svn switch --relocate svn.openfoundry.org/pugs svn.pugscode.org/pugs/ | run.pugscode.org | spec.pugscode.org | paste: sial.org/pbot/perl6 | pugs.blogs.com
Set by avar on 16 November 2006.
00:02 buetow joined 00:07 Psyche^ is now known as Patterner
putter anyone around? 00:07
luqui yeah 00:08
putter I'm puzzling over release backends. It looks like the current candidates are p5, parrot, hs(?), whatever vm someone writes and emitter for. that sound about right? 00:09
And of those, the only one which could conceivably not be mostly written in p6 is hs, yes? 00:10
luqui I'm not really sure what you're talking about... 00:12
are you wondering what to call the backends for the release?
putter It Xmas. We have a releasable perl6. There seem several possibilities. Its vm could be p5, parrot, ... 00:15
00:16 cognominal joined
putter Though I think question is whether anything by a hs-based p6 might not be implemented in p6 itself. At one time self hosting was said to be a post 6.0 concept. 00:16
But it looks like the only current senario where that seems plausible is a hs-based release. 00:17
I'm looking for a reality check, plus any comments like "selfhosting is now speced for 6.0", or "hs-based release has been removed from consideration". 00:18
or even "vm is now required to be p5-based for compatibility/deployment reasons". 00:19
any idea?
luqui as far as spec goes
I think it's whichever one happens
we would like it to be parrot based 00:20
putter k
ok
thanks
00:27 nipra joined 00:30 nipotaway is now known as nipotan 00:37 mako132_ joined 00:40 diakopte1 joined, diakopte1 left 01:02 lyokato joined 01:04 b00t joined 01:11 autark_ joined
putter Anyone know if www.perlfoundation.org/gc/grants/20...posal.html is still the parrot roadmap? 01:12
lambdabot Title: The Perl Foundation
01:18 lisppaste3 joined
luqui dunnos 01:20
putter tnx
luqui putter, i searched the meeting notes for "roadmap", and nothing consequential came up 01:23
so either it's still in effect, or it has been gradually phased out without anybody noticing :-)
01:24 lisppaste3 joined 01:34 diakopter joined 01:41 shobadobs joined 01:43 shobadobs left 01:44 rashakil joined
putter luqui: lol. my thanks. 01:46
sigh. dsl's without a full language - just dandy until you hit the inevitable screw case. :/ 01:56
revised www.vendian.org/mncharity/Temporary...leteme.png and .dot 02:20
& 02:22
02:24 lisppaste3 joined
luqui that's a nice graph. it's motivating ;-) 02:25
02:43 penk joined 02:45 vel joined 02:55 ProperNoun joined 03:15 mdiep joined
luqui @tell putter what can I do to help on mp6/kp6? 03:16
lambdabot Consider it noted.
03:17 b00t joined 03:26 mdiep joined
luqui @tell putter whoops, fglock is doing that stuff, not you, right? hmm, I'll ask him 03:27
lambdabot Consider it noted.
luqui @tell fglock what can I do to help on mp6/kp6?
lambdabot Consider it noted.
03:43 leed joined 03:51 mako132_ joined 03:52 Aankh|Clone joined 04:36 nipra joined 04:47 nperez joined 05:00 thestarslookdown joined 06:15 BooK joined 06:22 GabrielVieira2 joined 06:33 avarab joined 06:46 avarab_ joined 06:48 avarab_ is now known as avar 07:11 ilogger2 joined 07:15 Aankh|Clone joined 07:33 avarab joined 07:39 iblechbot joined 07:43 avarab_ joined 07:46 xinming joined 08:01 Aankhen`` joined 08:30 spx2 joined 08:37 Lunchy_ joined 09:04 elmex joined 09:29 ruoso joined 10:03 takesako__ joined 10:16 mj41 joined 10:19 fglock joined
masak c2.com/cgi-bin/wiki?MultiMethods 10:22
lambdabot Title: Multi Methods 10:23
fglock luqui: re kp6 - right now, it would be nice to have someone review the design decisions 10:32
lambdabot fglock: You have 2 new messages. '/msg lambdabot @messages' to read them.
