pugscode.org <Overview Journal Logs> | smoke/win2k (occasional): xrl.us/fqum (867/3781, r2479), Mac OS X (62/4506, r2362), Linux (164/4642, 2484) | pugs.kwiki.org
Set by obra on 29 April 2005.
Shillo just updated. rev 2500, nice, round number. :) Hmm, so far new breakage in math/int and io_in_for_loops 00:00
Juerd theorb: Doesn't COMMITTER include all necessary bits, such as IMPATIENCE and LAZINESS and HUBRIS and CLUE?
Limbic_Region is running make test on 2500 now
Shillo And all t/data_types
Shillo still has flaky hardware. So this could really be bad compile. 00:01
Limbic_Region had a fair amount more failures than usual 00:03
(3 subtests UNEXPECTEDLY SUCCEEDED), 33 subtests skipped.
theorb Autrijus said when he gave me a metacommitter bit that I should give anybody who demonstrated a willingness to produce patches a committer bit.
Limbic_Region Failed 38/234 test scripts, 83.76% okay. 153/4668 subtests failed, 96.72% okay.
theorb Schwern recently made todo be ignored except in release builds. 00:04
Limbic_Region yeah 00:05
I saw that
Shillo (3 subtests UNEXPECTEDLY SUCCEEDED), 15 subtests skipped. 00:06
Failed 45/234 test scripts, 80.77% okay. 145/3667 subtests failed, 96.05% okay.
Lots of scripts died on me, though.
theorb Hm, that's not good.
Shillo t/syntax/single_quoted_strings...............dubious 00:07
Test returned status 0 (wstat 11, 0xb)
Same for multi_sub,
And 09-dash-uppercase-c
*sigh* This kinda sucks. I need a new CPU before I can tell for sure that these really failed. 00:08
Shillo essentially doesn't trust his build of ghc, in the first place.
theorb Oh, wstat 11... that'd be segfault, I think, which seems unlikely. 00:09
Shillo *** Error: Cannot cast into Hash: VRef <Array>: NonTerm (MkPos "t/subroutines/sub_return_values.t" 209 4 209 29) at t/subroutines/sub_return_values.t line 209, column 1-63
theorb Works for me, shillo. 00:10
Shillo Can I rerun an individual test? 00:11
theorb The easy way is PERL6LIB=./blib6/lib ./pugs t/subroutines/sub_return_values.t
Shillo actually tries rebuild/retest 00:12
Just to see exactly how nondeterministic this is.
Computer shops open in 7 hours... ;)
(2 am here)
Limbic_Region prefers to copy Test.pm to .
but different strokes for different folks
Limbic_Region is trying to verify a bug ATM 00:13
theorb On a saturday morning, Shillo?
Shillo nods.
My favorite time for the component raids. :) 00:14
But I prolly won't, to be honest.
Because I still have this annoying problem of not wanting to move to Athlon 64 just yet.
Limbic_Region ahhh - I did find a bug
Shillo And if I buy a new Athlon XP just now, well... 00:15
Limbic_Region: about sub_return_values.t ?
Limbic_Region no
Shillo Oh.
theorb Oh, you think your CPU is toasted?
Limbic_Region needs to figure out how to fix it though 00:16
his bug that is
Shillo theorb: I -know-. :) 00:17
theorb: Once its temp reaches 60, it goes x-eyed.
(Celsius)
theorb Ah. 00:18
Stick more fans on the case, so it doesn't do that?
Shillo theorb: Won't help much, it's 20-30 degrees above ambient. 00:19
Shillo has the vintage. Palomino, the room-heater.
theorb Ah.
Limbic_Region ok - bugs fixed now I just need to remember how to check something in 00:20
theorb svn ci 00:22
Limbic_Region theorb - I remember that part
I just don't remember how to mark it as being needed
svn save <file?
theorb Unless it's a new file, you don't need to.
Limbic_Region svn save <file>
ahhh
theorb If it is, svn add filename
Limbic_Region gotcha
so it will just magically no what file I modified? 00:23
Shillo But do svn up, make sure it still builds if it pulled some other changes, then svn diff, and review to make sure you don't commit more than you wanted.
Limbic_Region Shillo - I just modified a test
ok - committed 00:25
and I am off
TTFN all - have a great weekend
Shillo sub_return_values still fails the same way. 00:27
Failed 45/234 test scripts, 80.77% okay. 154/4603 subtests failed, 96.65% okay. 00:28
Fewer segfaults, in different places. Well, got my answer. Can't use the machine for testing, really.
jabbot pugs - 2501 - Fixed broken test in t/statements/next.t 00:31
Shillo Later, all. 00:48
theorb_ OK, time for sleep. 02:02
jabbot pugs - 2502 - More type signatures on top-levels. 02:11
autrijus theorbtwo forgot to check in util/munge_haddock.pl :-/ 06:10
jabbot pugs - 2503 - * repair mandel.p6 compilation to parrot 06:21
gaal good morrow. 06:26
autrijus hey gaal. 06:31
gaal hey. i saw you committed some fixes for position - some things were is different places from where i was hacking 06:35
autrijus yeah 06:36
position annotation is now just like context annotation
gaal i couldn't make sense of that unfortunately ;-)
autrijus instead of tackled on the statement structure
and statement is just a simple linked list
so where we had 06:37
Statements [(Exp1, Pos1), (Exp2, Pos2), (Exp3, Pos3)]
we now have
Stmts (Pos Pos1 Exp1) (Stmts (Pos Pos2 Exp2) (Pos Pos3 Exp3)) 06:38
and now all tokens can have their own positions
including Syn and App
and the position is now a range instead of just the beginning pos
see the data definition for Pos
gaal hmm, that also makes possible that idea i had about annotating literal whitespace, no? 06:40
(for making mad Eclipse-like IDEs)
autrijus sure
you can annotate a Noop with Pos 06:41
for example
or add Syn "#"
for comments
gaal looking at data Exp now.. Syn stands for syntax?
autrijus yeah 06:42
please add comments
and check them in:)
gaal application / syntax / context / position / lexical pad / symbol / statements / primitive / value / variable / parentheses / non term? :) 06:43
autrijus yeah :) 06:44
non term == parse error
Parens is currently not used in normal parsing
otherwise all correct
gaal what does Syntax mean then? how is it an *object* in the AST?
autrijus Syn means something that can not be expressed as an App 06:45
like
gaal oh, like a block or a sub?
autrijus Syn "block" [...block body...]
Syn "=" [lhs, rhs]
gaal gotcha :)
autrijus basically, anything that involves non-normal reduction order
i.e. anything call-by-name
or stranger than that
gaal so eg, it can represent a class definition, where traits can come in the signature or in the body? 06:47
autrijus aye
gaal Is App the whole applicaiton, or a current compilation unit? 06:48
(does it matter?)
autrijus App is just a function application. 06:49
:D
gaal ah, so what represents the top level?
autrijus function($invocant: $argument)
there's no toplevel marker 06:50
a toplevel is just an Exp like anything
gaal like not having volume letters. good! :)
autrijus yeah :)
good analogy, that
gaal this is futurish, but how can a parser deal forgivingly with errors? again using the Eclipse example, when it encounters a NonTerm, it often (not always) can skip ahead until it manages to synchronize and keep parsing successfully, only marking eg one or two lines as errors. 06:55
i mean, code is not generallt self-synchronizing like utf-8 is
autrijus I have no idea. 06:58
maybe instead of parsefail 06:59
we always return NonTerm
then the upper parsing unit can carry on.
so e.g.
print((3 +), 5)
will be
App "&print" [] [NonTerm (...failure...), Val (VInt 5))]
that will let us report on multiple syntax errors too 07:00
which may or may not be a win
(I think it's a win)
woot, I think I just accidentally speeded up pugs again.
gaal ...since so often it's the first that causes the later? *nod*
lol
you keep saying that :)
autrijus it's not like I can help it :) 07:01
commit your comments
so I can merge?
gaal one sec
(beh, got into a conflict, sec) 07:04
r2504 07:09
jabbot pugs - 2504 - Add comment explaining Exp data types 07:11
autrijus ok, I'm committing mine 07:12
done as r2508 07:16
I wonder if it actually makes things faster.
gaal if smoke works, we'll find out soon :) 07:17
r2496 - 221 sec on my box
assuming no significant changes in thests, etc... 07:18
perhaps we need a more stable benchmark, if anyone's interested in writing one. 07:19
autrijus I usually just use mandelbrot 07:21
but it's not typical
jabbot pugs - 2505 - * "make optimised" is now OK too
pugs - 2506 - * $_ defaulting now uses the lexical $_
pugs - 2507 - * `for @a { ... }` is now always parsed
pugs - 2508 - * Add strictness to core monad and objec
gaal i guess smoke isn't typical either, because it heavily biases compilation and other startups. 07:22
uh, tests are b0rken :(
oh, no it may be my tree is diryt, sec 07:23
autrijus hey 07:24
skew hi
gaal yeah, false alarm, i had leftover hacks in Lexer.hs
autrijus skew: so "make smoke" dies?
skew okay, so I've got Test::TAP::Model installed, but make smoke can't find it
gaal do you have several perl5s on your system? 07:25
skew and I know absolutely nothing about perl or pugs
autrijus you need to run "Makefile.PL"
using the perl that you installed T::T::M with.
skew I got that far
autrijus then "make smoke"
skew I'll look for spare copies of perl
obra how goes?
mischan,
skew how would I check if some perl interpreter can see T:T:M? 07:26
I typed perl, now I've got blank space
gaal skew: perl -MTest::TAP::Model -e1 07:27
skew If you all haven't guessed, I'm coming from Haskell-land.
gaal you should get back to the shell w/o an error message.
skew okay, that works
does Makefile.PL remember which perl it was run with somehow? 07:28
gaal yes: it generates a Makefile (no .PL) that's keyed to it.
skew everything seems to refer to 5.8.6
which is the perl that just found T:T:M, according to your test
gaal we use perl 5.x in building pugs 07:29
that's all ok
skew progress - now it's Test::TAP::HTMLMatrix. Okay, I know how to fetch stuff from CPAN
gaal that one has a few dependencies - use CPAN's dep feature to help you with that. 07:30
skew, what OS are you on?
Corion Good morning - yesterdays backlog shows what I think is the cause for the scoping error with $_ in $_="Goodbye"; for "Hello" { say }; say; - but I won't be online for the next 8 hours and didn't find out how to remove the problem :) 07:31
skew what's the option for the? force would be useful too. The manpage doesn't mention much
Well, I hit y enough times that it's grabbed everything
hey, looks like it's making smoke
oops, YAML.pm 07:32
is there any way to get all the stuff I need at once?
gaal that also works :) how you get the CPAN shell to d/l things automatically for you depends on whether you're using the regular one or CPANPLUS.
skew I think I've just got the plain one
gaal that's an excellent question, skew :) one hopes we'll get this right in perl6-land. 07:33
skew Will I hit new dependencies halfway through a smoke run?
gaal forum2.org/gaal/pugs/smoke-cygwin.html # looks like a 2.7% speedup from r2496. ignore the errors on line_basic.t 07:34
skew why does smoke rebuild the interpreter each run? 07:35
gaal skew, realistically no. theoretically in a perl5 project you might have, but here once smoke runs it's all perl6
because there's no good way to zero the interpreter state 07:36
skew especially when it seems to be trying to use one frmo /usr/bin
gaal really? that certainly shouldn't happen!
skew well, was spewing out errors about not being able to exec /usr/bin/pugs 07:37
t/data_types/catch_type_cast_mismatch.t
Can't exec /usr/bin/pugs at t/data_types/catch_type_cast_mismatch.t line 1.
gaal how did you invoke this, 'make smoke' ?
skew tyes
gaal ok, i'll investigate this, but let's try something in the meanwhile: 07:38
set HARNESS_PERL to ./pugs 07:39
and PERL6LIB to ext/Test/lib
(in your environment)
then run 'perl util/yaml_harness.pl'
what OS are you using? 07:40
skew linux 07:41
gentoo
it installed 5.8.6, I've been grabbing packages with cpan, though
autrijus I know why
gaal this is very weird, because 'run-smoke.pl' sets the env for you.
jabbot pugs - 2509 - * check for all prereq modules at once,
gaal why? :) 07:42
autrijus gaal: system("perl") 07:43
skew: fixed
r2510
gaal autrijus, but the environment is set before system 07:44
autrijus gaal: it's the wrong perl.
say, ancient perl
gaal oh, you think it should be $^X?
autrijus sure.
gaal gotcha.
ideally, Makefile's $(PERL) 07:45
autrijus gaal: hm, 2.7% isn't that bad.
...and I also did the same for Parser.hs
so maybe another 2.7%.
gaal: it is makefile's $(PERL)
castaway gives skew a "g-cpan.pl"
autrijus because run-smoke is invoked with that
gaal 2.7% isn't bad at all!
castaway (for gentoo) 07:46
gaal especially not after 11%, and 30%, and 50% :)
autrijus I think we should benchmark mandel on this run and on pugs 6.2.1
gaal d/ls ye ancient pugs 07:47
autrijus ancient... of less than one week ago
gaal 'swhat i'm sayin' 07:48
skew what is giving that 2.7?
autrijus skew: r2512 should do it 07:49
skew:
data SIO a = MkSTM !(STM a) | MkIO !(IO a) | MkSIO !a
and
IScalar :: ScalarClass a => !a -> IVar VScalar
the two ! there.
skew do these patchs have descriptions anywhere?
autrijus skew: sadly not if you are using darcs. 07:50
I'll make them carry descriptions when I find tuits
if you are willing to use svn or svk, though
svn.openfoundry.org/pugs
has descriptions
(our main repo)
simply "svn co svn.openfoundry.org/pugs" will do.
