o -name '*.pbc' \\! -name 'pbc_merge_t*.pbc' -print
Coke yes.
plobsing svnclobber the files from space. It's the only way to be sure.
Austin If only we could get that on windows.
Coke Austin: ... 'make svnclobber'
he_: you're right; if you do the distclean first, you'll be in a much better positions. 13:33
Sorry, was stuck in "broken record" mode. =-)
Austin Should that be part of realclean or distclean?
Coke Austin: hell no.
it's the nuclear option.
Austin Huh?
plobsing Austin: no! I like my untracked files!
he_ Well, the Makefile(s) were generated by the old version, so it should know at that point what to clean up after itself.
Austin And if you just typed "make realclean" it seems like you're saying "I don't like my untracked files much any more" 13:34
Coke he_: modulo bugs, yes.
Austin: no.
I love my untracked files.
Coke notes that 'svnclobber' is the nuclear option, not austin's script.
but austin's script still removes things I wouldn't want removed on a realclean. 13:36
Austin How so?
(Actually, since my script doesn't include '| xargs rm -f', it doesn't remove anything...)
Coke Austin: thbbthp.
we should only be removing files our build generated, unless we go to 'svnclobber'. (I'm pretty sure distclean is an alias for realclean atm.) 13:37
Austin But IIUC, the pbc files other than the various special testcase flavors are all generated from pir, no?
Coke that doesn't mean we shoudl remove ones the /user/ created.
Austin Well, clearly what you need is for IMCC to copy the value of the PBC_ORIGIN environment variable into each pbc, so we can tell the difference between build-generated ones and user-compiled ones. 13:39
Or we could figure that realclean means "yeah, rm stuff"
:) 13:40
13:40 mls joined
mls Hi, src/packfile.c looks suspicious in line 4446: 13:41
self->groups = self->groups = mem_gc_realloc...
gcc complains about "causes undefined operation" 13:42
Coke ooc, which version of gcc is complaining about that? 13:47
(fixing shortly)
mls gcc version 4.5.0 20100311 (experimental) [trunk revision 157384] 13:48
Coke spif. 13:49
let me know if there's any thing else that shows up on that compiler. =-) 13:51
Austin whiteknight: I think I've got a Program fix. Best part: fix is in the t/
Coke mls? 13:52
purl mls is Master of Library Science or Minnesota Language Systems or Multiple Listing Service (real estate) or mailing lists
mls Well, it also complained about the no-return-in-nonvoid-function in src/pmc.c, but there's already a comment in svn, so you probably already know about it
Coke packfile.c updated. Thanks. 13:54
13:54 theory joined
Austin msg whiteknight I pushed a src/Program and t/Program bundle to master just now. 13:54
purl Message for whiteknight stored.
mls (The no-return... was in Parrot_pmc_new_init_int, I guess it just needs to return the result of the VTABLE_instantiate() call) 13:55
whiteknight nice 13:57
Coke mls - that's a relatively new function 13:58
Austin You might get some bonus crap if you pull, but the fix is that I explicitly reset the global registry, rather than trying to drain the queues.
dalek kapo: 91fa256 | austin++ | (2 files):
Modified src/Program.nqp to remove inheritance from Library. Modified t/Program.nqp to reset globals before running tests. I think this fixes the intermittent hang that seemed to be test-order dependent.
mls Who uses parrot_config's optimize var? It doesn't seem to be used by pbc_to_exe. Is that a bug? 13:59
Coke very likely.
dalek rrot: r44983 | coke++ | trunk/src/packfile.c:
remove useless self-assignment, courtesy mls++
Coke msg tewk - can you fix the XXX comment in src/pmc's new_pmc_init_int ? needs an explicit return in there, but I don't know what it should be returning. 14:00
purl Message for tewk stored.
whiteknight mls: I don't hink it's a bug. Just because parrot was optimized doesnt necessarily mean you want to optimize everything else 14:05
Coke whiteknight: pbc_to_exe uses "whatever parrot did" to build.
whiteknight Coke: and that's what we want?
Coke so we should try to pass through as many of those compiler option as possible.
whiteknight: unless you want to build a second config system, yes. =-)
whiteknight okay, I won't argue it then. Giving pbc_to_exe some commandline args to control optimizations might be nice 14:06
a -O option would pull Parrot's optimize flags
Coke ... of course, if parrot wasn't compiled with optimization, it'd be useless.
I think having a second set of options there is overkill.
DRY (unless you really need to) 14:07
Coke tries something:
coke_at_work drifts off into the corner.
14:07 patspam joined
coke_at_work SPAMMER! 14:07
purl Spammer, spanner. Spanner, spammer. "Ow!"
coke_at_work patspam? 14:08
purl, bad bot.
purl chuckles evilly
14:09 parthm left
coke_at_work (optimize) - it probably used to work this way until I factored out the warnings & the the optimize flag to make them overridable. I missed this usage because it's in PIR, not a makefile. 14:14
where this way == using the optimize flags.
the config flags are 'ccwarn' and 'optimize', and they should be includable like the other cc flags if someone wants to fix that. 14:15
14:15 bubaflub joined
dalek TT #1513 created by Austin_Hastings++: Test::Builder formats TODO tests wrong/poorly 14:23
14:31 davidfetter joined
nopaste "Whiteknight" at 173.12.37.77 pasted "Current test results after t/Program.nqp fix for Austin++" (89 lines) at nopaste.snit.ch/19982 14:35
Austin Looks just like mine. Whiteknight++
whiteknight nice 14:36
good progress is good
Austin You haven't found out yet that you're boned.
Crap.
I forgot you were depending on Nqp 14:37
whiteknight: You need to pull again to get a re-added File.nqp, which should make the Nqp.nqp testcase work, which I think you need for your external testcase class. 14:43
coke_at_work (yo, I heard you like nqp, so I put some nqp in your nqp?) 14:46
dalek kapo: 8151d15 | austin++ | (3 files):
Reinstated File.nqp to get Nqp.nqp working, for Whiteknight++
Austin Hell, yeah.