fglock putter: re Data::Bind - it looked like a problem with perl itself, and there was a bug report to p5p 10:33
10:55 yves joined
svnbot6 r15077 | fglock++ | kp6 - added 'Assign' to the grammar 10:56
r15078 | fglock++ | kp6 - fix last commit 10:59
11:07 xinming joined 11:15 xinming joined
svnbot6 r15079 | fglock++ | kp6 - added ChangeLog 11:17
fglock re S06: how does an anonymous our-subroutine work? it can't be added to the namespace 11:27
11:29 f0rth joined
fglock also re S06: s/package GLOBAL/module GLOBAL/ 11:31
11:33 cmarcelo joined, avarab_ is now known as avar
fglock re S02: how about replacing 'package' with 'namespace', to avoid confusion with p5 'package' 12:01
12:53 ofer0 joined, Limbic_Region joined
svnbot6 r15080 | kudra++ | Corrected spelling of a word 13:06
r15081 | fglock++ | kp6 - added a Perl5-emitter plug-in 13:15
pasteling "evalbot_r15080" at 194.145.200.126 pasted "Pugs build failure" (375 lines, 20.9K) at sial.org/pbot/22428 13:19
Juerd Oh my... evalbot pastebots :) 13:22
Incredibly clever :)
rgs when will it begin to propose patches :) 13:23
Juerd Or just commit them :P
13:25 buetow joined
svnbot6 r15082 | fglock++ | kp6 - added kp6-perl5.pl script - compiles kp6 to perl5 13:30
13:30 weinig joined 13:32 iblechbot joined 13:42 gnuvince joined 13:45 thepler joined 14:09 [particle] joined 14:13 VanilleBert joined
masak ?eval [ 'a' .. 'z', 0 .. 9 ].pick(10).join() 14:14
evalbot_r15082 "etk2pr1oxv"
masak very well
Juerd pick takes an argument. Heh.
masak but pick(10) doesn't put them back, does it?
Juerd What do you mean by "put back"?
?eval ~~['a'..'z',0..9].pick:10 14:15
masak random selection with/without replacement
evalbot_r15082 Error: ā¤Unexpected "~~['"ā¤expecting program
Juerd ?eval ~~(['a'..'z',0..9].pick:10)
evalbot_r15082 Error: ā¤Unexpected "~~(['"ā¤expecting program
Juerd ?eval ~['a'..'z',0..9].pick:10
evalbot_r15082 Error: ā¤Unexpected ":10"ā¤expecting term postfix or operator
masak if I want replacement, I have to do this:
Juerd ?eval ~['a'..'z',0..9].pick: 10
evalbot_r15082 "w k f h 8 j o y t u"
masak ?eval ( { ['a' .. 'z', 0 .. 9 ].pick() } xx 10 ).map:{ .() }.join()
evalbot_r15082 "8g5w3muf20"
masak loves p6
Juerd What's the difference?
masak slight, but there 14:16
Juerd Which is the difference?
:)
masak in the latter, I could get the same letter twice
in the former, I never will
Juerd Ahh
masak hence "with replacement"
like an urn, you know :)
Juerd ?eval ~['a'..'z'].pick(26).join
evalbot_r15082 "shunzcmjtpofixqgldkvbaeryw"
Juerd That ~ wasn't needed :) 14:17
masak yep, all there
Juerd ?eval ['a'..'z'].pick(52).join
evalbot_r15082 "fmwydzeuatonqrlcbjighkpxvs"
masak heh
can't fool pick
Juerd Too bad
?eval { ['a'..'z',0..9].pick } xx 10
evalbot_r15082 ({Syn "block" {App &pick (Syn "\\[]" {Syn "," {App &infix:.. (: ā¤ "a", "z");ā¤ App &infix:.. (: 0, 9)}}: )}}, {Syn "block" {App &pick (Syn "\\[]" {Syn "," {App &infix:.. (: ā¤ "a", "z");ā¤ App &infix:.. (: 0, 9)}}: )}}, {Syn "block" {Ap
masak yow 14:18
you need to eval them
with .()
s/eval/apply/
Juerd Then I'd still expect 10 subs.