I apologize for this inconvenience :-/ 07:51
jabbot pugs - 2510 - * make run-smoke run using the configure
pugs - 2511 - * more unboxing and explicit data struct
pugs - 2512 - * oops.
skew The most recent patch I see is r5186 07:52
gaal skew, by 5186 we'll be solving the halting problem. 07:53
autrijus gaal: heh
r5186 (orig r2604): autrijus | 2005-04-30 15:49:38 +0800
yeah, and the numbers are out of sync.
skew Sat Apr 30 00:46:07 PDT 2005 [email@hidden.address]
* r5184
autrijus gah. I think I need to fix
skew well, as long as I have the patch that fixes smoke so it will actually test my changes 07:55
is that 2.7% from running around unboxing things?
autrijus ok, fixed.
skew: yes.
we've already unboxed most of our data structure
skew that's what I was going to try
autrijus which brought 30% before
the 2.7% is just even more strictness on two places 07:56
skew unpack or use unboxed stuff like INt#?
autrijus btw, sync.pl fixed
not explicit unboxing
just -funbox-strict-fields
skew what does "sync.pl fixed" mean?
maybe I'll find something for SPECIALIZE ... 07:57
autrijus skew: sync.pl is the thing that syncs svn to darcs
skew oh, okay
autrijus so if you pull now 07:58
it should say r2605
try it?
skew no changes left to pull
autrijus weird. what's your last change?
skew "r5186" 07:59
autrijus k sec
oh btw, what is your real name that I can add to AUTHORS? 08:01
skew that's a bit premature :)
autrijus no, you already helped a lot on the SIO monad :)
jabbot pugs - 2513 - * double oops.
autrijus "Brandon Michael Moore" ? 08:02
skew yep
that's me
hey, I'm getting something that syas "tagged snapshot"
autrijus yeah, now pull again... it should say r2514
skew it still says "No remote changes" 08:03
whatever, as long as I've got your patch
autrijus uh, lag?
try again :)
please :)
skew okay, r2514 now
autrijus yay.
skew++
skew is there collected profiling data anywhere? 08:04
gaal mandel benchmark between Pugs 6.2.1 and r2513 show a 37.55% speedup.
autrijus no... you can do "make clean; perl Makefile.PL; make profiled"
to get a profiled pugs
skew then I suppose I need to run it on tests myself to pass profiling flags and do something sensible with the profiling output? 08:05
autrijus sure. you can do this one-liner
(after building a profiled pugs)
skew make smoke would probably be way too slow with the profiled version anyway
autrijus find t -type f | xargs ./pugs -e 'require $_ for @ARGS'
that will run all tests in one run
giving you the profiled output
hm, or not 08:06
skew I'm just trying to figure out where to go to look for performance
you need to pass some +RTS options to get it to actually do any profiling
and it's been too long since I last read the user's guide, so I don't know where the output goes 08:07
gaal (from 2m26.750s real to 2m26.750s)
autrijus find t -type f | grep -v Dialect | grep t$ | xargs ./pugs -e 'require $_ for @ARGS'
right 08:08
find t -type f | grep -v Dialect | grep t$ | xargs ./pugs -e 'require $_ for @ARGS' +RTS -p
skew gaal: that doesn't seem so impressive
autrijus that's it.
gaal: that looks like the same
gaal and grep -v Disabled also
autrijus find t -type f | grep -v D | grep t$ | xargs ./pugs -e 'require $_ for @ARGS' +RTS -p
skew I can think of some interesting things to do with all that data
gaal autrijus, looks like the same as what?
skew (from 2m26.750s real to 2m26.750s) 08:09
gaal oh
sorry :)
to 1m31.637s that is.
i had an, er, paste malfunction. 08:10
that sounds like 30% + 11% + 2.7% plus change. 08:11
jabbot pugs - 2514 - * add skew to authors
skew that's a bit more impressive 08:12
gaal forages
autrijus r2515 has "make pugs.prof"
skew is the plan to eventually port pugs to C by optimizing untill it's just a mass of Ptr and MutableByteArray#?
autrijus need to run "make clean; perl Makefile.PL" for that to work
skew: no, not really :) 08:14
skew: a fast interpreter is nice
but pugs is a compiler too.
skew autrijus: one problem with that xargs line is that it continually overwrites pugs.prof
autrijus so it doesn't quite matter :)
skew: huh? it only runs pugs once
skew does it?
oh, you run it with lots of options 08:15
Okay, I misread it, and thought I should check where the output went before I sat through a smoke run
gaal well, it does potentially run it several times... 08:16
if there are lots of tests.
skew that would be bad
gaal that would be xargs
autrijus fixes welcome. 08:17
pjcj autrijus: you mentioned a while back that you had some thoughts on how to introduce code coverage collection without impacting performance when it is not needed - how is Eval.hs looking in that regard now?
autrijus (not terribly worred now)
pjcj: yes, yesterday's rewrite adds the instrumentation framework
pjcj: now we have the info on what line/column ranges have been executed.
I gather that's part of what you want? 08:18
pjcj excellent, I wondered whether that was the case
autrijus as we can reasonably think that two tokens sharing the same line/column range are in fact the same
so that should be a pretty good start for a coverage tool
pjcj right - I'll have a poke around 08:19
any pointers on where to start?
the other thing that's needed is some info about all the constructs in the program - an AST dump or something 08:20
is anything like that already in place
autrijus sure
pugs -CPugs -e'say 1'
will give you a dump.ast 08:21
that contains two parts, glob and ast
glob is the global pad when the parsing starts
ast is the ast to run from that point on
jabbot pugs - 2515 - * fix testgraph; add "make pugs.prof"
pugs - 2516 - * finish "make pugs.prof" support.
pugs - 2517 - * four unexpected successes
pjcj ok - looks like we are in better shape than last time I looked ;-) 08:22
autrijus++ # I need another reason?
autrijus so look at Eval.hs line 245 08:23
you can see reduction rule for Pos
there you can record that the Pos has been executed
by writing into some magical global var
or using the envDebug stash
or something.
if you need more instrumentation
pjcj what is a Pos -- position? 08:24
autrijus it's a range
AST.hs line 539
if you need more instrumentation
just add them to Pos structure
but I think the range is sufficient for now
the "exp" in Eval line 245
is the expression that is about to be run
"pos" is its source range 08:25
pjcj great - that looks as though I should have enough information floating around
it's a shame there doesn't seem to be any easy way of covering the haskell source code though 08:26
and somewhat surprising 08:27
autrijus no, it's because Pugs is too cutting edge :) 08:28
pjcj I think it's a wonderful project in that respect :-) 08:29
autrijus :) 08:31
I think Pugs has used all major new subsystems GHC 6.4 brings
jabbot pugs - 2518 - * oops, forgot to add pugs.prof line
skew oh, of course you're using STM if you need to fuse it into SIO. Did it solve any interesting concurrency problems? 08:38
autrijus skew: yes. have you read my journal?
pugscode.org, the "Journal" link on the left
skew do I need to rebuild again with this 2518?
autrijus skew: no, just perl Makefile.PL ; make pugs.prof
is enough
(well, that is a rebuild. never mind me) 08:39
it's GHC will skip all the files if you have already built things
skew I just finshed rebuilding under r2517
autrijus using "make profiled"?
skew I'll never get around to actually profiling if I have to keep rebuilding
no, make pugs.prof seemed to build something 08:40
autrijus yeah. give me a sec
ok, I think you'll be fine
skew the pugs I've got now is built with profiling
autrijus under r2518, "perl Makefile.PL ; make pugs.prof" should just take a second
it's just fixing one line in Makefile 08:41
and it will run the profiled test run
I think.
at least it does it here
also, the profiled pugs is not built with -O
skew linking pugs each time through takes my system a while
autrijus as regular "make smoke" pugs is
oh. in that case you can just run this by hand 08:42
find t -type f | grep -v D | grep -v R | xargs ./pugs -e 'require() for @ARGS'
skew too late - it's not that long
autrijus :
:p
skew that saves profiling output? 08:43
autrijus hm it didn't
weird. give me a sec, sorry
skew +RTS -p -RTS
I remember, I wanted to try sticking INLINE on some of the monad methods 08:44
autrijus in that case, if full run is taking too long 08:46
you can always just do
./pugs examples/mandel.p6 +RTS -p
mandelbrot being the canonical benchmarker
skew is that a good benchmark?
autrijus it's the one I use mostly 08:47
so yeah
gaal autrijus, find | .... | ./pugs -e '$*IN is chomped; require for =$*IN' # except that 'is chomped' doesn't work yet 08:48
skew well, I don't know what an async or a pad is
autrijus gaal: anyawy, the full time is 08:49
find t -type f | grep -v D | grep -v R | grep -v pugsrun | xargs ./pugs +RTS -p -RTS -e 'my sub exit {}; require() for \@ARGS'
skew: async{} is forkIO
skew: the pad is the lexical pad.
gaal i just checked the GNU xargs sources - on some systems the limit is as low as 470 - much too short for our tests
autrijus i.e. the visible things from your current let-binding
gaal: fixes to Makefile.PL welcome. 08:50
gaal (however, "some systems" == "SYSV", so maybe this is not a practical limit)
skew and these are mutable structures that are now safe to use because of STM?
autrijus skew: yeah. basically
async { atomic { $x = $y * 10; $y = $x * 10 } };
skew there was no locking before?
autrijus there was no lock(), no. 08:51
skew hash tables I can see
autrijus we can put them in now using TMVar.
skew okay, I've got a pugs.prof now
jabbot pugs - 2519 - * add the RTS options.
autrijus skew: yeah, and it is not terribly useful 08:52
most time are spent on the parser :-/
the tests themselves are all short 08:53
gaal beh, isn't there some global .clone method? not being able to require($_.chomp) for =$*IN because stdin is r/o is annoying
autrijus for =$*IN -> $x is copy { }
I need to run for food. 08:54
be back later
skew: Pugs.Context.findList takes a whooping 6% 09:02
and it's wrong as well because it only handles single inheritance
I'd be very grateful if you (or other people) can refactor it into Data.Graph.Inductive
& 09:03
skew mandel.p6 seems to give more reasonable results. Heavy on Pugs.Prim.op2 like I would expect, then fairly generic sounding stuff 09:12
(that is, I expected mandel to be pretty heavy on math and op2 sounds like the bit that does the math)
gaal does pugs.prof work for you guys? i get a coredump. 09:21
well, a segfault.
well, a gpf.
this is cygwin :)
skew works for me
gaal did you have to install anything special? or just a usual ghc 6.4? 09:22
skew yeah
see if profiling works at all with cygwin
gaal what should i try? (i'm new to haskell)
skew wait, pugs.prof is supposed to do the tests on it's own. Building the profiling version and running it with pugs +RTS -p -RTS worked 09:23
gaal doesn't work for me. 09:24
skew just barely saved my pugs.prof from mandel.p6. make pugs.prof seems to be working
gaal i'll try this on linux then. will take a while.
skew make some little Haskell program, toss in the -prof option when you compile it, then run it with +RTS -p 09:25
see if it gives you a <progname>.prof (good) or a segfault (bad)
enough of a beginner that "make a little Haskell program" is an issue?
gaal no, i can do that. :) 09:26
skew how long have you been learning, and what can you do?
gaal i can add trivial Prims. i still fear monads. 09:27
but recursion and e.g. fold are ok. 09:28
skew From visitors to #haskell I get the impression that some people take a long time to learn monads, I'm wondering if the story is different around pugs with more experienced perl programmers and less first-year students
gaal how long is harder to answer :)
urp? on linux, 09:29
Chasing modules from: src/Main.hs
Could not find module `Data.Map':
skew do you have ghc 6.4?
gaal yes.
skew hmm
is this building pugs?
gaal yes, i wanted to do that in the bg while trying profiling on cygwin. 09:30
oh wait, my tree was dirty.
but no, that doesn't fix it. 09:31
skew is there any file that describes which modules do what? 09:32
gaal not that i know. there's been work in the past few days to haddock things. 09:33
skew well, the pugs work just claimed a second screen for browsing the tree then
gaal be 09:35
*brb
skew is there any pattern about which functions have type signatures? 09:36
gaal i don't think so -- i see no reason not to include them 09:41
except where the author thought they're too trivial.
skew okay. doBoolIO took about 15% of the profiled run of the tests and didn't have a signature
gaal but i don't have much to say about style in a language i don't know :)
does having a signature have any semantic difference? 09:42
i thought it was mostly a documentation hint.
skew sticking on a signature with the same type that would be inferred doesn't
but some things the type checker is not smart enough to figure out without a hit (that pesky halting problem, you know) 09:43
gaal yeah
but has this anytrhing to do with profiling?
skew also, you can give it a more restrive type than would be otherwise inferred 09:44
gaal i'd think signatures had only compile time effect?
yes, i see that.
aha! on my cygwin, any profiling fails. 09:45
skew relevant to profiling, if a function ends up polymorphic with a class constraint it is Bad for performance
gaal "my cygwin" is really ghg-mingw
skew rule one of optimization is to find and squish laziness and polymorphism 09:47
gaal skew, so could we benefit from narrowing doBoolIO to (possibly several) more tightly typed functions?
skew I think it already gets a specific type
but that findList autrijus mentioned is polymorphic in Eq, and only used at one type!
do you know how the typeclass stuff works at runtime? 09:48
for that matter, does autrijus?
gaal sorry, no.
beh, on linux 'make optimized' works but regular 'make' breaks. 09:49
skew if you have a type like (Eq a) => a -> [a] -> Bool, then internally it takes an extra argument, a record full of the methods for an Eq instance, and uses functions from that record every time it sees a method from Eq
gaal ...and the prof target is broken as well
skew at every call, the compiler passes along the right record for whatever type a is is that that call uses 09:50
gaal sounds like virtual method calls :)
eg c++ vtable hack
ok
skew pretty close, except usually (C++ template magic aside) the language can't go out and automatically define a new subclass for your problem 09:51
but anyway, if the compiler has a specific type, then it can call those methods directly, and more than that it can inline them and apply all sorts of crazy GHC optimizations to them
actually, typeclasses are more like delegation and the coding style in an untyped language 09:52
they just formalize "anything that has method X" 09:53
gaal okay, i made ghci and did :t doBoolIO, and got something extremely hairy
skew you did? 09:54
I expected something like a -> (a -> Eval Bool) -> IO Bool?