(Also, here's at least one function for interacting with the NQP compiler from within nqp code...) 14:47
Nqp::compile_file( 'foo.nqp' )
dalek kudo: 02ba26c | snarkyboojum++ | (3 files):
An attempt at getting .rotate working
 14:49
kudo: d0f934f | (Solomon Foster)++ | src/core/Seq.pm:
Delete trailing whitespace.
whiteknight Austin: I've posted a patch to fix TT #1144. Assuming no complaints, I'll apply it today 14:55
Austin I saw that. Do all the parrot testcases pass?
14:56 PacoLinux joined
Austin okay 14:57
whiteknight yeah, all tests pass 15:02
so barring complaints from the powers-that-be, it's in
too bad I can't retroactively apply it to 2.2
or "2.0" as the locals call it :) 15:03
Austin Heh.
My birthday's in April, so it'll make a nice gift.
whiteknight As a present, I'll play your personal code monkey and try to fix some of these millions of tickets you're filing 15:04
Austin Just using Kakapo has been doing a bunch of good for me, in terms of getting some real-world exposure. 15:05
coke_at_work if there's a 2.2.1 to fixup rakudo build issues, we can cherry pick some post 2.2 commits. 15:07
Austin whiteknight, fyi: I just pushed a Loader fix. You probably don't need it (the way you *need* the File.nqp fix) but it makes that testcase pass. 15:10
whiteknight nice 15:11
coke_at_work: that's cool, but I would be hesitant to change too much in the 2.2.1
Austin BTW, did you read the diagnostic on the MockFS failure you nopasted?
oh, never mind 15:12
It was the wrong test...
I just got this: t/Cuculinae/MockFS.nqp ................ 1/2 # Don't know how to check if String exists. Use a String or Path
Because of the _ versus String thing...
bubaflub i don't know if cardinal people have seen this, but ruby 1.9.2 has passed all the RubySpec tests and the 1.9.2 spec itself will be frozen on March 31st. 15:19
Austin Heh. 15:24
Ironically, the OS.pmc won't tell me what OS I'm using.
How do I do OS-based behavior in Parrot?
coke_at_work Austin: what are you trying to do ? 15:25
Austin Create a different class to support different filesystem semantics
coke_at_work ok. we don't probe for the fs type yet, I don't think.
but you might want to check File.pmc to see if it gives you the methods you want. If you have to roll your own, ,you can key off the config object to at least figure out what OS you're on. 15:26
Austin This is at runtime, not build time. But maybe looking in %config is the right thing?
coke_at_work s/right/closest/, yes.
kind of like us peeking into perl5's build config is the closest parrot has in some places. 15:27
but since we don't really support x-platform yet, it's probably good enough for some time.
Austin What's the origin of the config <osname> key? 15:28
Mine says 'linux', but uname reports 'Linux'... 15:29
15:35 Psyche^ joined
coke_at_work Austin: you can figure that out by acking through the config dir for osname. checking... 15:37
I don't see anything that explicitly sets 'osname', so it's probably coming from P5's Config.
Austin auto/arch.pm 15:38
which comes from get('archname')
coke_at_work ah, it was a multiline. =-)
Austin which appears in p5_keys_whitelist? 15:39
Which appears to mean it gets passed through 15:42
coke_at_work it's not a straight passthrough. 15:43
Austin heh
coke_at_work auto/arch.pm processes it.
not sure what the whitelist is doing there. 15:44
but the source is, ultimately, p5.
15:44 whiteknight joined, Andy joined
Austin Woot. Contextuals look for both caller lexicals and namespace globals. 15:51
Pmichaud++
davidfetter hello 15:53
purl hey, davidfetter.
davidfetter so i'm confused about the latest in pdd16 (nci) 15:54
15:54 janus joined
davidfetter in particular plobsing's patch for same, which appears to change docs so far. does the code need to change along with? 15:55
Austin Changing the PDD implies that any behavior that does not conform to the PDD is in error. 15:56
Which is why it's a mailing list discussion, and not a 3am commit...
:)
15:58 he_ joined
plobsing Austin: it will turn into a 3am commit unless someone provides some feedback, it's not much of a "discussion" at the moment 16:09
Austin Peter, I've seen a few messages go by - maybe 10-15. Maybe you haven't got them, yet? (I sometimes get mail a week late from the list :( 16:12
davidfetter oh, the person in quesion :) 16:14
plobsing runs and hides
davidfetter heh
do you have any interest in postgres?
plobsing I don't have much skill with databases. So not directly. 16:15
16:16 bacek joined
plobsing the PL/Parrot stuff just happens to be running into the same problems I've run into a couple times now. 16:16
davidfetter this all came up because we're trying to make parrot (eventually) pg's language engine 16:17
:)
who's, "we?"
er, where are you running into these problems?
<-- ETOOLITTLECOFFEE 16:18
plobsing 1.) frame builder 2.) Parrot::Embed 3.) vim-parrot (personal project, not yet published)
davidfetter <-- vim user :) 16:19
cotto_work bacek or bacek_at_work, ping 16:24
16:25 kjeldahl joined 16:32 jsut joined
plobsing davidfetter: also I have a vendetta against varargs 16:33
Austin D'oh. Storing contextuals doesn't check for dynamic vs. global.
davidfetter sends plobsing up the nung river in a pt boat
16:38 whiteknight joined
whiteknight internet connection is teh suxxor 16:40
allison: ping 16:41
dalek kapo: 70ee5af | austin++ | (3 files):
Fixed ordering test in Loader.nqp

Signed-off-by: Austin Hastings <Austin_Hastings@Yahoo.com>
16:47 Mokurai1 joined
dukeleto 'ello 16:49
bubaflub afternoon duke 16:50
dukeleto bubaflub: howdy
plobsing: yes, PL/Parrot is running into lots of sharp corners of the Parrot embedding API :)
plobsing dukeleto: embedding parrot, its hard not to run into sharp corners. 16:52
16:52 chromatic joined
dalek TT #1514 created by Austin_Hastings++: NQP-rx doesn't check storage mode for contextuals 16:52
whiteknight I would like it if dalek would print a link to the ticket 16:53
Austin You know, I suggested making that a bot 16:54
But they told me, "Oh, that's in the irclogs."
whiteknight what's in the logs? 16:55
Austin Apparently, nobody but us suxxors actually follows IRC. Everyone else follows the logs.
darbelo Austin: That's not the same as 'no'.