Or is this how pugs stringifies closures? :) 14:19
14:19 vel joined
masak guess so 14:19
Juerd ?eval { ['a'..'z',0..9].pick } xx 10 >>.()
evalbot_r15082 Error: ā¤Unexpected ">>.()"ā¤expecting operator
masak I think the >>ops are broken 14:20
been so a while now 14:21
prove6 t/operators/hyper.t has a few :todos in it 14:22
14:25 chris2 joined 14:26 [particle] left 14:30 weinig is now known as weinig|bbl 14:37 theorb joined 14:46 nipra joined 14:47 elmex_ joined 14:51 theorb is now known as theorbtwo 14:52 jferrero joined 14:53 [particle] joined 14:56 bonesss joined 14:59 fglock joined 15:24 jferrero joined 15:30 hexmode joined
hexmode km;/ 15:31
15:35 fglock_ joined 15:40 fglock joined 15:44 xinming joined
TimToady pick has an optional boolean parameter specifying :repl 15:47
not implemented though
15:48 CardinalNumber joined
TimToady and "package" is the Perl word for "namespace" 15:48
I don't see any reason to make another alias for that.
masak TimToady: :repl, great 15:52
TimToady I do kinda wish I could say ['a'..'z',0..9].[rand(*) xx 10] though
?eval ['a'..'z',0..9].[map {rand 32}, ^10] 15:53
evalbot_r15082 ["h", "e", "b", "e", "x", 4, "q", 5, 0, "h"]
TimToady ?eval ['a'..'z',0..9].[map {rand 32}, ^10]
evalbot_r15082 ["b", "h", "a", "y", "e", "e", "q", 3, 0, "t"]
TimToady that works
er, except it should be 36... :/ 15:54
[particle] of course, that's base 36 15:55
so, can you get the size of the list inside the selector?
TimToady the list might not even be generated yet 15:56
[particle] of course it would make it unlazy
TimToady ?eval ['a'..'z',0..*].[map {rand 36}, ^10]
15:56 xinming_ joined
evalbot_r15082 pugs: out of memory (requested 1048576 bytes) 15:56
TimToady oops
[particle] notes, 'stop saying "of course"'
TimToady if @a[*;1] has * meaning all of the current dimension, then maybe +* is the size of the current dimension. 15:59
not quite sure if ['a'..'z',0..9].[map {rand 32}, ^10] 16:00
er
not quite sure if ['a'..'z',0..9].[map {rand +*}, ^10] would work though
[particle] i think it needs something explicit to act upon 16:01
TimToady well, I'm not gonna think about that much. this morning I'm much more worried about how we carry the longest-token rule through a grammar consistently. 16:04
I'm not sure the <%foo|%bar|%baz> notation is sufficient, quite.
16:05 diakopter joined
[particle] where does it fall down? 16:05
TimToady several ways 16:06
the inheritance problem could be worked around saying %.baz instead of %baz
16:06 diakopter left
TimToady and saying that if you have "rule baz", it can be used as a hash 16:06
in which case it is required to be able to cough up a table of its initial tokens as a set of keys 16:07
16:07 diakopter joined
TimToady there's also problems in the followup to the initial token, however 16:07
at the beginning of a statment, for example, you are in a superposition of wants 16:08
statement_control
declarator
prefix
term
etc.
some of those have to fire off additional rules
the hash allows a rule for the value of the hash
but offhand I don't see how to recognize a prefix or term and fire off the operator precedence rule 16:09
there's a token lookahead problem in general
most recursive descent parsers assume a tokener with one token lookahead 16:10
but the definition of token depends on our current state.