could you past it in for me?
pasteling "gaal" at 192.115.25.249 pasted "something extremely hairy" (10 lines, 370B) at sial.org/pbot/9782
skew Okay, it's not as bad as 10 lines sounds 09:55
gaal one of the things that haskell and perl have in common is that 10 lines can be a lot of hariness. :)
skew I wonder how many instances of Value there are running around
gaal i wish i knew more haskell to be able to answer that. 09:56
skew I'd put my money on Haskell for sheer mind bending. I should look over the IOHCC winners and see if there was any sophisticated entries, or it was mostly just ugly syntax
autrijus yo. 09:58
skew hi
autrijus gaal, skew: please feel free to commit more type signatures.
skew: committing via darcs is currently broken, but you can send patches, or (again) if you are willing to suffer svn, I can give you committer right there
skew: and no, I had no idea that typeclasses are vtabled at compile time. 09:59
and yes, there are lots and lots of Value instances, so we can't do without that
skew I think I've got a copy around. I can figure out how to use it
that's not the only way to solve the problem
gaal skew, the classic perl one-liner nuke isn't even very obfuscated: 10:00
perl -wle '(1 x $_) !~ /^(11+)\1+$/ && print while ++ $_'
skew hmm, mabye it is
autrijus I'm fixing Context.hs to say Type instead of Eq a => a
skew is most of Pugs polymorphic ic Value?
autrijus: hold off on that for a bit
I'm trying something a bit different
autrijus skew: yes, most of the primitives are based on fromVal 10:03
which pulls out a polymorphic Value from a Val
pasteling "gaal" at 192.115.25.249 pasted "xargs elimination" (16 lines, 781B) at sial.org/pbot/9783
autrijus skew: too late, I've already checked in Context.hs 10:04
skew there is a SPECIALIZE pragma which should be as good for some of the stuff in Context
gaal autrijus, or someone, mind trying this patch? i do't have a system to test it on myself.
autrijus skew: the real problem is Context.hs wants a rewrite using Graph.
skew: Tree is wrong -- Perl 6 need multiple inheritance
gaal (autrijus, plain "make" is broken on linux)
autrijus gaal: broken how 10:05
gaal LD_RUN_PATH="" cc -shared -L/usr/local/lib Pugs.o -o /home/roo/src/pugs/blib/arch/auto/Perl6/Pugs/Pugs.so
cc: Pugs.o: No such file or directory
cc: no input files
make: *** [/home/roo/src/pugs/blib/arch/auto/Perl6/Pugs/Pugs.so] Error 1
but make optimized does work.
skew wait, does make install install under the name perl? 10:06
gaal skew, it should not!
as /usr/bin/pugs 10:07
autrijus gaal: it needs to say =$*IN
skew oh, it was a message from ghc-pkg about plugins
autrijus gaal: checked in
gaal er, yes, thanks :)
skew I thought for a moment it was pugs complaining 10:08
autrijus gaal: that is very wrong. your tree is clean? is this perl 5.8.1+?
you have a sane makemaker?
skew make seems to be working just fine here
gaal 5.8.4, i'll make reallyreallyclean again.
autrijus ditto here
gaal hey, how come rm -rf * ; svn up does something? :) does * not glob .svn? 10:09
hmm, the real clean make helped. 10:10
skew SPECIALIZE tells GHC to generate an optimized version for the types you specify, and use that wherever it can 10:11
jabbot pugs - 2520 - * add some signatures to Context.hs upon
pugs - 2521 - * remove xargs from "make pugs.prof" fro
skew do the signatures help performance any? 10:12
autrijus dunno, haven't tested. I'm working on $work now 10:13
skew: I don't think we want SPECIALIZE for things that are monommorphic anyway. 10:14
it can help for the functions that has (Value a) polymorphism.
skew if there are few enough instances that it's workable to specialize everything for all of them 10:15
actually, I don't think it helps at all if you are using fromVal all over the place
because then it doesn't know at compile time what type it's getting so it has to use the dictionary packed up in the Val
autrijus right. 10:16
skew Hopefully jhc makes all these worries obsolete
do you ever use code at some particular known Value insatnce? 10:17
that's the only way SPECIALIZE would help
I think
autrijus well, yes I do 10:19
all the stuff in Prim.hs
skew Hey, I think UNPACK is different from ! and -funbox-strict-fields
autrijus how so?
GHC manual says they are the same
skew are they?
autrijus yeah.
"See also the -funbox-strict-fields flag, which essentially has the effect of adding {-# UNPACK #-} to every strict constructor field."
in 7.9, Pragmas
skew okay 10:20
so there's not another howevermany percent sitting there :(
autrijus nope :)
personally I'm already quite happy with how fast Pugs is.
skew I guess unboxed values would have to be stuck right into the constructor
autrijus i.e. works for my production needs already
skew that's nice.
How much slower than perl5 is it? 10:21
autrijus the operator dispatch was ~100 times slower
skew I heard two orders of magnitude more instructions per op, but that was a while ago
autrijus since we only improved to 5x speed or so
I'd say still much slower.
skew how does that op stuff work? 10:22
autrijus and I'm happy really, because to get any faster
I'll use the parrot and/or ghc compiler backend
that is
pugscc --runparrot examples/mandel.p6
which is already _faster_ than peerl 5.
all the op stuff is in Prim.hs
and the dispatching happens on Eval.hs 10:23
skew I hope the string matching only happens once
I mean op2 and such
autrijus yeah.
I have no idea
skew make pugs.prof now fails in linking for me 10:25
why should findList be happening a lot anyway? 10:27
autrijus because it is used to do subtype matching 10:29
via isaType
i.e. a Int is a Rat
a Rat is a Num
and so on and so forth
many op dispatch needs to perform typecasing to determine which MMD instance to dispatch into 10:30
see the large tree at end of Context.hs
skew does mandel.p6 have to be written that strangely? I thought I could just find a line number somewhere and tweak that to make it only draw half the thing 10:33
where did you get the C anyway? The IOCCC? 10:35
autrijus skew: line 36 10:36
make it > 0
I committed the change
skew I changed y=30 to y=10
autrijus skew: I have no idea. it's from parrot tree
skew it seems to be drawing the bottom third now
you guys know those are examples of bad C style right? Even if it is more readable than the perl ... 10:37
though some of the blame definitely goes to the default color my editor gives those $ variables
gaal skew: "This is badly hacked to get Parrot compiler going" 10:38
skew yeah, that's what I was asking about
autrijus right, except I think we can unhack it now
because our parrot compiler is much better now 10:39
but it will probably need to remain in the ugly C-based style
just to get things fairly benchmarked
(against pasm, p5, C, etc)
woot, I've sent Pugs's Haskell Community Report.
# autrijus.org/hca2005may-pugs.txt 10:41
jabbot pugs - 2522 - * only draw mandelbrot once
autrijus bbl. :) & 10:44
skew single character $ variables are pretty unreadable in a light blue color where my eyes blur everything into an S-shaped blob with something sticking off to the right
what did that patch change? 10:47
gaal skew, what editor /font/colorscheme do you use?
skew vim, with one of the stock colorschemes (I've forgotten which) modified not to show some important Haskell construct in a similarly unreadable color 10:48
gaal fwiw, i like the koehler colorscheme in vim 10:49
skew though the unreadability might have been an artifact of coding in daylight hours
gaal not used to that, eh? :) 10:50
skew no, I was doing that way back when I made the colorscheme
evening, with green instead of bright yellow 10:51
gaal and what font? i use MS Courier New, though i am looking to change it. (But I require unicode, so it isn't easy to find something) 10:53
need to try dejavu.sourceforge.net/wiki/index.php/Main_Page 10:54
skew whatever my terminal is using
gaal if you say your $s blur a lot, maybe a different font can help.
skew it's fine with something like koehler 10:55
some colors are just harder to resolve
I think it's time to stop letting new patches interfere with my work 10:57
I'm going the make my copy of pugs run my copy of mandel.p6 blazingly fast, and you want to make changes you can optimize them yourself :) 11:00
I wonder how much I would disturb the timing if I learned Ruby on Rails at the same time 11:04
I can always just chat on IRC while stuff builds and times, I guess :)
I have a friend with many neat ideas for things to write, but he's too caught up in coding them all up to discover for himself that there are better languages than C# and ASP.NET out there 11:06
I figure if I can reimplement a fair chunk of one of his web things with an order of magnitude less code that might get his attention 11:07
gaal avoid html as much as possible if you want to manage that :) 11:08
skew I think not counting HTML is pretty fair 11:09
gaal hey, avoid *http* as long as possible :)
castaway skew, heh, poor guy 11:12
skew what compiler backends are there? 11:13
never mind, I found the pugscc script 11:14
maybe interning strings could help 11:16
== on [Char] can be a bit slow 11:17
how much value is there in making pugs fast? 11:24
gaal values fast 11:25
as long as it doesn't obfuscate the code so much it's too hard to maintain
skew okay
gaal hey, i'm just another perl6 hacker though :) 11:26
skew actually, a lot of the complexity I'm contemplating would be hidden from most of the code 11:28
wow, runparrot is a *lot* faster than the interpreter
more than 60x faster 11:30
gaal skew: www.mail-archive.com/perl6-compiler...00417.html but i guess you figured all that out by now. 11:32
it isn't so surprising that the compiled versions run faster than interpreted code, is it? 11:33
skew of course, the C cde is 100x faster yet
pjcj is that with an optimised parrot build? 11:35
gaal and hand tuned assembly might be faster still, but at some level it stops mattering :) 11:36
pjcj IIRC the default is unoptimised
skew wait, I was looking at real time. The C is 300x faster 11:38
I doubt there's more than about 10x to gain with clever assembly
pjcj in the end, real time is all that matters though, right? 11:39
skew isn't it wonderful programs can be 1000x suboptimal and still fast enough for all practical purposes 11:40
well, the real time for C is dominated by output
gaal it's especially possible if it takes 1000x less time to code them :)
skew I just realized I've been timing the unoptomzied pugs 11:43
I wonder if anyone has tried to build pugs with jhc? 11:44
gaal does it have all the newest ghc extensions? 11:45
skew heh
no
It's something somebody wrote themselves to learn Haskell and try out some ideas 11:46
gaal sounds familiar :)
skew it has an interesting implementation of type classes that might help pugs a lot 11:47
it doesn't do much optimization yet, though
actually, the extensions might not be that much work 11:48
STM is just an issue of the runtime and you can take that from GHC
as for the fancy types, the internals are already geared up for dependent types 11:49
It's mostly an issue of parsing, typechecking, not having any optimizations, and already taking a gig of ram to compile hello world
gaal cheerfully has a gig of ram on his main box :) 11:50
skew implementing separate compilation should help bring that down
also it's a bit odd in that fancier optimizations improve compiler performance
gaal "separate compilation" == compiling separate units then linking them?
skew because a simple program is sent to the bits that are really slow 11:51
though I think John just found a paper about how to do it in linear time instead of exponential or so
gaal: was
yes
theorbtwo john == the jhc guy?
skew basically, all the function pointers and typeclasses and stuff are inlined together into case statements 11:52
yes
JohnMeacham, sometimes on #haskell
gaal brb
skew something like that anyway
If I understood him right, higher order functions from the library need to be recompiled against the code in your program that calls them 11:53
but apparently there is some way to save lots of info to make this fast
after all, the linker still has to copy over the library definitions under GHC
anyway, typeclass stuff compiles to functions that take a representation of the type they are being applied to, and pattern match on that 11:56
so with something like unpacking Val it can theoretically do a single case, then go to code for dealing with that particular type
gaal skew, can you help with a parsec Q? 12:05
skew sure
gaal inside a rule, i'm trying to get and optinal next bit of data. 12:06
skew that should be easy
gaal what does this do when it fails? 12:07
lineNum <- try $ many1 digit
skew fails and tries the alternative
except writing it like that it seems you're in the middle of a do block
gaal lineNum <- try $ do { many1 digit } <|> return () 12:08
skew try is a mark for backtracking
gaal ?
skew oh, that's just precedence error
try (many1 digit) <|> return () is the right idea
except use Maybe instead of trying to return either a number or unit 12:09
gaal and then lineNum will be null when the lookahead failed?
skew no, then ghc will tell you that numbers and () are a very different type
gaal oh: how?
skew well, there should be a function called option somewhere that already does all that
gaal can't blame it :)
skew do you understand the LL parsing stuff? 12:10
gaal nope, no theoretical background :(
skew basically, at any choice Parsec goes with the first parser that is willing to take the next character of input
gaal yes, which is why i mustn't consume things in the first parser if i want to try the second 12:11
skew so if the second character gives a parse error, that's too bad because it's already chosen that alternative
try changes it so it will try the next if there's ever a parse error
really it's more like a parser uses lazy evaluation to return DontMatch or StartedParsing <thunk> right away 12:12
then the inside evaluates to either Fail <reason> or Success <value> (in some Either lookalike type) 12:15
gaal i unfortunately only have a wisp of your gist :) 12:16
skew 1) by default, once a parser takes a character you're committed 2) try changes that so if it ever fails you will go to the next alternative 3) the implementation is really elegant and uses neat Haskell-stuff 12:17
gaal :) 12:18
why does this parsefail on "="?
lineNum = option (-1) (do { n <- many1 digits ; return read n })
skew oh, and 4) There's probably already a combinator to do whatever you want in the library
I don't think "=" is a digit
oh, ghc doesn't like it. Are you in a do block? 12:19
gaal ...and is there a way to use that... combinator to evaluate to either Nothing or Just someNumber?
skew it's var <- code to run monad stuff, or let var = expr to name something
gaal let's see...
skew sure, just use Nothing as the default and wrap your answer in Just. 12:20
or, make your own option
gaal i am in a do, and this fails: lineNum <- option Nothing (do { n <- many1 digit ; return $ Just (read n) }) 12:21
i can't return a Maybe? 12:22
skew that should work 12:23
gaal ah, it was getting confued because i hadn't changed my next line to use a Maybe
skew I got my two monitors from people moving away for about $150 total, and got the third from the CS department's abandoned stuff pile 12:25
gaal are they all connected to one computer? 12:26
skew yes
though I was explaining to tuomov on #haskell what great use I'm making of ion
theorbtwo wants one of those.