Austin The TT# -> link converter
whiteknight Ah, that would be a cool bot
plobsing or you could add a script to your irc client to match patterns like that and make links
whiteknight even cooler if it were written in Parrot
an IRC library would be a cool project
plobsing: I use at least two IRC clients 16:56
Austin plobsing: Or I could use firefox's built-in keyword expansion stuff to make tt# xxx work.
plobsing whiteknight: that just means you get to do *more* scripting! scripting is fun right? 16:58
dukeleto Austin: i would very much like a link to the TT as well 16:59
dalek ?
purl dalek is #parrot's spammy little rss bot or (see: dalek plugins)
dukeleto dalek plugins ? 17:00
purl i heard dalek plugins was github.com/Infinoid/dalek-plugins/tree/master
Austin I was thinking more of a live rewriter, that would translate anyone's tt# into a link. 17:03
I say tt#2
and ttbot says: TT# 2 is at trac.parrot.org/parrot/ticket/2 17:04
(Or whatever)
darbelo ttbot?
purl ttbot is TapTinder build bot owned by mj41 and reporting tt.taptinder.org/buildstatus/pr-Parrot/rp-trunk build errors.
Austin I wonder if purl is smart enough for that
So ttbot is already taken... 17:05
It was the obvious name..
purl help?
purl #perl is not a help channel, and I'm not a help bot. If you want Perl help, try #perl-help or #metallica. or (see the 'help channel' factoid as well) or
Austin fna
bubaflub we had something like that for our companies ticket system
i might be able to resurrect the code
all i'd need is a server to run it on so it could listen in the chanel 17:06
everytime we'd say #1234
it would spit out the link to our internal tracker
Austin I'm wondering if purl is smart enough for that..
whiteknight I had a bot like that for Wikipedia. Long time ago
Austin What software is she running?
cotto_work infobot
probably with many additions 17:07
Austin If this is magnet, she's flooterbuck. 17:08
17:14 cosimo joined
dalek rrot: r44984 | gerd++ | trunk (5 files):
add the option: "perl Configure.pl --pkgconfigdir=DIR"
 17:16
Austin tinyurl example.com/test 17:25
purl Your tinyurl is tinyurl.com/4o6x2s
Austin Thanks, purl.
17:28 ruoso joined
allison whiteknight: pong 17:29
whiteknight allison: can you take a look at my patch for TT #1144
fixes the ticket and passes all tests, but I want an eye to double-check the approach 17:30
there's a question about whether this is even the intended behavior, which I think it should be
allison sure
whiteknight thanks! 17:31
17:31 ash_ joined
allison whiteknight: looks great! 17:33
whiteknight awesome, thanks!
allison++
dukeleto i thinks making dalek spew the TT link when it announces a new ticket would be the best + easiest 17:34
Austin It's a start. 17:35
dalek kapo: 5c3b3a6 | austin++ | (2 files):
Got the string-oriented parts of Path working, tested.
 17:42
chromatic +1 on the MMD patch. I'm not sure why it didn't work that way before. 17:44
whiteknight thanks. I'll commit it in a few minutes 17:47
I'm working on a patch right now for the RO stuff in tt389_fix
looks like it fixes at least a few broken tests
chromatic The fact that the RO vtable doesn't have the proxy in pmc_class? 17:48
whiteknight yeah
I'm basically sharing it. vt->ro_variant_vtable->pmc_class = vt->pmc_class
17:48 brooksbp joined
whiteknight a little ugly, but workable 17:48
coke_at_work Would anyone care if I removed all the docs-specific targets? 17:49
chromatic That was my best idea too for now.
whiteknight I'm doing a comparison right now to double-check that I'm fixing tests and not regressing anywhere 17:50
I'm still seeing plenty of places in the code where things are being looked up in vtable->_namespace 17:51
and those are leading to Null PMC access errors
I think if we find and hunt down the rest of those we will be in better shape
chromatic Places that look up methods? 17:53
whiteknight yeah
also, the issue with inheritance
If you could throw together a function like you were talking about a while ago to walk the MRO and look for methods, we could plug that in to the places that need it 17:54
probably a bunch of cut+paste from Object.pmc:find_method
chromatic Can probably do that. 17:55
dukeleto looks like github.com/Infinoid/dalek-plugins/b...cketlog.pm needs to be slightly modified to emit a TT URL
bubaflub dukeleto: i think it's calling the emit_karama_message in karmalog.pm 17:57
github.com/Infinoid/dalek-plugins/b...og.pm#L167 17:58
right after that
coke_at_work aooga: 18:02
The OSUOSL will be performing an IOS upgrade on our primary Cisco router
this coming Wednesday, March 24, 2010 between 7:00-7:30PM PDT (0200-0230
UTC). Networking for *ALL* of the OSUOSL will be down for approximately
5 minutes during the outage and will be restored once the update is
complete.
purl Nothing ever is.
coke_at_work (this means us)
Austin Heh. 5 minutes. 18:03
whiteknight yeah, 5 minutes in IT time is damn near all day 18:04
cotto_work panics
dalek rrot: r44985 | whiteknight++ | branches/tt389_fix (2 files):
share pmc_class between a vtable and the ->ro_variant_vtable. Fixes 6 tests in t/dynpmc/rotest.t
 18:06
rrot: r44986 | whiteknight++ | trunk/src/multidispatch.c:
[mmd] change the mmd distance calculating algorithm to count autoboxing as one step, and implicit type matching as a second. calling with an S register should match (String) more closely than (_). Fixes TT #1144
dukeleto bubaflub: mad props to you if you add TT URLs :)
dalek TT #1144 closed by whiteknight++: MMD incorrectly matches _ instead of String 18:14
whiteknight yay karma! 18:15
bubaflub dukeleto: i've got a patch for it to show TT URLs (which I *think* should work); where should i send it? 18:16
or should i just do a pull request from github?