and the tokens that are legal in a particular parser state 16:11
may be specified not just by a top set of %grammatical_categories but by various subrules
[particle] to be clear, <%foo|%bar> is a true superposition (junction), and not logical-or 16:12
TimToady you need the transitive closure of all the head tokens of all the rules.
not a junction
it could be any character there to indicate a sequence of hashes
<%foo+%bar> would work as well
[particle] you're right, not a junction
TimToady the point is that the hashes as a whole follow longest token 16:13
[particle] do the keys of %foo get precedence over %bar
TimToady while ties are broken in favor of the earlier hash
[particle] ok, i see
longest token overall
with stable sort 16:14
TimToady so it's a bit like the MMD policy that a "my multi" can hide a particular long name, but other long names can be pulled in from outside
but if we have to go as far as interrogating each head rule for its initial tokens, then the hash notation is superfluous
all we need is something like [ :longest <a> | <b> | <c> ] 16:15
[particle] sure, but it's much shorter
it's almost as if the operator is spelled |%
TimToady or a different |
yes 16:16
|+ maybe
[ %statement_control |+ %declarator |+ <op_prec_parser> ] 16:17
[particle] that stands out better visually 16:18
TimToady though probably %.foo form instead for inheritance ease 16:19
and then the derived grammar doesn't have to temporize the hash for a lexical scope 16:20
[particle] oic. yes, that works.
it makes the syntax a little clunky, but for good reason 16:21
[ %.stmt_ctrl |+ %.decl |+ <opp> ]
TimToady actually, declarator is just another term, so it belongs down in opp as well. 16:24
we could finesse it for Standard Perl without the transitive property by making sure that statment_control is always an identifier 16:25
but if someone were to add a punctuational statement_control, then we don't have the safety of whitespace anymore
and it should really abide by the longest token rule, even if the longest token has to bubble up out of the <opp> rule. 16:26
oh, which also has to be parameterized by its terminator, so another strike against using a hash for that.
the other approach would be to use a set of hashes to peek ahead and figure out, the call <opp> with "pos" before the peeked token. 16:28
16:28 kanru joined
TimToady but that doesn't seem as clean 16:28
s/the call/then call/
[particle] what if jumping between parser types (rec-desc, opp, custom) is undefined behavior in hash combinators 16:30
TimToady problem is we'd have to use different hashes for the same thing in different contexts
so %prefix in the opp would have to just parse the operator, but at statement level would have to parse the whole expression 16:31
[particle] that's no good :( 16:32
TimToady and I think it's just a lot niftier if the rule engine can calculate the transitive differences from a DRY specification
than if we have to copy stuff around
[particle] agreed.
[particle] goes to get wet
TimToady but it does seem to call for |+ or some such. 16:33
yeah, ich auch. &
16:49 CardinalNumber is now known as ProperNoun 16:54 CardinalNumber joined
svnbot6 r15083 | fglock++ | kp6 - working 'hyper-to-map' transform 17:01
17:13 ashleyb joined 17:26 GabrielVieira joined 17:46 fglock joined 17:54 xinming joined 18:09 ProperNoun joined
TimToady I have majorly bitten the bullet. 18:13
see p6l for change of | semantics to longest token.
now we use || for old | semantics
[particle] saw that 18:14
TimToady for a while it was in the transitional form where only :ratchet reversed the meaning, but I decided consistency was more important with how | and || work outside
and we get || for free because null patterns are illegal 18:15
[particle] that paragraph is a bit dense. an example might help
gaal say, $x in "for @list -> $x { ... }" is always my, right? is there a way to specify it otherwise?
TimToady it's a formal parameter 18:16
gaal not that having it available is necessarily a good thing, but the one use case i'm thinking of is if you want to look at the last iterator value if you break early from the loop.