(A CS department's abandoned stuff pile, that is.) 12:27
gaal ion kinda disagrees with the many screens philosophy :)
skew how so?
I thought the idea was to efficiently use all your space
gaal ion is the maximized window wm, no?
skew I go for tiling 12:28
to me it's about efficient use of screen real-estate and making it easy to get around with keys
there's nothing wrong with having more real-estate to use efficiently 12:29
gaal my eyes, not being what they used to, really appreciate maximized xterms
yay! i another test passes.
skew++; # help with parsec
okay, this must be utterly trivial, but how do i implement followedBy? 12:39
ie, i want to assert that the next character isSpace 12:40
but not consume it.
it it just "try isSpace"?
skew no, that will consume it
gaal really? i thought the whole idea with try is that it doesn't consume? 12:41
man, this is so much easier with regexps :)
skew no, the idea with try is that it can still backtrack and try another alternative even if it does fail
gaal hmmm. 12:42
skew how about lookahead? 12:54
gaal not mentioned in www.cs.uu.nl/~daan/download/parsec/parsec.html 12:55
maybe l'm looking at an old doc?
skew it's documented with the rest of the libraries now
gaal oh 12:56
skew there are no descriptions for many functions, but they are listed
gaal not mentiond here either: haskell.org/ghc/docs/latest/html/li...dex-L.html 12:57
nothingmuch ingy: ping 13:03
gaal: is thirst in hebrew a male noun? 13:07
the page has 'zama kitsoni'
and i don't know if it should be kitsonit
gaal yes.
nothingmuch i'm 90% sure
yes male or yes female?
gaal yes SIR!
yes masculine.
nothingmuch okay, so when you say 'zama kizonit' it's because you really mean 'thushat zama kizonit'? 13:08
skew Text.ParserCombinators.Parsec.Combinator.lookAhead
gaal nm: i guess.
nothingmuch grazie señor
gaal thanks, skew! 13:09
i can't believe how long it's taking me to do /("?)(.*)\1/ :) 13:10
xerox Is it missing a '/' ? 13:11
gaal no :)
xerox Oh, what does it mean?
gaal it matches a sometihng optionally surrounded by quotes 13:12
and the interesting something goes to $2.
skew how about between (char '"') (char '"') (many anyToken) <|> (many anyToken)?
xerox Oh yes! :_)
gaal :)
complications: there may be trailing whitespace, which i want to ignore; 13:13
skew porting regular expressions is going to give you very strange parsers
gaal and i certainly don't wan tto consume the newline that may be there.
skew: i'm doing #line, which is one of three things:
skew oh, anyChar should be what you want in the middle then 13:14
gaal #line NUM -> setPosion pos { sourceLine = NUM } (basically)
#line NUM "FILE NAME" -> setPosition pos{ souceLine = NUM, sourceFile = FILE NAME} 13:15
skew if this is using Token, then just stick lexeme around anything you want to each trailing whitespace
gaal and this has to have a VERY SOFT failure mode, because if this fails, eg
#line oh no that's not what i meant
or
#line 123fooled you!
then this falls back on a comment :)
skew you mean you want it to be very strict? 13:16
gaal let say,
using Java terminology, :)
skew sounds like you want it to be very picky about what it recognizes, and use try to fall over the comment 13:17
gaal i want to surround the whole rule with a try { } catch { treat this line as comment }
skew that would be <|>
gaal but that means i really muistn't consume things until the last moment.
skew I think maybe you have used regular expressions when you should have used a real parser too many times
gaal don't you mean Real Parser? :-p 13:18
skew why were you even trying to parse something that optionally had quotes around it?
gaal: nah, that's probably writing the state transition table by hand
gaal well, we're all learning here :) how would you suggest this should be done? 13:19
skew do { string "#line"; whitespace; many1 digit; option (return 'q') (between (char '"') (char '"') (many anyChar)); whitespace } 13:21
see if that parses what you want
do you know if you are using Token?
Where does the parser live?
gaal wait, first of all the quotes are optional
skew are spaces allowed in the filename then? 13:22
gaal and whitespace, doesn't that include \n?
spaces are allowed if you use quotes :)
skew maybe I mean spaces
gaal the code is in Pugs/Lexer.hs
oneLineComment 13:23
which i think is on the style of oneEyedPirate
(arrr)
hey, aren't matches greedy? 13:25
oh, scratch that, you were anho 13:26
anchoring against quotes.
skew I don't know if that works
gaal the versiony you have in your source tree didn't go very far 13:27
i've hacked it a bit since.
skew oh, I'm thinking of the UU combinators, which do have a non-greedy many
gaal well, worst case, we can try against your rule 13:28
and if it failes, try against a rule with a single unquoted token
and if it fails, try against #line NUM which deals with a comment fallback internalky. 13:29
skew I would just replace anyChar with something a bit more specific
if escapes are allowed in the filename you could parse it as a string
gaal i guess they have to be 13:30
otherwise you won't be able to say
skew don't try to do everything in a single expression
gaal #line 123 "file\nwith\nan EXCEPTIONALLY sneaky name"
key, it's not me 13:31
s/k/h/
it's the semantics defined by p5
skew if it's getting too complicated, start a new definition for quotedFilename, or whatever
gaal where, of course, this is implemented in c
well, that's where i get confused about not consuming input when i fail.
skew test little bits of the parser separately 13:32
write more complete parsers that refer to the little ones
the magic word is compositionality
stevan is openfoundry.org down? 13:36
theorbtwo I just checked in the missing util/munge_haddock.pl a moment ago. 13:37
(I really need to be more careful...)
stevan hm, I cant seem to connect to it
oh well
gaal i've had problems this morning, stevan, but it works now
stevan hmm, this happened once before, it was an DNS issue,... which would explain why it works for you guys and not for me 13:40
oh well I was going to just mess around with test.pm anyway
anyone have any opinions on the :todo<bug> :todo<feature> idea?
bugs would fail unless PUGS_RELEASE was on 13:41
jabbot pugs - 2523 - Whoops, add missing munge_haddock.pl!
stevan and features would just show TODO (like regular) 13:42
(at least this is how my caffine deprived brain is viewing it right now)
gaal stevan, i never really understood the aim of release-time TODO and what the solutions were trying to achieve. maybe a rationale-type doc can help? 13:43
stevan gaal: the whole idea is quite simple 13:44
to avoid bug reports for problems we know about :)
its bad QA practice (which is why Schwern was so offended by it), but it is practical given the release schedule 13:45
and the fact that even if we explained that, people done read, so there will still be a glut of bug reports
gaal okay; what's wrong with making sure the candidate has all fixable bugs accounted for and then 13:46
make test | force_todoer > force_todo_list
?
stevan a force_todoer script would be just about impossible to write 13:47
well actually maybe it wouldnt
gaal why? the release manager looks at the last smoke, decides it's as good as it's going to be, and registers those tests as force_todo 13:48
(modulo parsefauils, of course)
stevan so you are basically saying, when we decide to release 13:49
we gather all the failures (through some means)
and then just force_todo them
gaal presumably those failures are exactly those we "know about" and dont' want bug reports on, no?>
stevan yes
I mean this is pretty much what happens now 13:50
just manually
gaal there's a minor issue of platform dependent bugs
the yaml output contains all failing tests in easy-to-use form
stevan hmmm 13:51
gaal so what are the shortcomings of this approach?
stevan nothing
it is an automation of what we are doing manually right now
gaal but you mentioned todo<bug> and todo<feature> 13:52
what are those?
stevan there are really two levels of TODO
gaal known bug and futures?
stevan the stuff in t/oo is :todo<feature> since they are no where near done 13:53
gaal ("futures" eq "features to be implemented in the future" :)
stevan yes
:todo<bug> is someting which should be implemented, but is not (for whatever reason)
like 'for @list &sub'
gaal and todo<bug>s are not bugs, they're you-know-what :)
all right; this obviously needs human interaction. 13:54
stevan yes
yeah I am not sure <bug> is a good name
gaal okay, so i do see a possible problem
we want to have the test results not say things are okay, but rather "it's okay, i knwo this doesn't work." 13:55
but that's just a display issue in the test harness
and the user doesn't care if it's because of a known bug or a missing feauture
have we forked off test::harness already? or is ours a pristine copy of something on cpan? 13:57
stevan it is still a pristine copy
gaal well, maybe it's time to change that. because the problem isn't in Test.pm, i think, but rather in the harness. 13:58
stevan yeah I think you are right
the test harness should be able to report number of TODOs like it does number of SKIPS
gaal we have our copy anyway, might as well moose it up.
stevan at a minimum
well is this a TAP issue? or a Test::Harness issue? 13:59
gaal alternatively, we *could* cheat and overwrite the make test target to run the yaml harness ;-)
but not really, because we don't want to bundle yaml.pm too
stevan :)
gaal TAP does say something is TODO
nothingmuch notices buzzwords and decides to join i 14:00
n
gaal but it doesn't differentiate <bug> vs. <feature>
stevan looks at the TAP doc
gaal really must grab a bite
scook0 finally manages to build all the haddocked documentation -- thanks, theorbtwo!
stevan gaal: I wonder if in TAP we can just add support for reporting TODOs
gaal we can do it as an undocumented extension
stevan and then in the yaml harness we can get more detailed/pugs specific 14:01
gaal TAP doesn't *report* anything
but it does say a test is TODO
stevan gaal: true
gaal it's the harness that decides to interpret not ok # TODO as "ok"
stevan ok, so we patch Test::Harness
gaal which is why
exactly
stevan I guess what I am saying though is this
reportnig TODOs is useful for others 14:02
the different type of TODOs are only useful to us
so if Test::Harness reports TODOs
that will be good enough for us
and then the YAML harness can differentiate
since that is what we all use for development anyway
gaal okay, so if i understand correctly, this is just a small display change in Test::Harness. 14:03
stevan yes
which I am sure the perl-qa folk would like and apply
nothingmuch see actual_ok 14:04
in TAP::Model
gaal and this is orthogonal to our force_todoings.
stevan gaal: yes
gaal actual_ok == false for not_ok # TODO, right nm?
nothingmuch yep
stevan Schwern's issue with our force_todo policy is that no one knows there are errors 14:05
gaal that's entirely our problem
nothingmuch actual_uk is really just the 'ok' or 'not ok' with no meaning
gaal by which i mean
not a T:H issue
stevan IMO if we show our TODOs then we are still being honest, but avoiding uneeded bug reports
gaal okay, let's hack this then. 14:06
stevan gaal: ok I am lookin at T::H::run_all_tests
gaal stevan, mind if i disappear for a few miunutes? i really have to make myself a sandwich. 14:07
stevan gaal: go ahead, i will mess around with this
gaal beb & 14:08
s/e/r/
rehi 14:18
stevan gaal: 14:21
this is pretty simple :)
I think i have it already
gaal good :)
stevan basically it now outputs like this
t/builtins/arrays/end....ok
1/16 skipped: just skippin, 2/16 TODO test(s), 1/16 unexpectedly succeeded
All tests successful (1 subtest UNEXPECTEDLY SUCCEEDED) (2 subtests TODO), 1 subtest skipped.
I added a skip and an unexpected succeedd 14:22
gaal woot
stevan just to see how itall looked
but it basically just reports TODOs much like it reports SKIPS
:)
autrijus yo. 14:24
shapr y0
stevan hey
autrijus stevan: you can start doing changelogs :)
stevan autrijus: sure thing boss :)
autrijus release eta less than 30 hours from now
gaal yay! release soon? 14:25
autrijus no other major changes planned
stevan: thanks!
gaal: sure... it's weekend again
stevan autrijus: I am not having luck connecting to openfoundry though
gaal stevan: i'm your backup man, yo
stevan gaal: :)
autrijus stevan: hm worksforme
stevan: is it dns?
stevan autrijus: I think so
autrijus try switch relocate to openfoundry.dyndns.org
stevan it will probably work soon, gaal has similar issues earlier
theorbtwo autrijus: sorry I've been a bit sloppy on my checkins of late -- I shall endevor to do better. 14:26
stevan ok
autrijus svn switch --relocate
theorbtwo: that's just fine, I do that all the time
:)
theorbtwo: think you can make "make haddock" work on 6.2.2?
(pugs 6.2.2 that is)
stevan openfoundry.dyndns.org works :) thanks 14:27
autrijus np
gaal oh crap, stevan i just noticed what day it is tomorrow and i may not be available as much as i'd like. 14:28
theorbtwo Ooof, makefile hacking. 14:29
stevan gaal: you going "up the river" as they say?