whiteknight ah, my last tt389_fix commit fixed more bugs than I thought. the dynpmcs don't rebuild when we make changes to Pmc2c 18:17
dukeleto bubaflub: Infinoid. yeah, a pull request is probably best 18:18
bubaflub "Hardcore Forking Action". ah, github.
dukeleto bubaflub: yeah, that always makes me happy :) 18:19
bubaflub ok, pull request sent to Infinoid through github 18:21
we'll see what happens
Austin bubaflub: What's the modified structure of the dalek messages?
bubaflub Austin: just a second line below the first 18:22
with the URL
Austin Okay.
bubaflub github.com/bubaflub/dalek-plugins/commit/37807a
to see the changes, pretty simple really.
Austin Moritz' irclog stuff processes the contents to make #123 into a link, so you'll want to make sure that stays in.
bubaflub okey dokey 18:23
if we wanted to get the ticket URLs from other places i could make some quick edits as well 18:24
whiteknight chromatic: I think all remaining test failures in tt389_fix are because of inheritance issues
we get that worked out, everything should be fixed.
Infinoid bubaflub: thanks for the patch. Any objection to including $prefix on the link line too? 18:25
bubaflub Infinoid: not at all.
(at least from me) 18:26
so something on the second line like:
Infinoid cool. I'll try it out tonight and merge it if it looks good
bubaflub TT #1513: trac.parrot.org/parrot/ticket/1513
dukeleto +1 to making the TT link on the next line
Infinoid yeah. I think I may need some special : handling, but it'll look close to that
dukeleto Infinoid++ bubaflub++
Infinoid should be trivial :)
18:27 iblechbot joined
chromatic whiteknight, are the inheritance failures obvious from the failing tests? 18:36
whiteknight chromatic: I think all the remaining failures are inheritance failures 18:37
the only one I think might be a different issue is the failure in t/pmc/objects.t
chromatic I see. 18:38
whiteknight It's not finding a vtable override in a subclass of Integer, and I suspect it might be stored in a namespace somewhere
dalek rrot: r44987 | whiteknight++ | branches/tt389_fix/t/pmc/resizablepmcarray.t:
[tt389_fix] add exception handlers to one RPA test. When this test throws an unhndled exception the rest of the test file doesn't run at all. Now we see the one failure, and we also see that all other tests in the file pass
whiteknight or, the mechanism to find the vtable override isn't searching in the proxies correctly
chromatic Oh, right. 18:39
That might be a case where :vtable and :method interact badly.
whiteknight likely, yes
chromatic Did you fix something like that in trunk or on the branch?
dukeleto mj41++ # Code coverage on tapir2.ro.vutbr.cz now has GMP support
whiteknight in trunk. I fixed it so :anon :vtable was entered into the namespace correctly
this one isn't :anon, but I'm sure a similar mechanism is at play. 18:40
chromatic Right. 18:41
ash_ for nci, is parrot planning on using libffi?
whiteknight brb. VisualStudio demands the computer restart for an update
chromatic libffi is an option for an NCI backend. We haven't decided on one. 18:43
plobsing chromatic: are we going to decide on only *one*?
chromatic I don't know that.
plobsing althought if we were to choose 1, can't complain about libffi
chromatic The cost of supporting multiple might be significant. 18:44
See also: runcores, JITs, garbage collectors, PIR compilers, STRING subsystems....
ash_ i just heard that libffi doesn't work with win32, (well it does work with cygwin and ming) so i was wondering
plobsing ash_: url? 18:45
lucian ash_: i don't think that's true. for example python's ctypes works just fine on windows
chromatic If that's true, that's a point against it (or a point for us to encourage it to work well with Windows).
ash_ sourceware.org/libffi/ says Windows/Cygwin and Windows/MingW as supported platforms, i took that as it won't work with the windows compiler... 18:46
i could be wrong though
cotto_work msvc support is important on windows 18:47
lucian ash_: python-win32 is compiled with msvc and its ctypes (based on libffi) works fine
ash_ lucian: that may be, i am just saying thats what the libffi site says, i don't have a windows computer to test that though 18:48
lucian ash_: right. a quick google suggests that there's a set of patches maintained for msvc compat
and code.google.com/p/libffi-msvc/ 18:49
dukeleto +Inf to libffi
plobsing since we're talking about nci, has anyone had a look at my proposal for pdd16? 18:51
bubaflub libffi on msvc: sourceware.org/ml/libffi-discuss/20...00049.html
sent earlier this month from the maintainer of that google code project; seems like things are working
lucian bubaflub: so it really is a non-issue 18:52
bubaflub lucian: the follow up emails said they wanted to merge it in
lucian bubaflub: even if it doesn't get merged, maintaining a libffi patch is much easier than using anything else 18:53
bubaflub "I tried these patches on mswin32/MSVC9 and it works fine, though current github/atgreen/libffi/master doesn't work on my env."