TimToady so you can do anything to it that you could do to the $x in "sub ($x)"
gaal hmm, i don't see how that fact can be used to change the storage type 18:17
[particle] $OUTER::x?
TimToady while my $x = =@list {...}
gaal yeah, okay. :) 18:18
TimToady my $x = first {...}, @list 18:21
gaal oh, that was a yeah of assent, not sarcasm :-) 18:22
TimToady np 18:23
gaal (hmm, typed affirmative calculus)
TimToady my $x = @list.first: {...} 18:24
I think I like this language. :)
kolibrie timtowtdi, obviously
gaal hee
TimToady anyone care to hack .first into Functions.pod as a variant of grep? 18:26
18:28 avar joined
kolibrie the Little Red Hen must be baking bread 18:39
nothingmuch the giant will eat her any moment 18:40
TimToady "Not I" said the Little Fluorescent Green Pig. 18:41
stevan looks in the fridge and sees that all the blue ice cubes are gon 18:42
18:44 larsen_ joined, [particle] joined 18:45 bernhard joined
svnbot6 r15084 | lwall++ | Added first function. 19:02
19:04 neonse joined
kolibrie notes that TimToady is not a Little Fluorescent Green Pig 19:08
19:13 CardinalNumber joined
svnbot6 r15085 | fglock++ | kp6 - cleanup traverse/visitor 19:13
fglock the sun here was just like this: www.geol.ucsb.edu/faculty/sylvester...ges/18.jpg 19:17
lambdabot tinyurl.com/y8k8sj
19:32 ashleyb joined 19:38 CardinalNumber is now known as ProperNoun 19:39 ziggy joined
ziggy hello, just a quick complete noob question: is there any tutorial/book teaching perl6? and i don't mean bits and pieces, something coherent and complete 19:41
Juerd No, not yet 19:42
[particle] perl 6 design isn't coherent and complete yet
ziggy hehe, guess i'll have to wait another 365 days :(
Juerd No, christmas is before then. 19:43
ziggy last christmas and the one before that i heard the same :D 19:45
[particle] one day maybe you'll believe it ;) 19:47
ziggy how did you guys learn perl6?
[particle] by writing it. seriously. read the specs, and write some tests. 19:48
ziggy wow
[particle] the tests are all written in perl6
wanna help?
ziggy so there are specs?
Juerd ziggy: I learned Perl 6 by reading Apocalypses, and trying to stay up to date after that
ziggy: But if you start today, I would recommend that you read only the Synopses.
[particle] spec.pugscode.org 19:49
lambdabot Title: Official Perl 6 Documentation
Juerd ziggy: See topic :)
[particle] ziggy: want a commit bit, so you can help?
Juerd ziggy: And/or access to feather, so you can try before you "buy"? :) 19:50
ziggy wow, not so fast, let me chew a few synopses, try pugs, and then i'll ask if i consider myself usefull :)
what's feather? 19:51
10x for the link
Juerd ziggy: feather is a machine
[particle] ziggy: anyone can get a commit bit, even you
Juerd ziggy: People who want to try or work on Perl 6 (that includes Pugs) can get a shell account.
ziggy: This was initially because many people found it hard to compile Pugs, as GHC is rather demanding during compilation. 19:52
ziggy nice, too bad i can't use it, because i've got a lame isp that drops 20% packets and ssh is so sloooooow
i have no alternative, i must compile it myself
Juerd Good luck, then! :)
ziggy what do you mean by "demanding"?
Juerd It's not that hard, but it takes some time.
ziggy: It uses lots of RAM and CPU.
(The first time you compile it, that is. Subsequent compiles are much faster if not much was changed) 19:53
ziggy aren't there any binaries available?
Juerd There are some, but they're releases, and always out of date. 19:54
ziggy ah, ok
Juerd There is also feather.perl6.nl:8080/runpugs/ if you want to try simple stuff (not entire programs) 19:55
lambdabot Title: Run Perl 6 Now -- in your browser!