Odin- May day?
theorbtwo find src -name \*.hs|perl -pe 's/hs$/hpp/'|xargs make && haddock -h -o docs/haddock/ `find src -name \*.hpp` is what it should do. 14:30
stevan autrijus: we are hacking Test::Harness to show TODOs
t/builtins/arrays/end....ok
1/14 TODO test
autrijus stevan: right, I saw that. wunderbar
stevan as an alternative to making everything die
gaal stevan, heh, no, though that might still happen :/ 14:31
stevan I am going to submit the patch to the perl-qa folks
gaal: passover related then?
gaal nah. 14:32
theorbtwo Doesn't passover end today and not tomorrow in Israel? 14:38
gaal isn't the expert
nothingmuch it's ending! woohoo!
i just finished the pitas today
gaal tomorrow is a work day, but some people have some more celebrations
nothingmuch but i have some regular bread
nothingmuch survived yet another year
shapr yay 14:39
theorbtwo doesn't keep pasah dicha any more.
gaal oh, between isn't greedy? oooof. 14:41
theorbtwo "between isn't greedy"?
autrijus gaal: why, it is 14:42
gaal theorbtwo: parsec's between
autrijus open; x <- p; close
gaal oh! good. :)
autrijus it looks greedy to me
theorbtwo Ah.
autrijus i.e. if "p" here is "many1 anyChar" 14:43
theorbtwo BTW: I only know because I checked wikipedia.
autrijus then between will always fail
so it's not only greedy, it's committed
(i.e. not back tracking)
gaal er, does that mean i can't use it with <|> ? 14:44
autrijus hm?
sure can
as long as your "p" does not always match "close"
stevan cool, petdance is going to take the Test::Harness patch :) 14:52
gaal yay!
stevan I just sent him a patch for Test::Memory::Cycle this morning too :)
shapr Oh hey, I advertised pugs a bit at the end of a monads overview I posted on the pragmatic programmers yahoo mailing list - article.gmane.org/gmane.comp.progra...-year/3286 14:57
autrijus nice. shapr++
instead of "container" I'd say "action" 14:58
but I realise monad has this dual nature :) 14:59
shapr Yes, I know it's a 'lie for teaching' but I don't know how to describe it as both without overwhelming the listener.
autrijus "how to turn a value into an action", and "how to combine two actions together" 15:00
shapr Hm, I'll try that on the next few people who ask for a monads tutorial on #haskell.
The containers comparison seems easier for those used to objects.
autrijus yeah but IO is a more interesting monad than State. :) 15:01
I've run into people who thought State is all there is about monads
shapr Yeah, that's surprisingly common.
autrijus and came away thinking that trivial things in other languages takes lots of time in haskell.
jabbot pugs - 2524 - * same-named variables declared in inner
autrijus for me, (ReaderT IO a) is the canonical example when I talk about monad and transformers. 15:02
it has a certain elegance. :)
shapr I try to start with the Maybe monad because I think everyone has wished to abstract a huge stack of if-then-else.
autrijus er, no, people usually just use block exit.
aka return()
shapr Hm, I hadn't thought of that. 15:03
autrijus because most languages actually have ContT built in (but without callCC exposed)
shapr Maybe I'm pickier with my control flow.
In that case, I really need to rethink my monads tutorial.
autrijus well, there will be people who think more like you, I'm sure. :) 15:04
we need all sorts of tutorials.
shapr I'm not convinced, most people seem instantly confused when I describe my thinking.
autrijus strangely it hadn't happened to me :) 15:05
shapr You're not most people ;-)
autrijus ...but all people are not most people...
shapr excellent point
I truly hope I can end up hanging out with you for a few days in person at some point. 15:06
I think I will learn a lot.
autrijus ICFP, likely.
shapr I'm not coming this year, sadly.
autrijus I'll submit to Haskell workshop.
oh. then Euro OSCON?
shapr Oh good idea!
shapr googles to check dates
autrijus Euro OSCON could be expensive though, so make sure you get a talk accepted :)
shapr ICFP is also expensive, I might go if it were cheaper. 15:07
I had *lots* of fun at ICFP2003.
autrijus brb 15:08
shapr Various real life difficulties have encouraged me to cut back on my output and take some time to search for balance, center, whatever you call it. 15:09
Will you be giving a Perl talk at EOSCON?
autrijus yeah. 15:10
likely a 3hr tutorial about Pugs
at least.
shapr Spiffy. I could give a talk on QuickCheck and ports.
stevan autrijus: do you mind if I remove the PUG_RELEASE stuff? since TODOs are now reported?
gaal why is the inferred type of fileName () ?
fileName <- choice
[ do { (between (char '"') (char '"') (many anyChar)) ; trailingWhiteSpace }
, do { many1 anyChar ; trailingWhiteSpace }
]
autrijus tentatively named "Perl 6 Bootstrap Camp"
stevan: sure, go ahead
shapr gaal: you haven't returned fileName 15:11
gaal ah!
silly me :)
jabbot pugs - 2525 - adding some more tests for end.t 15:12
webmind wishes perl6 was here
shapr It is here!
webmind not at the speed I like for production 15:14
webmind could code his product in perl6 15:15
but it won't be fast
stevan webmind: you could,.. at rate we are going, it will be fast enough
webmind what rate ?
shapr Yeah, did you see autrijus journal from yesterday?
stevan rate of development
webmind yeah
how fast is it ?
autrijus webmind: actually, if you have your application coded
in pugs interpreter mode 15:16
stevan webmind: this week alone has seen 50%+ speed increase
webmind nice
autrijus chances are that we will help tailoring Parrot backend for you
to support the constructs you use
then it is guaranteed to be faster than perl 5.
webmind hm
how do you mean?
stevan there is a feature for tests policy here 15:17
autrijus webmind: pugscc --parrot
stevan you want a feature,.. write a test
autrijus it generates parrot assembly code
that is very very fast.
webmind hmmm
autrijus almost guaranteed to be faster than perl5.
webmind ok
autrijus so you can use the interpreter mode to test corectness
webmind so how would that work with the perl6 modules I'm using now ?
er
perl5 modules 15:18
autrijus you can load them via ponie
talk to Nicholas on irc.perl.org #parrot for how
or, to test them with interpreter, you can use eval_perl5() support in pugs
(which needs some work, so tests welcome)
webmind hm
gaal webmind: or port the interesting ones to p6 :) 15:19
autrijus that too.
webmind gaal, that's a bit ovwer my head
over
:) 15:20
shapr Might not be, you never know until you try.
stevan webmind: what is your application?
gaal webmind, it's a great way to learn perl6
because they already run
webmind stevan, making a decent photo album
flexible, yet not too loaded 15:21
gaal, uhuh
I know it would
be
gaal well, if you have to do lots of fancy image manipulation
then the task sure ain't trivial
autrijus you can always call out to C libraries
gaal and you may need to embed c
autrijus actually I'm thinking about linking in libtap :)
(probably not)
webmind not really.. it's more having a good way to order your pictures, and having flexible interface 15:22
which it is..
stevan webmind: well we alreayd have CGI.pm, and File::Spec, that should be a good start
webmind ah no
stevan :P
webmind I'm using the tempplate toolkit
-p
stevan ah
gaal that would be nice to have in p6 :) 15:23
autrijus hm, I wonder if Test.TAP will make a good haskell library
gaal: indeed :)
webmind template toolkit and Image::MetaData::JPEG are the main modules I'm using
gaal autrijus, for haskell itself, or for pugs?
webmind gaal, yes
Image::MetaData::JPEG is still experimental though
autrijus gaal: for haskell itself, _and_ for pugs
shapr Oh hey, I need to hook up with Alias again next time he's online and ask him for more HaskellDemo suggestions.
gaal my OR was not exclusiuve ;-) 15:24
autrijus as nice as quickcheck is, sometimes you want simple harnesses
oh. :)
gaal although ", or" is often parsed as XOR in english, isn't it.
shapr Yes, I think so.
autrijus $ perl -e 'undef = 3' 15:25
Modification of a read-only value attempted at -e line 1.
$ perl -e '(undef) = 3'
# ok
sigh.
I wonder if perl 6 is going to keep this behaviour ;)
gaal what got assigned as 4 :)
err,
4--
ahem. :)
autrijus :)
gaal oh, it's there so you can do 15:26
(undef, $x) = something()
autrijus yeah.
stevan autrijus: I see the (undef) = 3 being somewhat similar to the _ in some functional langauges (Erlang is where I know it from) 15:27
autrijus right.
it's just overloading what undef stands for.
stevan although I would prefer something more like _
autrijus yeah.
stevan but _ is already taken
autrijus is it?
stevan $_, @_, etc
and _ is the filehandle right?
-e _
gaal before i knew about this syntax, i'd spell that as ($junk, $x) = ...
stevan gaal: me too :)
gaal _ is an annoying keyword in p5. it's always in main:: 15:28
autrijus, i think i send you a patch to gettext ages ago about that :)
autrijus that may be the case :)
gaal defininf sub _ { } doesn't look like it works
although it really sorta does. 15:29
stevan gaal: this works perl -e 'sub _ { print "hey" }; _();'
gaal stevan, try doing that in package Moose. 15:30
stevan ah 15:31
I see
gaal what do you know, it works. maybe in older perls it didn't?
stevan it just died in 5.8.6 for me 15:32
gaal ahhh, right, i remember the problem.
perl -le 'package Moose; sub _ { print "hey" } package Elk; _()'
and anyway, now filiesh _ is screwed for life. :)
i think when i first encountered this problem i had to trace perl to figure it out 15:33
anyway: what's ambiguous about the read here? 15:34
lineNum <- option Nothing (do { n <- many1 digit ; return $ Just (read n) })
....and is there a prettier way of saying this? 15:35
stevan ok the patched Test::Harness and the Test.pm with PUGS_RELEASE taken out are commited 15:36
gaal yay. stevan, want me to write an extract_todo tool?
stevan I will mess around with the :todo<bug> :todo<feature> stuff after this release
gaal: what would it do?
autrijus gaal: you can use the "integer" builtin 15:37
gaal remember we said we wanted make test | force_todos > todos
autrijus or "decimal"
I think "decimal" is what you want
gaal autrijus; okay, but still what's wrong with what i tried? 15:38
autrijus gaal: nothing, it's just you have to annotate
(read n :: VInt)
stevan gaal: let me work out the :todo<attr> thing first
autrijus gaal: since you had not used lineNum anywhere
once you have
then that is unneccessayr
as it will know the type
gaal ah, i see!
autrijus actually, :: Int is right
VInt is not it
gaal thanks :)
autrijus the joy of inferencing :) 15:39
gaal i suspected something like that was happening because that very same line worked until i commented out something later
stevan I am off to do my weekend errands, bye all 15:40
autrijus: I will get that changelog to you later this evening (6-8 hours)
theorbtwo is back.
castaway is forward
gaal bye, stevan! 15:41
gaal is ever sideways
theorbtwo BTW, I'm going through and adding explicit signatures, since haddock documents signatures, and not so much the actual functions/instances. 15:43
autrijus stevan: cool 15:44
theorbtwo: yes, please go ahead
as complete as possible -- ghci is your friend
but please drop module name prefixes.
i.e. not Pugs.AST.Exp
and if you see room of narrowing 15:45
like
(Eq a) => [a] -> a
where it is only used for Val or Exp or Type
then please narrow it
it will buy us much performance according to skew.
(and is more readable anyway)
theorbtwo Can't use ghci. Have been using ghc --show-iface=src/Pugs/foo.hi some, but that doesn't list everything (presumably it only lists exported functions when there is an explicit export list). 15:46
Will keep the narrowing advice in mind.
Have mostly been using error messages.
autrijus k.
"make ghci" 15:47
and then
:m Pugs.Prim
:t foo
has worked for me
gaal is what :t does a callable funciton? 15:48
if so we can dump this for theorbtwo pretty easy
autrijus I don't know, I think hsplugins may have access for it
gaal find -name \*.hs| xargs perl -le 'print $1 if /^(\S+) =/ 15:49
but don't tell skew i used a regexp for that ;)
theorbtwo GHCi runtime linker: fatal error: I found a duplicate definition for symbol blkcmp 15:50
autrijus seen corion 15:51
jabbot autrijus: I havn't seen corion , autrijus
autrijus hm. I was about to ask whether he can write to p6l for me
or rather, for himself
castaway he's awake on PM, autrijus
autrijus Corion reported that <code>for @array { say }</code> broke, because <code>$_</code> is not the implicit argument in the inner block -- <code>@_</code> is. I have special-cased to turn the block used in <code>for</code> to default to <code>$_</code> for now, pending further clarification.
gaal to2, rm src/UnicodeC.o
castaway shall I poke him?
autrijus castaway: or you can write p6l :)
castaway meee?
autrijus sure, you!
castaway hides.
autrijus the problem needing clarification is a simple one :) 15:52
castaway last I read, it should topicalise to $_
castaway looks where that was
autrijus pjcj: paul at pjcj dot net? 15:57
invitation sent :)
theorbtwo Don't forget to add yourself to AUTHORS. 15:58
castaway S06, under Switch Statements .. ?
".. any block that sets $_, including a for loop" 16:00
gaal autrijus, sorry for not just understanding this parsec thing already :), but: in oneLineComment, if somewhere i evaluate skipToLineEnd, that consumes the input and quits the parser successfully, right?
...but not if i'm in a nested do 16:01
...in that case i have to return skpToLineEnd, yes?
autrijus return skipToLineEnd? 16:02
gaal or maybe
skipToLineEnd
return ()
?
the number of returns must match the number of dos? 16:03
autrijus well
if you say
skipToLineEnd
it is the same as 16:04
foo <- skipToLineEnd
return foo
if you want to do something but ignore what it captured
you do
thatSomething
return ()
read existing Parser.hs code for enlightenment (or more confusion) :)
I need to sleep.
theorbtwo G'night, autrijus. 16:05
gaal i'm of the 'be confused until you're enlightened' school.
castaway night autrijus :)
gaal i just hope my eventual enlightenment matches my current confusion :) 16:06
autrijus 'nite!
gaal night.
autrijus gaal: I'll pray for your nirvana :)
gaal thank you!
TheHunter hi, what's the easiest way to get this Test::TAP::Model required for pugs' make smoke installed under debian? 16:07
theorbtwo Easiest? sudo cpan Test::TAP::Model 16:08
gaal TheHunter: do you have cpan?