is in the follow up email;
i think it's safe to say that MSVC stuff works for libffi and there are active commits in the google code project i.e. commits this month 18:54
though we should still investigate, i think it's premature to say that it *doesn't* work with MSVC 18:55
cotto_work "It's also possible to build libffi on Windows platforms with 18:58
Microsoft's Visual C++ compiler." from github.com/atgreen/libffi
smells like a winner
bubaflub that's a bingo 18:59
19:02 whiteknight joined 19:07 eric_j joined 19:19 TiMBuS joined
lucian allison: ping 19:23
19:23 AndyA joined
dalek TT #895 closed by NotFound++: Towards automatic allocation and deallocation of PMC attributes 19:36
TT #1493 closed by NotFound++: Avoid repeated STRING constans with the same value
NotFound sdl? 19:37
purl sdl is Simple Directmedia Layer aka DirectX for Unix except with 59% less evil. There's Perl bindings. Frozen Bubble is written using it. or or allows that nifty doom game to run on ayrnieu's *Zaurus*
19:41 AndyA joined
whiteknight chromatic: I think I may have the tt389_fix issue figured out 19:43
and it's much much easier than I expected
19:44 eric_j joined
whiteknight lib/Parrot/Pmc2c/PMCEmitter.pm:get_mro_func, the generated vtable->mro for a built-in type is an RSA of type names. I'm converting that to an RPA of PMCProxies. src/oo/find_method_direct_1 should now walk the MRO correctly and find methods as appropriate 19:44
chromatic Ah, that sounds reasonable. 19:46
coke_at_work whiteknight++ 19:47
whiteknight that might also explain a handful of other bugs if subclasses are inheriting that "MRO" and not being able to lookup methods in it
chromatic Right.
We'll probably be able to delete a lot of code too.
whiteknight of course, it may create some bugs if anybody is relying on this lousyness 19:48
...and miniparrot segfaults. So, there's my answer
allison lucian: pong 19:49
lucian allison: it seems i was somewhat wrong about parrot's objects described in PDD 15
allison lucian: how so? 19:50
lucian allison: they're less than ideal (as in i'd love to see some changes), but it should be possible to implement python objects on top without horrible hacks or perf hits
allison lucian: excellent! 19:51
purl EGG-see-lent!
lucian allison: the description makes them sound like java objects and not fitting everything-is-an-object languages at all
allison: but they expose their vtables, so it should be fine
allison lucian: well, in Parrot we have to deal with many different kinds of languages 19:52
lucian allison: yes, which is odd that the object system is so perl-oriented
allison lucian: some are everything-is-an-object, some aren't
lucian: is it perl oriented? it was designed to be pretty generic
lucian allison: it's less generic than it should be 19:53
allison (and the folks developing Perl say it's not perl oriented)
lucian oh. then i probably don't know perl enough (which i don't)
NotFound lucian: I don't see 'bless' in parrot ;)
allison Rakudo is prototype-based, so using a different overlay on top of Parrot objects
19:54 joeri joined
allison lucian: but, I'd agree it's not strictly Python objects, and may need overlays there too 19:54
lucian allison: i'm thinking python's object will inherit Object and for example python's str will inherit both object and String
allison: in any case, it seems parrot's objects are a mix of high and low level objects 19:55
allison: i think i should just start trying to implement object and see how it goes 19:56
allison: what i don't understand is how interop will work with parrot's objects as they are now 19:57
allison lucian: interop is via vtables 19:59
at least "interop" in the sense of accessing attributes, calling methods, etc
take a look at docs/pdds/draft/pdd31_hll.pod for the broader sense of interop 20:00
TimToady phone, I guess
allison lucian: loading modules across languages, etc
TimToady: dialing
20:00 cxreg joined
whiteknight chromatic: ah, I see what's going on. Parrot_*_get_mro is creating an RSA of type names, then Parrot_create_mro is walking that list and creating namespaces and PMCProxies from that list 20:00
chromatic Seems like two steps that could be one. 20:01
lucian allison: so each object/class puts things it might want others to call in its vtable?
whiteknight that's what I'm thinking, but will require some rewriting
chromatic lucian, are you familiar with message passing? 20:02
lucian chromatic: yes
whiteknight chromatic: I'm heading home from work now. I'll tackle this more when I get home. Talk to you then
lucian chromatic: so it's smalltalk-style?
chromatic No, but I was going to explain by analogy. 20:03
lucian chromatic: i meant smalltalk-style as in you call a method on an object and it decides what to do 20:04
chromatic Yes, like that.
Except that "find a method" is also a method. 20:05
lucian chromatic: right. so you could even handle conversions between naming conventions? 20:06
chromatic Sure, however you store and look up methods depends on your object/class.
Tene you'd need to be able to subclass Class to have a class that handled that differently than Class.pmc does, though. 20:10
Which we can't do.
lucian Tene: then how are metaclasses implemented? 20:11
allison: i wasn't sure how i could implement this gist.github.com/335672 20:13
20:24 bacek joined
cotto_work bacek, what was that ping about? 20:25
bacek o hai
cotto_work helllo
bacek cotto, about your last commit into ops_pct. You changed number of parsed ops in test. 20:26
cotto_work That could have been an incorrect change. 20:27
I'd like to make that test less brittle. It's non-trivial to verify if the test is failing or if it's just broken. 20:28
bacek If no ops were removed test should pass. 20:29
cotto_work all tests pass with the opsc-built core_ops.c so I'm inclined to blame the test.
bacek :)
cotto_work That also makes me suspicious.
allison lucian: wasn't sure you could implement which? Do you mean "A.print2 = f"? 20:32
lucian allison: that and print3
NotFound Someone took a look at TT #857 ? I'm checking with Strawberry perl in XP Home and the result of the snippet looks good to me. 20:33
lucian allison: in fact, that's how class definition works in the first place, the rest is sugar 20:35
allison lucian: adding a method to a class at runtime is the 'add_method' vtable function 20:37
lucian allison: yes, i found that out in the meantime
cotto_work unfortunately the way to verify it is to make ops2c process those files and see how many ops it generates 20:41
btw, initial (if rough) self-hosting is accomplished by the bootstrap-ops makefile target. That item can be taken off the TODO list. 20:42
lucian allison: also, i'd like to see pynie as self-hosted as possible 20:43
cotto_work For verifying that the proper number of ops were produced, it'd probably be best to have a dedicated .ops file in t/ .
allison lucian: self-hosting as in written in python?