Juerd ?eval 1 + 5 # This works too, but please don't use it a lot :)
evalbot_r15085 6
ziggy wow, nice bot :P
?eval say "hello" 19:56
evalbot_r15085 OUTPUT[helloā¤] Bool::True
19:57 ofer0 joined
[particle] ?eval "hello".say 19:57
evalbot_r15085 OUTPUT[helloā¤] Bool::True
[particle] perl6++ 19:58
ziggy say is a string class method? :D
[particle] ?eval 4.say 19:59
evalbot_r15085 OUTPUT[4ā¤] Bool::True
[particle] everything is an object
ziggy can you add methods to builtin classes, like in Ruby?
[particle] yes 20:00
ruby's older sister just got an extreme makeover :)
ziggy ok, you convinced me to read the specs :P 20:01
[particle] you need a commit bit. msg me your email address 20:02
ziggy ugh, i can't, i have to register a nick first
[particle] /msg nickserv identify <some_passwd_here> 20:03
hrmm, maybe i can start a private chat...
20:04 ziggy is now known as Ziggy6
TimToady you can always use the "foo at whatever dot where" approach... 20:06
Juerd TimToady: Which is utterly ineffective.
[particle] Ziggy6 msg'd me successfully. bit sent.
Juerd TimToady: Spammers also like Perl and its regexes :) 20:07
Ziggy6 thanks, i got the email
20:08 polettix joined
kolibrie Ziggy6: after you checkout pugs, add yourself to the AUTHORS file 20:08
Khisanth [particle]: you can also set a mode so unregistered people can send you messages
[particle] how can i cpan upload from feather? 20:09
Juerd [particle]: I don't understand the question.
[particle] i need to upload parrot-0.4.8.tar.gz to cpan
it's on feather
i'm on windows, using putty
Juerd [particle]: There are several ways.
[particle]: You can use a command line utility for it, there are several on cpan 20:10
[particle]: You can create an SSH tunnel and use the web interface
[particle]: You can use w3m for the web interface
[particle]: Easiest is probably to put the file in your ~/public_html, and have PAUSE download it from there
[particle] ah, yep, will do
first, lemme make sure ithe tar is working 20:13
gaal [particle]: also, on the putty d/l page you'll find many ssh-based command line tools, eg for scp and sftp etc.
[particle] thx, gaal, i'll look at that, too
gaal also there's the cpan-upload tool, which is great when you're like me and find the most bugs in your code immediately after releaseing it 20:14
[particle] i figured i'd release 0.4.8_01 first for sanity check
gaal I've done !cpa:p and then ^04^05 (or similar) more than once... 20:15
[particle] besides, i'm not sure if i have proper rights to upload parrot yet
20:17 cognominal joined 20:25 trym joined 20:32 fglock joined
TimToady fglock: hello, I hope the new | semantics will not cause you much heartburn. :) 20:33
It means you don't have to implement <%foo|%bar|%baz> as a special case. :) 20:34
merely as a general case.
since <%foo> | <%bar> | <%baz> is now guaranteed to behave the same. 20:35
20:37 Ziggy6 left, pmichaud joined
pmichaud TimToady: ping 20:38
[particle] pmichaud: he was just here, i think he's hiding behind something 20:40
20:41 luqui joined
pmichaud okay 20:41
I have a ton of questions on the latest S05 changes :-) 20:42
TimToady whistles
pmichaud oh wait, do I hear him?
TimToady It is pitch dark. If you proceed you are likely to be eaten by a grue. 20:43
luqui I, also, have a ton of questions on the latest S05 changes
Well, it's only one question, actually, but it weighs a ton 20:44
TimToady cool, fire at will, just don't fire at me.
pmichaud well, I'll start with the simple one first -- does @foo act like @foo[0] | @foo[1] | ... or @foo[0] || @foo[1] || ... ?