TheHunter no.
castaway you do, unless you have a perl older than 5.6.0
(or possibly older)
TheHunter oh, it's v5.8.4 16:09
gaal actually the one with all the deps is Test::TAP::HTMLMatrix
so do what theorbtwo said but with that module instead.
castaway then just type "cpan" on in a dos prompt
gaal takes a break &
castaway umm.. s/dos prompt/terminal/ sorry
theorbtwo Best? apt-get install dh-make debhelper; mkdir dh-make; cd dh-make; dh-make --cpan Test::TAP::Model; cd Test-Tap-Model-<tab>; sudo debuild; cd ..; sudo dpkg --install ./libtest-tap-model-perl<tab> 16:10
(Or something a lot like that.)
TheHunter ok, thanks.
shapr castaway said dos prompt! <insert beavis/butthead 'uhuhu' quote> 16:13
castaway someone on another channel was looking for windows info, honest!
shapr grins
castaway ruffles shapr, long time no see 16:14
shapr Hiya castaway! How's the island?
Work is getting in the way of my free time.
castaway bah, work
tis purty good tanks 16:15
TheHunter ok, the best way is too painful. I'm gonna have to live with /usr/local being messed up by some perl stuff. 16:21
castaway then just to cpan <RET> install Test::TAP::Model 16:22
theorbtwo Hint: When it asks you if you want to do manual configuration, "no" is an acceptable answer, and it'll try to autoconfigure. 16:23
TheHunter oh, for the love of god. "make test had returned bad status, won't install without force"
using force acceptable 16:24
theorbtwo Yep.
shapr gets the hammer
TheHunter oki.
castaway hammer?
shapr Use enough force Luke! 16:25
TheHunter ok, thanks all. 16:26
putter With a few-minutes-old pugs, my make test is massively failing with "strange object 68" "Please report this as a bug to [email@hidden.address] Is this a known problem? x86_64. 16:58
shapr There are a few known GHC on x86_64 bugs, but I don't know if this is one of them. 16:59
putter tnx. checking... 17:00
theorbtwo Uff. Suddenly, for no reason I can see, util/munge_haddock.pl (with an argument!) is now hanging until I hit C-d. 17:05
ingy nothingmuch, gaal: ping 17:15
nothingmuch ingy: pong
i was wondering if there was a kwid impl that can do full kwid->html 17:16
but then i figured out Kwiki::Formatter->new->text_to_html($foo) was good enough for me
ingy nothingmuch: working on it
nothingmuch: meet obk 17:17
obk waves
ingy obk: this is nothingmuch
nothingmuch hola obk
obk Hi
nothingmuch too many of us here
ingy I think you guys are countrymen
nothingmuch israeli imperialism ;-)
see also gaal, lightstep, lumi, roie_m 17:18
obk Brian thinks we know each other
ingy obk wrote the YAML spec
obk True, Israel is about the size of Seattle, but still :-)
nothingmuch well, supposedly everyone knows everyone in israel, but we know that isn't really true ;-)
obk I hear you are doing a great job on the Perl6 interpreter
ingy but you guys all look the same ;)
nothingmuch gah, fscking petal... why won't it reencode the entities it decodes 17:19
obk Ingy, I don't look Israeli - if there is such a thing
ingy I am trying to persuade obk to open up his great work on the YAML parser, ala pugs style
obk promises to open it up next weekend
nothingmuch does that mean it'll be in haskell or p6?
obk Nope. Straight ANSI C. 17:20
nothingmuch ah
ingy his impl is in C
autrijus wants to do a haskell version
but autrijus wants to do too many things :)
nothingmuch yeah, /me always does that too
so you mean the develpment model?
ingy aye
nothingmuch sure, i'm going to learn svk
and hack on pugs 17:21
and whatever
obk I'm certain it would be easier to write it in a functional language. I wrote a complete programming parser/lexer in a few 100 lines of dfcp...
ingy oren's work is quite amazing from what I've seen
nothingmuch at least i learned about catalyst this weekend
obk Its actually a horrible object-like framework - with coroutines - on top of ANSI C. Yuck.
nothingmuch obk: what are you using now?
YACC?
obk Nope, Yacc is useless for a streaming parser.
In fact, it pisses me to no end there's no common streaming parser generator out there. 17:22
nothingmuch doesn't know either =P
don't know YACC, that is
ingy autrijus: ping
obk The difference is between "getNextToken" and "parse(tokenCallback)"
ingy is she asleep yet? 17:23
obk I ended up writing a framework for a getNextToken parser myself, and I'm now applying it.
I don't know whether the Haskell compiler knows how to do this sort of thing automatically - thread-less producer/consumer? 17:24
nothingmuch god damnit, i think i've read about 2000 lines of code this weekend
sick and tired
ingy autrijus: if you are around you should meld minds with obk for a few minutes
obk Ah, reading other people's code is the best way to get sick and tired...
nothingmuch at least search.cpan gives you a recursive grep
ooh, i think i can hack this in 17:27
nothingmuch hopes no one will ever read the source code to the beyahad website
castaway ingy, he went to bed a while back 17:29
ingy castaway: ok
nothingmuch obk: want to be the first person to see it that actually reads hebrew, apart from family and my boss?
boss being friend who needed this website
woobling.org:81/ 17:30
obk is looking
nothingmuch brb, i'm going to shoot it 17:31
obk What do you mean, "read Hebrew"? It seems like a normal web site. What am I missing?
nothingmuch see if it will encode now
most people i showed this too are english speakers =)
obk It does display in Hebrew in firefox under Gentoo...
Ah. 17:32
nothingmuch it better... i worked hard getting catalyst to pipe out utf8
theorbtwo "The connection was refused when attempting to contact woobling.org:81."
obk nothingmuch turned it off
nothingmuch ok, up once more 17:33
nothingmuch bitches and cries
theorbtwo Looks nifty. 17:34
nothingmuch grazie, theorbtwo
obk What is character 0u0097? It doesn't exist in my Hebrew font.
nothingmuch probably mdash
or copy
i'm trying to get Petal to re-encode the entities it so graciously expanded for me 17:35
theorbtwo Hm, my copy of unicode says that 0x97 is nonexistant.
obk Mope, it seems to be some sort of control character. I got it in woobling.org:81/page/%D7%A7%D7%91%D...7%99%D7%AA 17:36
theorbtwo Em dash is U+2014
crysflame aiee
obk It appears right after "autonomy" in the first sentence of the second subject.
theorbtwo I see the box-with-0097 in my firefox.
My, that's a really obnixious-looking URL. 17:37
obk Hebrew encoded as UTF-8 and then escaped using URL %xx. As ugly as it gets...
BTW, does anyone know if there's a way (a font name or something) to force the display of arbitrary characters as these numeric-value unicode boxes? It is useful for my debugging of YAML parsing errors. 17:38
nothingmuch obk: yeah, i'll make the URLs bilingual eventually
obk: that's an mdash 17:39
theorbtwo Hm, should work to set it to show in a font that doesn't include non-ascii chars.
nothingmuch theorbtwo: i'm trying to fix that right now by making MkDoc::XML::Encode use a better hash
theorbtwo www-950.ibm.com/software/globalizat...ch&us= lists 0x97 as "control". 17:40
nothingmuch and not do only 'amp', 'lt', 'gt' and that other thing
hah
Corion nothingmuch: You did stuff in Catalyst, right? Is there a guided tour of Catalyst somewhere? I feel/fear that mostly, Catalyst creates a lot of scripts for you once, and if you change stuff, you have to recreate. Where do I look/start?
theorbtwo I think you got snookered by a Microsoft perversion of a standard charset.
nothingmuch then no wonder it's broken
Corion: look at ::Tutorial briefly 17:41
then ::Intro briefly
generate, and look at all the .pm files
they're simple
Corion willdo :)
nothingmuch then read through ::Intro more carefully
it's very streight forward
Maypole was a heap, but this is very simple stuff
Corion nothingmuch: I didn't find Maypole hard. In the end, I mean. But I want-need to implement some good browser for a dataset, and it screams for a webapp (import/export/transport browsing) 17:43
nothingmuch it's easy
Corion ... that is, import/transport/export of data between programs
nothingmuch: I'll look into it
nothingmuch but when i tried it was unfun to hack
catalyst's code is much more streight-forward
at least that's what maypole was when I tried it 17:44
damnit! why is it being loaded after me 17:45
nothingmuch sighs 17:47
nothingmuch types 'sudo gvim /usr/local/lib/perl/site_porl/...' for the 100th or so today 17:49
mauke hoho, porl
theorbtwo It might work better if you s/porl/perl/.
nothingmuch oh! MKDoc not MkDoc
i hate case insensitive FSs 17:50
theorbtwo: try p<tab>
obk Well, got to go... Keep up the good work.
nothingmuch ciao
putter is there an easy way to disable gc? 17:59
elmex ? 18:00
putter sigh. regret to inform that 6.4 is no longer sufficient to run pugs on x86_64. 18:03
elmex can't connect to svn.openfoundry.org ?!
putter allegedly a compacting gc bug, fixed in HEAD and eventually in 6.4.1. 18:04
make optimized is also failing (differently?). 18:05
elmex has the svn rep of pugs moved?
svn: PROPFIND of '/pugs': could not connect to server (svn.openfoundry.org)
theorbtwo I got that a minute ago, apparently people have been having problems all day. 18:06
Somebody had a svn switch command, which I was going to find and then got distracted (thanks).
Corion I have that problem too, but kept silent about it :) No pugs hacking this weekend. Or so I think ;) 18:07
svn switch
or something. The backlog has it.
Or svn relocate
I don't know
elmex hmm...i tested pugs the day before yesterday... and somehow experienced not that much speed ...
castaway openfoundry.dyndns.org 18:08
svn switch --relocate
elmex ./pugs -e 'my ($i, $a); loop ($i = 0; $i < 10000; $i++) { $a += 3 } say $a' takes 5.4 seconds for me...
putter aside from make, make optimized, and make profiled, is there any other pugs variant I can try to hopefullly sidestep the gc problem? 18:11
(clarification: I haven't actually tried profiled yet) 18:12
theorbtwo svn: PROPFIND of '/pugs': could not connect to server (openfoundry.dyndns.org)
There goes that idea.
shapr You could try the darcs repo. 18:13
theorbtwo Is it bidirectional?
I have stuff I want to check in.
shapr It depends. 18:14
You can send patches by email just fine.
Also, if your repo is http accessible, other people can just update from your repo as well to get patches.
And you can do the same from theirs, so darcs doesn't really need a server at all.
I don't think autrijus has set up the darcs repo to put patches back into svn, but it will let you work with the pugs sources in a source control system and share patches with others. 18:15
Does that answer your question sufficiently? Or did I leave out some useful details? 18:17
castaway ooh, Episode IV 18:18
elmex btw: a small quest what does this mean in perl6: 18:27
sub { my $fo = open ... or die ...; print $fo "blubb\n"; } - does it mean: write "blubb\n" to a file and close it, OR does it mean: write "blubb\n" to a file and forget the filehandle and leak it? 18:28
theorbtwo elmex: Write "blubb\n" to a file and close it... when we get around to it. 18:35
(Except that's say $foo: "blubb", or $foo.say("blubb").) 18:36
elmex theorbtwo: well... perl6 doesn't gurantee timely destruction
theorbtwo Right. 18:37
elmex and now imagine i call that sub 40000 times
and the gc doesn't run
becaus i give it 10000 mb ram
or maybe just 30mb
=> no memleak, but my progam will run out of filehandles
theorbtwo It'll probably do a gc run when it runs out of filehandles. 18:38
elmex what do filehandles have to do with memory? i thoguht a GC is there for memory
castaway anything thats garbage .. 18:39
theorbtwo Exactly.
If doing a gc run will sometimes free up filehandles, then doing it when we're out of filehandles seems smart.
elmex theorbtwo: then imagine i do open database-handles or other kinds of handles 18:40
maybe a user defined handle type :)
theorbtwo Opening and closing things over and over is always a poor programming pratice. 18:41
Now there's just another reason it's bad.
elmex theorbtwo: well, it's not _this_ particular example 18:42
theorbtwo: what about modules i use? if they do stuff with handles?
theorbtwo Use the source, luke.
castaway how many is it likely to have that you run out?
elmex theorbtwo: i do have to know the source of all modules i use? 18:43
theorbtwo The untimely distruction the pugs people were talking about was possibly a few ms late under normal conditions.
castaway or at least what they do
revdiablo so, relying on destruction for closing filehandles might not be the best practice
s/file//
elmex castaway: that wont matter. imagine a deamon that has enough ram, so that the GC won't have to run often... will it crash because the handles run out after 10 days where no GC ran? 18:44
revdiablo: so, it's a bug
revdiablo: it will invite perl5 programmers to do that bug. and it's hard to find. and it's not do-what-i-mean
revdiablo imagines elmex has been waiting to say that for 10 minutes
elmex i havent 18:45
theorbtwo It's a sad trade-off.
elmex it's not DWIM 18:46
propably ther will be:
sub { auto $foo = open .. }
castaway opening a handle and not closing it, but waiting for it to magically vanish, is DWIM ?
theorbtwo Yep, it is, castaway.
elmex castaway: in perl5 it's gurateed, YES
castaway ah, nevermind me then 18:47
elmex and a gc run for filehandles: how does the GC determine that there are too many open filehandles?
castaway then I suspect it'll work here eventually ,)
elmex anyway, it's a tradeoff
theorbtwo The way that perl5's GC works has the nice property of timely distruction.
It has lots of not-nice properties too.
elmex theorbtwo: perl5 uses refcounting 18:48
theorbtwo Correct, elemenx.
elmex circular ref's leak
theorbtwo It's slow.
Circual refs leak.
It's very error-prone.
elmex so parrot/perl6 will be fast and circ refs won't leak 18:49
theorbtwo It's very infectious -- it makes it very difficult to write XS code, because all code that touches anything perlish has to know about the refcounting.
elmex well, thats indeed a feature
but the tradeoff is timely destruction
theorbtwo Correct.