Austin Allison: what's the position on duplicate named arguments? I'm seeing arguments from calling, e.g., foo( :a(1), :a(2) ), but it strikes me that this is the obvious way to provide and override defaults...
allison lucian: or self-hosting as in not using NQP? 20:44
lucian allison: for example, all objects should be written in pynie
cotto_work bacek, could you update the TODO with that? I'm not sure how soon I'll have tuits.
lucian allison: i can't see how nqp could be avoided for the compiler
Austin *seeing arguments = seeing errors
allison lucian: if we went that route, we should just use PyPy
lucian allison: i just want to avoid nqp use if possible 20:45
allison lucian: not sure we're going to get the speed that way, though
lucian: aye, I'd like to use Pynie as an excuse for more advanced compiler tools
lucian allison: with object implemented, dict and the like should be doable in python
20:45 brooksbp_ joined
allison lucian: as in, parse the Python3 grammar directly and generate the compiler from it 20:45
lucian allison: there would be a need for an interface to parrot from pynie, like a fake module
allison lucian: I suspect they'll be slow, though 20:46
lucian: they're faster as low-level objects
lucian allison: why? they'd inherit parrot's Hash and ResizablePMCArray 20:47
allison: something like this gist.github.com/335708 20:48
allison lucian: sure, but right now they're written in PIR, as a thin overlay
lucian allison: would compiled python be much slower than pir? 20:49
allison lucian: at the moment, yes, because parsing python is slow
lucian: the parser is slow in general
lucian allison: parser aside, why would it be slow?
allison lucian: down the road it probably won't make much difference
lucian: but choosing between reimplementing features that are already working, and implementing more features, the second is a higher priority 20:50
lucian allison: dict and friends would have to be reimplemented anyway after objects exist
allison lucian: well, somewhere down the road try it out and benchmark the two 20:51
lucian allison: i don't much see the point of rewriting dict in pir to inherit from python's object than rewriting it in python directly 20:53
eric_j hi all, i would like to work on pynie. i would like to call set_key_type and set_value_type for the dictionary, but i can't figure out the syntax
allison lucian: honestly, I was thinking of moving all the Python core types to C, but didn't figure it was worth it yet, until we nail down the semantics further 20:55
lucian: figured PIR was good for prototyping
lucian allison: a good jit would have about the same performance as writing it in C 20:56
allison lucian: aye, which we may have by then anyway
lucian allison: the PyPy folks have showed that extensively with micronumpy and other microbenchmarks
allison: btw, i've read the jit page, has anything been decided? 20:58
allison lucian: it's been decided to try multiple toolkits and see which serves our needs best 21:00
lucian: still leaning toward LLVM, especially with the work the unladen-swallow team has put into it
lucian allison: well, the u-s guys have run into a lot of issues
allison: and there's no chance of u-s being anywhere near as fast as PyPy for example
allison lucian: they've resolved most of their issues 21:01
chromatic LLVM still has to address memory usage. 21:02
allison lucian: and the speed could be considerably enhanced by caching the compiled code
lucian allison: most of their, but not llvm's. it's still really bad at patching code
allison lucian: yup, so you can see why we're not committing to it
lucian: the other options are not much better
lucian allison: it may well be possible that with a mid-level layer (parrot) would make it possible to generate fast static binaries out of high level languages 21:03
using llvm
NotFound Talking about python and speed? Someone wants to revenge the pie in the face? X-)
lucian NotFound: what pie?
purl i heard pie was true. or www.piecouncil.org/national.htm or london.randomness.org.uk/wiki.cgi?a..._value=Pie or www.austinthirdgen.org/upload/piechart.jpg or www.weebls-stuff.com/wab/ or flickr.com/photos/cowfish/3137913195/ or dilbert.com/fast/2009-03-07/ or position-independent executable (like PIC libraries but more so)
lucian purl: bad bot
purl chuckles evilly
allison lucian: part of it is, state-of-the-art JIT techniques tend to focus on static languages
LLVM, for example makes all sorts of bad assumptions about whether code will be modified later
lucian allison: it would also be possible to rewrite parrot in RPython and use PyPy's jit :D 21:04
NotFound lucian: an old parrot tale
atrodo allison> I'm curious how so
allison lucian: yes, implementing a JIT once in Parrot is a potential advantage to a number of languages
lucian: ha-ha :)
lucian allison: i'm semi-serious about that :) 21:06
allison: it would be fast, but it's a LOT of work
allison lucian: seriously, though, implementing RPython in Parrot is a good goal 21:07
lucian allison: no, parrot in RPython :)
allison: actually, a RPython variation might be useful as a NQP-for-python
allison lucian: I'm entirely confident we can do better than PyPy in Parrot 21:08
lucian allison: performance-wise?
allison lucian: yup 21:09
lucian allison: i'm still skeptical. PyPy does loads of very interesting and hard things
allison: but having that mid-level abstraction layer might help
allison lucian: aye, those are interesting and hard things anyone can do, even Parrot 21:10
lucian: but, that's down the road 21:11
lucian: first, full feature support
(in the language implementations) 21:12
lucian allison: i know, i was just saying they have a head start]
allison lucian: in some areas, yeah
lucian: they've done great work
lucian allison: (they did the same thing, first a slow python-in-python, then a translation framework to C, then automatic JIT addition) 21:13
allison lucian: and, they're happy to share too
lucian allison: yes, i can't wait till we can run py.py on pynie to stress-test it
21:13 Whiteknight joined
lucian allison: i'd love to see PyPy become the reference implementation (especially since the interpreter is written in pure python) 21:14
allison lucian: aye, one of the reasons I'd steer clear of python-on-python in Pynie, is that PyPy has already done it so well
lucian: kind of "been there, done that"
lucian allison: yes, python-on-python would be better as a parrot backend to pypy 21:15
allison lucian: Parrot's competitive advantage is the middle-layer
lucian allison: but that would be a double interpreter
allison lucian: and keeping the individual language implementations as thin as possible on top of that middle-layer
lucian allison: yes, but i still think as much of the core types should be python :)
eric_j speaking of the full implementation, can someone help me calling set_key_type from PIR? i will add the answer to the wiki... 21:16
allison eric_j: got distracted in the conversation. what do you mean by set_key_type?