But untimely normally isn't that slow. 18:50
elmex of course
perl6 will be rockfast
theorbtwo Laggy, I should say, not slow.
elmex but... well :) poor perl5 programmer that 'updates' his perl to perl6 will probably run into leaking handles if he forgets some { auto $fo = ... } idiom or something like that... 18:51
in end effect: manual resource management 18:52
like java :)
(but with nicer syntax)
theorbtwo Not really, no.
shapr I think I'll enjoy using Perl6. 18:53
theorbtwo Like I said, untimely distruction will not be all that untimely, unless you mess with the gc parameters by hand, in which case you deserve what you get.
elmex well, but a GC doesn't gurantee it will run in the next 10 days if there is enough RAM... or does it also have to know how many mysql handles there are? well... 18:54
shapr: i probably will enjoy it too :)
shapr: there are a lot of cool features
shapr: and it' will probably be quite fast
shapr Though I sort of feel sorry for my future coworkers. If they don't understand Haskell and Perl, they'll be in for a bit of reading. 18:55
elmex :)
*g*
theorbtwo elmex: Perhaps DBI will be smart enough to start a GC run if it appears that it's out of handles.
elmex theorbtwo: so DBI will be fast but not leaky
err... s/fast/slow/
theorbtwo The idea is that a single gc run should be fast, so you won't notice it.
Nope, elmex.
A GC run can be very fast, if you have a very fast GC. 18:56
elmex theorbtwo: how will that be gurateed for 100000 objects? ;)
theorbtwo elmex: 100k objects ain't nothin'. The Parrot people had problems at that load, and fixed them, years ago. 18:57
Now it gets a bit slow when there's more then a few hundred million living objects.
elmex interesting
well, i'm not shure DBI will know whether it runs out of handles... but holding open 100 mysql connections until I run out of them (not caring about other processes or users).. isn't smart 18:59
all that there is to do: the module programmers will have to care about it
chromatic Who's coding today? I have a laundry list. 19:01
castaway nobody, svn is playing up again
castaway points at shapr anyway
shapr wha?
chromatic Seems okay for me at rev #2622.
shapr What did I do?
castaway I'm nominating you. 19:02
shapr I'm busy organizing the second issue of The Monad.Reader
elmex theorbtwo: i also don't really belive that the GC can really be that fast... java GC takes 4+ seconds when taking up 250Mb ram here ...
shapr escapes!
castaway nooo
elmex theorbtwo: but parrot is free surprise me there in the real perl6 world when it's done :) 19:03
chromatic I have here a (theoretically) working port of Test::Builder to Perl 6. 19:04
It awaits some review from the rest of @Larry for a few syntax and implementation issues of which I remain blissfully confused.
However, I have a list of all of the features of Perl 6 it needs.
castaway :)
theorbtwo Theoretically valid p6 (you think), but not working in parrot, you mean?
castaway add them to the todo tests ? 19:05
chromatic I'm not sure how to answer that. It's Perl 6 code, not Parrot code, so no, but it requires nothing from Parrot that's not already there (or emulable).
theorbtwo Err. 19:06
but not working in pugs, you mean.
chromatic Right, I believe so.
theorbtwo is actually surprised to learn that 'chromatic' !~ @larry. 19:07
chromatic Technically, I am: dev.perl.org/perl6/people.html
castaway ooh, secretary, highly technical job that 19:08
chromatic It means that I type faster than anyone else.
castaway heh
theorbtwo wonders if the secretary is responsible for the index. 19:09
chromatic No. That's why I suggested "amanuensis".
nothingmuch mofo! 19:10
nothingmuch hates HTML, and XML, and templating!
*whine* *bitch* *cry* *blame parents*
chromatic Anyway.
castaway ruffles nothingmuch 19:11
chromatic If I were to name the features I needed, are there people who could say "We have that!" or "Write a test in the file named _____.t"?
autrijus 03:04 < chromatic> However, I have a list of all of the features of Perl 6 it
needs
castaway good timing autrijus
autrijus show me the list so I can sleep? :)
castaway *g*
chromatic Line at a time in the window so it shows up in the log or uploaded somewhere? 19:12
autrijus er, nopaste.
castaway wonders how long
autrijus or uploaded somewher.
but best, mail to p6c/p6l
chromatic 32 lines
autrijus perlbot: nopaste
perlbot Paste your code here and #<channel> will be able to view it: sial.org/pbot/<channel>
pasteling "chromatic" at 63.105.17.30 pasted "Perl 6 Features Needed for the Test::Builder Port" (32 lines, 940B) at sial.org/pbot/9786 19:13
elmex chromatic: ah, you've wrote a mail about finalization. do you think that finalizers, that run very late (probably after you run out of filehandles), are of much use? 19:14
autrijus oh, chromatic , as you are here, a quick question
for @array -> { $_++ }
is it legal?
chromatic elmex: it depends
autrijus is $_ somehow special that it has a "is rw" tackled on?
nothingmuch theorbtwo: can you please try that URL again, and tell me if the mdash displays properly?
chromatic autrijus, I'm not sure. I *think* so, but I'm not sure.
nothingmuch then maybe i can forgive myself for what I did to get that working 19:15
elmex chromatic: you have to make shure that your finalizers, who close filehandles, run before you run out of filehandles. and not when the gc decides that it runs out of mem
chromatic elmex, that's why I think it's important to make a clear distinction between finalization and destruction.
castaway we dont have classes yet, do we? 19:16
chromatic Finalization is for clearing up used resources, such as filehandles, database connections, sockets, and unmanaged memory. Destruction is for cleaning up managed memory.
autrijus ok, let me shorten that list to things that Pugs doesn't yet have.
castaway autrijus: didnt I answer that yesterday/earlier ?
chromatic That's a little complicated when you finalize something that you're holding onto elsewhere, but that's a problem with destruction anyway.
autrijus castaway: yes you did
castaway (yes it defaults to rw)
theorbtwo nm, I still see the 0x97 box.
Even after a force RW.
autrijus what is "hash key setting" 19:17
nothingmuch in that case I think your font doesn't have '&#x97;' and i should change from it
chromatic $status{"skip"} = 1;
elmex chromatic: i think, to have some kind of mostly-timely destruction, you need something like: { auto $fh = open "foo.txt" or die; ... print $fh; } - note that a my $fh will be collected on next gc run. => either on scope-exit a run or when you run outof mem.. if you do scope-exit collection, you have to have a fsckin fast gc
autrijus oh. sure.
P5-style global substitution with negative-width lookahead
if you can settle for rx:P5// 19:18
then we are all set
chromatic $line ~~ s:perl5:g{\n(?!#)}{\n#};
autrijus cool.
theorbtwo That's not suprising; like I said, it's not actually in latin-1 or unicode, just in MS's perversions thereof.
nothingmuch oh, didn't know that
i think i missed that part in chat
chromatic elmex, I think people want timely *finalization*, not timely destruction.
nothingmuch got any reccomendation?
autrijus ok, 8 things. 19:19
object attributes of Bool, Int, Str
private class attributes marked "is rw"
invocantless attribute access in methods
empty classes
class inheritance
method redispatch through *WALK to supermethod
submethod allowed
delegation of methods to objects stored in object attributes
nothingmuch &#8212; ?
theorbtwo www-950.ibm.com/software/globalizat...o/unicode, search for "m dash" at the bottom under "name".
autrijus other things are either already there or trivially true after the 8 things are implemented.
that doesn't sound so bad.
theorbtwo U+2041.
chromatic It's not particularly complicated code.
autrijus now if we can compile that to parrot.
elmex chromatic: well, yes, of course. but how do you do timely finalization? if you want to do it with a GC it hast to be fast. if you use some kind of syntactic sugar like 'auto' it's kind of manual resource management
autrijus I think from the list that it's quite doable too. 19:20
chromatic I could probably rewrite parts of it in IMCC, but I don't know how that translates to Parrot AST.
autrijus I thought the point of parrot ast
is that you don't have to know that.
you can drop to imcc level at any time
chromatic elmex, it's an end-of-scope action. In theory all it has to do is register the auto-finalize stuff with that hook which can be very fast. 19:21
autrijus but, hey, if you can translate it to IMCC, that's wonderful
chromatic It doesn't have to mark and sweep the world or scan the current generation.
autrijus because my IMC generator monad will just use what your translation offers.
when th^W^Wif that works, then the Pugs AST _is_ Parrot AST.
chromatic autrijus, the advantages of IMCC are that it works and it has documentation. 19:22
autrijus sigh, laggy connection
elmex chromatic: yes, i hope that will work if i do { my $fh = open ... } ... objects need a special tag... and there needs to be a system for that...
autrijus chromatic: as compared to Perl6? :)
nothingmuch try again?
chromatic Let me rephrase that then. The advantages of IMCC is that I already know how to write it or, at worst, read the code.
s/is/are/
autrijus hm. and you can't do that for perl 6? 19:23
I'm puzzled :)
chromatic See, now you're just picking on me. 19:24
Where's the respect for your elders?
autrijus pouts. I'm genuinely curious. :)
Corion . o O ( Pugs didn't get started out of respect :) )
autrijus but, in all seriousness. I'd love to see your code, and if you can translate whatever part of it to IMCC 19:25
then I'll make sure that Pugs can compile those parts to equivalent IMCC.
how about that. :)
chromatic Sure, that's pretty easy.
autrijus woot.
chromatic I've asked Damian and Luke to review the code to pick up any changes I've missed and I'll send it in as soon as they do. 19:26
autrijus oh btw.
my $foo = { say $_ };
what is signature of $foo? 19:27
*$_ or $_ ?
or neither? both?
chromatic $_. 19:28
It doesn't splat.
It only takes the first argument.
autrijus so
$foo(1, 2)
is fatal?
chromatic It depends on how strict signature checking is. 19:29
autrijus because that's what happens when you have a nonsplatty scalar.
"Too many arguments" as defined in S06
chromatic There might be a restrictive pragma in effect.
autrijus ok. but I'll make it nonsplatty. thanks!
chromatic Anytime!
autrijus oh, a related question 19:30
chromatic Go ahead.
autrijus sub { say "123" }
it has a *@_ signature? 19:31
sub { say }
what about this?
chromatic I don't know about the first one, but I guess that the second has *@_, or at least propagates the signature of say(). 19:32
That's just a guess, not a Mosaic ruling.
autrijus sub { say; say; say }
I don't think that propagates anything, does it?
so, in this
for 1..10 -> { say }
does the pointy there also have *@_? 19:33
and how does say pick up @_[0] as $_?
chromatic Good question.
autrijus is for() explicitly topicalizing?
but if it is merely a statement:<for>, how does it do that?
chromatic In the absence of placeholders, how do you know the count?
autrijus right.
chromatic I can only agree that that's the question! :)
autrijus my current heuristic is that signatureness closures, when used in "fo" 19:34
"for"
assumes a ($_)
since that is what most people expect.
chromatic That seems reasonable in the absence of any other ruling.
autrijus can you check it for me, or should we bring it to p6l?
for 1..10 -> () { say "you won't see this" } # and this infinite loops... right?
chromatic I expect not, but I don't know. 19:35
theorbtwo I don't understand why that'd inf-loop.
chromatic I expect for to consume the looping, not the arguments.
autrijus hm? 19:36
theorbtwo Oh -- each time through the loop, it eats 0 arguments, until 0 are left of the original 10.
autrijus for 1..10 -> ($x, $y) { ... } # this loops 5 times.
and 10 / 0 is ... Inf
theorbtwo Right, was being momentarally stupid.
chromatic It takes no parameters, not zero parameters.
It's one of those "this is important to math" differences, I guess. 19:37
autrijus not understanding
castaway 10 / undef ?
chromatic If there are multiple parameters, it does eat up multiple values.
If there are no parameters, it eats up one value per iteration, but you can't access them by name. 19:38
Eating up multiple parameters is a special case, like multiplayer Pac-Man.
castaway and default is 1 ?
chromatic Yes.
theorbtwo If there's no explicit signature, it eats one parameter (into $_).
castaway makes some sense
theorbtwo But there is an explicit signature.
chromatic So no parameter.
Inside the block.
autrijus ok. 19:39
autrijus slaps a (min 1) somewhere
theorbtwo Making it an error?
chromatic Think of a number, any ten numbers, but don't tell me what they are!
castaway ,)
gaal other oprions are run never or raise compile time error?
chromatic I don't know about the error. I could argue either way. 19:40
theorbtwo Oh. So it takes one parameter, and throws it away?
chromatic theorbtwo, yes, though I might say it throws away a potential parameter before it even thinks about taking it.
It steadfastly refuses to take a parameter.
gaal if we take the run never option it might be useful as a quick way to comment out a block.
autrijus I'll do the "min 1" ;)
someone write a test? 19:41
castaway stubborn thing ,)
gaal cygterm + home.tiscali.de/emilio.lopes/screen/screen.html == almost non sucky windows system! 19:43
Corion gaal: Why would anybody want to do that? :-) 19:44
gaal because the windows terminal sucks rotting wooden donkey arse. 19:45
theorbtwo At some point, you get to the point of "just run linux already, you silly idiot".
Corion gaal: I'm quite content with it...
theorbtwo: But then you lose the hardware, printing, gaming and about everything else :-)
gaal theorbtwo, you don't understand
i already do run linux
nothingmuch Corion: i remember you wanted something like google complete
Catalyst does that
gaal on the other machine.
nothingmuch one second, i'll show you 19:46
gaal but the other machine is slow :)
Corion Am I doing something wrong, or is my network connection broken? "svn switch --relocate svn.openfoundry.org openfoundry.dyndns.org" gives me no love
nothingmuch: I know (S)ajax
theorbtwo Gives me no love either, Corion.
nothingmuch it's tré´s´´coó´l
Corion theorbtwo: Ah :)
gaal Corion, svn has been acting up since ~12 hours ago.
on and off.
Corion theorbtwo: But that would be bad anyway, 'cause you got your love at home ;)
gaal ah, you know that. 19:47
theorbtwo Oy.
theorbtwo plays with his çômþößé kèÿ. 19:48
shapr w00
castaway ,)
theorbtwo BTW, X is stupid -- compose,i,j does not give ÿ.
shapr Shouldn't it give the Dutch connected ij ? 19:50
Oh wait, y uml is the same thing, right?
Yeah, I remember, y uml was a mistaken form of the Dutch ij.
autrijus really zzz & 19:56
castaway sleep well, autrijus
theorbtwo Sleep well, Autrijus-San.
(Or is that s/San/san/?)
Corion -sama :)) 19:59
ingy Perl6-Bible-0.16.tar.gz on cpan now 20:40
obra hey ingy 20:46
ingy yes?
obra how goes? 20:47
that was a hello, not a "and I need..."
ingy it all goes
obra ;)
ingy oh
hey there obra ;)
I'm doing good 20:48
out with the old in with the, well, older actually. you know
is the pugs repos down again? 20:49
mauke does anyone here know of a binary distribution of ghc 6.4 for x86 linux that doesn't require glibc 2.3? 20:50
castaway It::All::Goes ? 20:56
ingy castaway: that sounds like a cool nnew project 20:58
don't tempt me
I just released a new Test::Chunks
It is getting very cool 20:59
Can't wait to port to Perl6
need classes first...
hmm.
I should just preport it
wolverian I have been putting off coding things, waiting for pugs to get OO 21:00
not a very productive attitude, I guess
ingy wolverian: yeah
wolverian you too? :)
ingy I don't write anything but 1-liners that aren't OO these days
and i don't write OO that isn't Spiffy 21:01
of course Perl6 is already Spiffy
shapr I'm tempted to ask how you define OO...
ingy built right in
shapr But I should probably write worky code instead.
ingy shapr: ask away
shapr No, I am afraid :-)
ingy what is your concern?
haha
shapr I will never get my work done if I start an interesting conversation!
autrijus hm. I notice that tpe.freepan.org is never down when svn.openfoundry.org goes down on weekends 22:11
despite them being in the same building.
autrijus seriously considers switching over to freepan
Limbic_Region autrijus are you still up or have you slept already? 22:12
autrijus I have slept for... I think 40 minutes 22:13
I feel refreshed!
Limbic_Region my wife would kill me
jhorwitz is jealous -- needs 8 hours to feel refreshed. 22:14
autrijus eh?
castaway werent you in bed? ;)
Limbic_Region just that between work and playing on the computer when I get home - "bed time" is her time
she would be just a tad bit upset if I got up after only 40 minutes 22:15
autrijus ah. 22:16
Limbic_Region autrijus - the failing last.t test is due to last inside a nested loop exiting from all levels - not that I know the first thing about Haskell, but if I were going to start looking where would I start?
autrijus Limbic_Region: hm. enterLoop in Monads.hs defines it 22:17
I wonder if it's broken always or just recently
Limbic_Region AFAIK, there was no test testing it 22:18
until I added one
autrijus confirmed as broken 22:19
likely because the outer last is seen
Limbic_Region well if you know how to fix it without much thought be my guest 22:20
Limbic_Region will try and add more tests and find another bug
;-)
autrijus fixed. 22:21
castaway heh 22:22
Limbic_Region my idea to learn p6 better and spread advocacy is start answering questions at the Monastery with a p5 and p6 answer 22:24
much like Randal did in the early days of Perl
and when applicable, add the solution in examples
but - I really do have a lot to learn about p6 and everything about Haskel
is there any alarm() and sig handling in Pugs yet? 22:34
theorbtwo Yes. 22:36
Even with subsecond resolution and working on win32, IIRC.
(It starts a second thread, which sleeps and kills the master thread... again, IIRC.)
Limbic_Region hmmm
autrijus so the problem is quite insidious 22:42
for 1..10 { for 1..20 { last } } 22:43
the problem is that a closure is defined to capture the lexical pad upon its creation
so the { last } closure captures the &last of the outer loop
and because pad restoration happens _after_ enterLoop introduces the inner &last 22:44
the outer one is restored as the binding to use.
the moral is that &last should probably not be implemented as a lexical sub. 22:45
Limbic_Region autrijus - so probably even worse of a problem when last <label> is supported? 22:46
autrijus yeah. the proper fix is to switch to use VControl. 22:47
(as is intended)
Limbic_Region same deal with next then
autrijus right.
Limbic_Region well, and redo for that matter 22:48
Limbic_Region has only ever found a use for redo like twice
theorbtwo Should VError and VControl be that different? 22:49
autrijus theorbtwo: I think so; S06 (or S04)? makes it clea 22:50
clear
that control exceptions and error exceptions are different
theorbtwo I thought control exceptions were a subtype of error exceptions, with just slightly different semantics... but it's been a while. 22:51
Huh? op2DefinedOr = undefined ? 22:53
nothingmuch beh 22:54
autrijus ok, I have a
genSymCC "&next" $ \sym -> do
now.
it will do for now.
nothingmuch finally goes to try and do wrap
Limbic_Region 3 arg open isn't in Pugs right? I am looking at tests/examples and not seeing it 22:55
autrijus yup 22:56
nothingmuch what's the temp hostname for OF again?
Limbic_Region and "is chomped" is still todo
hmmm
I guess I will just answer the question as though those features were available today ;-) 22:57
Limbic_Region notes the time and heads off to watch some TV 23:00
TTFN all
castaway openfoundry.dyndns.org (in theory)
theorbtwo Last I tried neither worked. 23:01
autrijus that is right.
another 24 hours downtime.
I think I'll switch to freepan.org ;)
ingy, gugod: around?
Darren_Duncan are you talking about the non-working subversion server? 23:02
autrijus Darren_Duncan: yes.
it seems that they periodically has problems on weekends.
Darren_Duncan right, that's mainly what I came here to mention
nothingmuch mutters something about distributed VCS
castaway thinks she'll switch to the bedroom.. 23:03
g'night
Darren_Duncan also, for those not in the know, the Ars Mac OS X 10.4 review by John Siracusa is already out
at arstechnica.com/reviews/os/macosx-10.4.ars
theorbtwo thinks he'll follow castaway.
Darren_Duncan these are always excellent, and this year's is several weeks early
as soon as I get my copy I'll be able to test Pugs on it, in a couple weeks 23:04
specifically, whenever the ADC people send it out to non-seed-getting people 23:05
autrijus oh. I accidentally implemented next and redo.
theorbtwo Accidentally?
Darren_Duncan how unfortunate
autrijus yeah. was just debugging the last() problem
Darren_Duncan how long until you can recover? 23:06
autrijus and refactoring things
Darren_Duncan: I have no idea :-/
theorbtwo They just kind of slipped out of your fingers... you tried to hold them back, but they were gresey little bugguers...
autrijus exactly
Darren_Duncan those are a good compliment to last 23:07
jhorwitz autrijus: mod_pugs lives. crippled, but alive. 23:10
autrijus jhorwitz: oh. wow. wow. 23:11
Wow.
url?
jhorwitz autocompiles to IMC, but runs in a PHP-style mode until we have objects to use the apache API 23:12
autrijus fantastic.
if you can write pseudocode with objects
theorbtwo PHP-style mode ?
autrijus then I'll make that happen.
jhorwitz theorbtwo: content only -- can't use the apache API
autrijus: need to put on a public server, then will post the url 23:13
theorbtwo Woo, Pugs.Prims now has 100% top-level signature coverage. 23:14
autrijus woo indeed
say, we all have freepan committership here, right?
Juerd jhorwitz: Just please don't make the mod_perl mistake again.
autrijus Juerd: what would that mistake be?
jhorwitz Juerd: which one? ;-)
Juerd jhorwitz: Which is to make the content system and the powerful does-everything system the same thing.
mod_perl will never be as popular as php 23:15
autrijus ok, next/last/redo is consistently available.
Juerd *because* it is powerful.
It's hard to do simple things with mod_perl
theorbtwo Well, because it's hard to make it safe for multiuser systems without per-users apacheds. 23:16
Juerd And people don't trust it because they think it's insecure.
jhorwitz Juerd: mod_parrot is meant to be a framework for writing the higher level modules
Juerd theorbtwo: PHP has exactly the same problem and got popular regardless.
jhorwitz: There must be two mod_thingies, im{not at all}ho
jhorwitz ? 23:17
Juerd jhorwitz: One light weight and fast, php-ish
jhorwitz: One to write plugins for apache
theorbtwo I'm going to go go to bed, but I'll try to check in from there.
autrijus is migrating full history to freepan
Juerd The first should specifically be webserver independent
The other can be apache specific
autrijus pugs, that is
Juerd And the full blown, hard to grok thing should NOT be a dependency of the other
Only then can Perl ever reach PHP's ease of use.
autrijus the first one is spelled SpeedyCGI
or FastCGI 23:18
etc?
Juerd You have to approach it from a sysadmin's point of view as well as a programmer's point of view
theorbtwo &
jhorwitz Juerd: i *am* a sysadmin. :)
Juerd autrijus: No, all too much setup and things to be careful with.
autrijus: The thing that sucks most about PHP is its language. 23:19
autrijus oh. I see what you mean.
Juerd autrijus: The way it's installed in apache is very friendly.
autrijus ponders mod_mason
or mod_catalyst
Juerd autrijus: Perl 5 can't do it correctly because interpreters are very heavy and resetting them takes a lot of time.
autrijus Juerd: yeah, already grokked. 23:20
Juerd autrijus: Which means forking and then exiting after a single request is faster even :|
It takes practically no time to set up a PHP interpreter 23:21
It doesn't have a big bunch of states to be set to initial values
And PHP is internally more modular in these areas 23:22
autrijus *nod*
theorb Hmm, is password-changing implemented on freepan yet? 23:23
autrijus I thought the idea is for you to commit your password as a special prop
no idea if it's implemented or not.
but I figure the only way to raise peer pressure is to use it :)
(still migrating) 23:24
jhorwitz: I want to see mod_pugs :) 23:25
Juerd "Only Python modules under a proprietary licence are allowed. GPL and BSD licences are particularly forbidden."
HAHA
theorb wiki.freepan.org/index.pl?Changing...anPassword suggests you email ingy a md5 hash of your username and pass.
jhorwitz moving bits as fast as he can... 23:26
autrijus theorb: I've changed my password already this way, I think
still on r174. 23:27
nothingmuch arstechnica.com/reviews/os/macosx-10.4.ars/5 <-- is anyone grokking from this that say, postfix will be launched automatically when someone tries to talk to port 25? 23:29
theorb Hm, what you just described is inetd -- been around halfway to forever. 23:30
Trades having to keep daemons running vs slow response time.
nothingmuch right, but not exactly
delay packet, raise postfix, reroute packet 23:31
then kill it after a while
maybe
theorb Ah.
Interesting idea.
Not sure it sounds worth the effort.
nothingmuch but i'm not exactly sure what that is saying
and then fullfill deps by delaying dependant processes
by 'figure out dependencies automatically' i thought it would just try to start everything
but i have no clue 23:32
theorb Write your daemon well, start it on boot, let it swap out if it wants to be.
jhorwitz autrijus: did you want to see code or a demo? 23:35
autrijus jhorwitz: both.
code, preferably 23:36
but demo is cool too
jhorwitz autrijus: that's what i was afraid of. :) parrot on my public server is WAY out of date, so demo will be a while. code is simple enough to paste.
perlbot: nopaste? 23:37
perlbot Paste your code here and #<channel> will be able to view it: sial.org/pbot/<channel>
theorb Blasted flakey network. 23:39
autrijus hmm, corion's trick broke. 23:40
he used
pasteling "jhorwitz" at 68.236.4.147 pasted "a *very* simple pugs layer for mod_parrot" (57 lines, 1K) at sial.org/pbot/9794
autrijus map { $_.values } map { $_.values } (any(...) ~ any(...))
that is patent nonsense. :-/
and it's all my fault for letting it work before
I've now changed it to say
map -> Junction $_ { $_.values }
instead.
jhorwitz: wow. 23:41
jhorwitz autrijus: simple, eh?
autrijus so when I declare pugs to be just another compiler object inside parrot 23:42
like Tcl now does
you can cut the spawnw
jhorwitz exactly.
autrijus and directly feed code in, right?
nice.
jhorwitz++
jhorwitz actually, if that happens, you can write this layer in pugs
autrijus that's riight.
jhorwitz mod_pugs without a line of C 23:43
autrijus this was your original idea all along, isn't it.
jhorwitz nods
autrijus jhorwitz++
you'll of course post this to both p6c and p6i, right? :) 23:44
jhorwitz once i commit the changes and have a demo
theorb Hm, does this mean that we'll get eval_tcl for free, along with eval_imc ? 23:46
autrijus theorb: aye.
and eval_commonlisp
and eval_python
et cetera.
as was the plan all along.
plan 1..Inf 23:47
ok 1
ok 2
...
theorb laughs evily.
Forget about editors with embedded languages. 23:48
Our langauge can embed yi /and/ emacs!
At the same time!
theorb wonders if oneLineComment should be renamed, since it also parses things that look vaugely like comments, but aren't. 23:50
autrijus jhorwitz: so, what does the handler in p6 look like 23:51
jhorwitz autrijus: since we have no objects, mod_pugs just spits out whatever the handler returns 23:52
sub handler { return "Hello, world!
";}
autrijus: you mentioned pseudocode -- i need to call methods on the PMC argument passed to the pugs handler 23:53
it's completely opaque
autrijus that is fine. we actually have opaque objects. 23:54
jhorwitz then i guess method calls really won't be that hard to implement. :) 23:56
autrijus the problem is what to dispatch to.
do I just emit code that calls the method back? 23:57
i.e. if you call with an argument $foo and then invoke $foo.bar() 23:58
jhorwitz $r.puts("hello") would roughly translate to mypmc.'puts'("hello")
autrijus ok.
jhorwitz assuming mypmc was passed in as $r

Powered by Cro and the Raku® Programming Language.

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

Thank you!