eric_j allison: in dictobject.pir, i am trying to make parrot not coerce the keys to be strings. i understand that parrot's "hash" object has a method called set_key_type that will let you change the key type to be objects, but i can't figure out how to call it 21:18
allison lucian: in the long-term, it doesn't much bother me either way whether they're in Python or PIR or C. In the sort-term, it's a waste.
lucian allison: a waste to write them in C, true 21:19
allison eric_j: there is no way to do that at the moment in Pynie, as Parrot's core dict-like type only stores string keys
eric_j hmm
allison eric_j: so, we'll have to make extensions to that type
chromatic I thought bacek fixed that.
lucian allison: but since dict, list and type have to be rewritten in terms of objects anyway (and they're small), why not write them in python? i don't see the waste 21:20
eric_j i was hoping that that was what set_key_type does
allison chromatic: we talked about it, not sure that it happened
chromatic: it's at least not there in 2.0, which is what pynie runs on 21:21
bacek It was in 1.6 I think.
hash = .Hash_key_type_int
or .Hash_key_type_PMC 21:22
allison bacek: dicts have mixed string and integer keys
bacek: and preserve the types of those keys
eric_j in python, dicts can have any hashable key
bacek Then use PMC with autoboxing/unboxing 21:23
allison bacek: aye, we can do that
bacek Actually, Hash will autobox/unbox keys automatically. 21:24
ETOOMANYAUTO 21:25
eric_j bacek: can you help me with the PIR? i'm trying to write $P0.set_key_type (Hash_key_type_int), but evidently i'm in the wrong rabbit hole
allison bacek: is there a way to set that once for the subclass?
bacek allison, no afair.
allison bacek: that option seems to be set once you've already instantiated the object 21:26
lucian eric_j: $P0.'set_key_type' Hask_key_type_int
bacek eric_j, $P0 = .Hash_key_type_PMC
lucian eric_j: ignore me, i assumed something wrong
NotFound In a init vtable override in the subclass? 21:27
Austin eric_j: .include 'hash_key_type.pasm' ; $P0 = new 'Hash' ; $P0 = .Hash_key_type_PMC
bacek eric_j, $P0."set_key_type"(.Hash_key_type_int) should work as well
allison bacek: I can do it in the dictmaker function, which is what instantiates most dicts
eric_j thanks guys
bacek allison, should work. Or override SubclassedHash.init to do it always. 21:28
dalek TT #1515 created by Austin_Hastings++: Duplicate named args cause fatal error in subs 21:31
allison bacek: those constants are defined in 2.0, or at least aren't installed 21:34
bacek: *aren't*
bacek allison, runtime/parrot/include/hash_key_type.pasm 21:35
you have to manually include it.
allison bacek: aye, the file doesn't exist in a 2.0 install
bacek Ah, may be...
allison bacek: (found it in the current trunk no problem)
bacek: it's a generated file, might not be listed in the MANIFEST.generated 21:36
nopaste "NotFound" at 213.96.228.50 pasted "Setting hash key type in subclass init, winxed example" (17 lines) at nopaste.snit.ch/19988
cotto_work bacek, any thought on/objections to adding a fixed .ops file to compilers/opsc/t for use as data? 21:37
allison NotFound: what language is that?
purl I'm not sure... but it doesn't sound like any kind of English I know.
21:37 brooksbp joined
bacek cotto_work, nope 21:37
NotFound allison: Winxed 21:38
allison NotFound: interesting language 21:39
NotFound allison: thanks
nopaste "NotFound" at 213.96.228.50 pasted "Setting hash key type in subclass init, pir generated by the winxed example" (52 lines) at nopaste.snit.ch/19989 21:40
lucian NotFound: looks a bit like C#. not aesthetically nice, but interesting 21:42
NotFound lucian: the syntax is borrowed from javascript with bits of c++ and bits of C# 21:43
lucian NotFound: reading the code.g.c page
allison bacek: that was it, hash_key_type.pasm wasn't listed in MANIFEST.generated, so wasn't being installed 21:44
lucian: we love all kinds of languages here :) 21:45
bacek allison, "installed parrot" is totally unknown beast to me.
lucian allison: i'm not saying it's not nice, it just looks bad :)
allison: since most people are used to C-like syntax, it's a necessary evil 21:46
allison bacek: it's a good idea to use it installed whenever possible. Though, what we need most is to be able to run the full test-suite on an installed Parrot 21:47
NotFound lucian: specially when the language creator belongs to that group of people.
lucian NotFound: heh. i'm half-used to it myself. but i think syntax matters, and C syntax is a bad idea 21:48
NotFound lucian: so bad that it dominates during decades ;) 21:49
allison lucian: all syntacies have a place and a use
lucian NotFound: yes, it's unfortunate
allison: that doesn't mean they encourage good practice or they're readable 21:50
21:51 kjeldahl joined
allison lucian: I don't know, I've found people can write good or bad code in any language 21:51
NotFound lucian: No one person has yet asked what a winxed snippet posted here means. To me that qualifies as readable.,
allison lucian: and readability is partly a matter of writing good code, and partly a matter of what you're familiar with 21:52
lucian allison: of course. but why not help the programmer write better code if at all possible?
readability is relative to experience, yes. but any syntax has an inherent readability 21:53
allison lucian: absolutely, and for one programmer Python may be the answer to that, and for another programmer C, or Haskell, or Go might be the answer to that
dalek kudo: fed4687 | jonathan++ | docs/ROADMAP:
Tweak the ROADMAP with the results of The Meeting.
allison lucian: the polyglot programmer is on the rise 21:54
lucian allison: that doesn't make C syntax an more helpful
dalek rrot: r44988 | allison++ | trunk/MANIFEST.generated:
[install] Adding the hash-key type definitions to the manifest of
lucian allison: it's terse where it shouldn't be, it has too much noise
allison lucian: so it's not the language for you, no problem 21:55
NotFound lucian: if you want to rewrite C, you need a flux condenser or something,
allison lucian: I have spent far more time writing C than high-level languages the past 6 years, and it's grown on me
lucian: for specific kinds of tasks
lucian: though, I wouldn't recommend it as someone's first language 21:56
NotFound Yeah, Z80 assembler is much better X-)
lucian allison: almost everything about C is nice except part of the syntax
allison lucian: maybe an acquired taste, like black coffee, or 99% dark chocolate
lucian allison: for example, ; at the end of lines. there's no real reason for that 21:57
NotFound Mmmm... coffee, good idea.
lucian NotFound: mmm, dark chocolate
allison lucian: <shrug> character-delimited statements are easier to parse than newline delimited statements 21:58
lucian: and for a low-level language, I'm okay with some syntax decisions made for ease of implementation rather than ease of use
lucian allison: parsing isn't really an issue. and i don't think ease of implementation should ever dictate syntax 21:59
allison: perhaps at the time it made sense. but nowadays, it doesn't, at all 22:00
22:00 he_ joined
NotFound I don't think that ending a sentence without puntuaction is more readable. Books use it from hundred of years ago. 22:00
allison lucian: PIR, for example has no statement terminators, but it's also so dead simple that there are no compound statements 22:01
lucian: when a language has compound statements and multi-line statements, a character to mark off the end of a statement is pretty handy 22:02
lucian NotFound: books also don't use { } for paragraphs
allison lucian: the worst use of closing semicolons I've ever found is Matlab
lucian goes to read about matlab
allison lucian: where they're optional, but determine whether the result of the statement is automatically printed
lucian: language design is all about balancing 22:03
lucian allison: oh. that's bad
NotFound lucian: books uusally doesn't have loops. 22:04
lucian allison: i know, and i don't think C syntax is quite as well balanced as it could be
allison lucian: certainly annoying to someone who's working primarily in C and Python lately :)
lucian: I should send you a book chapter I once wrote about language design
lucian: it's an absolutely fascinating topic 22:05
lucian allison: indeed
allison lucian: it's like going from loving Rembrandt, to understanding the choices all painters make
(while still loving Rembrandt) 22:06
lucian NotFound: true, but in code you delimit statements with line breaks anyway, for style
allison: i do like several languages, while preferring python
22:07 cotto_work joined
NotFound lucian: yes, I like to put line breaks where I think looks good, not where some language designer wants to force me. 22:08
lucian NotFound: in this case you're the language designer
NotFound: and you can put line breaks anyway, but since breaking the convention is more rare than not, \\ for line continuation is less noisy than ; for line break 22:09
s/line/statement/
NotFound lucian: yes, and I don't want to make to others what I don't want others make on me.
lucian: oh... \\ for line continuation... just like C preprocessor? ;)
lucian NotFound: a bit, i guess 22:10
NotFound Anyway, the main reason for me to joining parrot is to avoid syntax wars. Write your code in the language you like, and use modules written in any other. 22:12
lucian NotFound: true. very nice 22:13
NotFound: not just syntax wars, but it's DSLs done right
NotFound: for example i'm fine with either ruby or python, even though most of their differences are syntactical 22:14
NotFound lucian: that's what some religions says: no religion wars, just follow the right god ;)
lucian NotFound: you have have read me right, i meant that you get real DSLs on parrot, not semi-DSLs as in ruby
s/you have read me right/you haven't read me right/ 22:15
NotFound Oh, I don't follow much the discussions about DSL, and don't know enough Ruby to understand the recent posts on that subject, 22:16
22:17 cotto_work joined
lucian NotFound: basically ruby syntax allows you to write stuff that's ruby, but looks a bit like another language 22:17
NotFound Yes, with parrot is easy to write your own langauge, either DS or general purpose.
lucian NotFound: yes, that's what i meant
NotFound I hope Winxed is a good example of the later. 22:18
lucian NotFound: i sure looks that way :) 22:19
s/i/it/ 22:20
NotFound At least is useful to write parrot features examples a lot shorter than in pir.
And with better runability than seudocode :D 22:21
22:22 GodFather joined
eric_j allison, just tried the set_key_type, it works pretty well. however, it looks like __init__ is not automatically being called. i.e., i have to do x={}; x.__init__() because i put the set_key_type in __init__ 22:40
lucian NotFound: a language that targets parrot's low level types is indeed useful 22:41
NotFound: i believe there was another such language, Close
NotFound lucian: yes, but looks like there is not much activity in it these last months.
allison eric_j: I'll get it built-in to pynie soon 22:47
eric_j: have the patch, but it won't work until Parrot 2.3 22:48
allison afk
eric_j allison: no problem, i'm working from subversion. if you have an unstable branch in subversion i'll switch to that 22:53
22:54 tetragon joined
dalek rrot: r44989 | NotFound++ | trunk/CREDITS:
fix my entries in CREDITS
 22:59
japhb plobsing, just sent a review of your NCI PDD patch to the list. 23:06
dukeleto japhb: nice writeup 23:17
japhb dukeleto, thanks!
chromatic dukeleto, could we have a configure step for configure build directory? 23:30
dukeleto chromatic: that sounds nice 23:32
chromatic: how would one go about doing that? 23:33
chromatic I'd ask kid51.
dukeleto chromatic: ok. i will make a TT
chromatic: that very well may help with the RTEMS port as well 23:34
chromatic I think so. 23:35
23:35 cotto_work joined
Whiteknight chromatic: I'm getting pretty close on tt389_fi 23:55
the new code is much smaller, probably faster 23:56
chromatic Good, I should have some spare time late tonight and tomorrow. 23:58
dalek TT #1516 created by dukeleto++: Allow a build directory to be specificed to Configure.pl 23:59
23:59 patspam joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!