[particle] thinks TimToady must be very far away, because he's hiding behind a spec 20:45
TimToady Seems like | would be nice
pmichaud I agree :-)
it would then match %foo
TimToady then you don't have to sort it
pmichaud or at least they would tend to act more similarly
TimToady on the other hand, how do you then write the ordered one?
luqui that does have implications for infinite lists...
pmichaud the other question I have is more about the implementation description in S05 20:46
TimToady k
pmichaud the "list of initial constant strings" has to be a runtime decision, yes? 20:47
TimToady luqui: is that your 2000lb question?
luqui no
I expect it is similar to pmichaud's coming question
so depending on what you were asking, yes
pmichaud i.e., in order for a rule to report back on the initial constant strings of an initial subrule, we'd have to know what grammar we're in at the time the match is being performed
TimToady yes, that's correct 20:48
but I see it much like mmd optimization
pmichaud why then we do we need to return the initial list? Couldn't we just have the rule report back "here's the initial constant string that I can match"?
rephrasing 20:49
why then we do we need to return the initial constant strings as a list? Couldn't we just have the rule report back "here's the longest initial constant string that I can match"?
TimToady doesn't allow for much optimization
pmichaud I'm not sure we're going to get much optimization anyway, since we have to determine it at runtime 20:50
luqui notes that that's how parsec does it
TimToady people will change the string to search much more often than the grammar used to search it
luqui isn't sure how fast parsec is
pmichaud so, we could compile the table of initial constant strings at runtime 20:51
i.e., each rule maintains a list based on the grammar of the match
TimToady then you just merge those as you know the connections
assuming lists are presorted longest to shortest 20:52
pmichaud do we have to be able to invalidate them if new rules are added via out-of-band means?
TimToady or build a trie
pmichaud I guess each grammar could keep track of them
TimToady I presume so, including adding entries to grammatical categories
pmichaud i.e., we can ask a grammar "what are the initial constant strings for rule xyz?" 20:53
TimToady but maybe the update can be done incrementally.
pmichaud which it then computes if it doesn't know already
TimToady or used to know and forgot...
pmichaud and it saves information as it goes so that the next time it's asked it doesn't have to recompute the whole thing
TimToady crudely, generation++ would work 20:54
but most changes via declaration are incremental
postfix:<!> and such
so could probably propagate the fact that you need to do an insertion on your table or trie
without rebuilding the whole thing 20:55
pmichaud do we need to specify the optimization mechanism in S05, or can we leave it up to the implementations to decide the best way to determine "longest initial constant match"?
TimToady I was wanting to leave that as open as possible
20:55 cdpruden joined
TimToady for some datasets it's probably fastest to falsify on character N rather than character 1 20:56
pmichaud okay, I think I've got it for now
luqui are the semantics of /<a> | <b>/ to match whichever of /<a>/ and /<b>/ is longer, or just to go to the one with the longest initial matching constant string?
TimToady oh, let's see, I'll also need to mumble something about longest tokens requiring \b after identifiers by default... 20:57
luqui and continue from there...
pmichaud as I read S05, try the one with the longest initial matching constant string first
TimToady that's correct
longest token :: rest of rule
pmichaud if it fails, backtrack to try another matching alternative according to the same criteria
TimToady and :: in fact specifically terminates "token" 20:58
luqui so it's concievable that /<a> | <b>/ could match <a> even if <b> ended up being longer...
TimToady so does key/value boundary in hash match
pmichaud does any quantifier terminate "token"?
TimToady believe so
pmichaud that's what I was working from
TimToady (to both of you)
luqui that seems incredibly brittle 20:59
pmichaud luqui: I think if it's brittle then use || instead
luqui i.e. optimizations have the potential to change semantics
that's a cop-out answer...
pmichaud agreed :-)
luqui why should the semantically unstable one be the shorter one?
(or more accurately, why are we including a semantically unstable operator in the language?) 21:00
TimToady I don't see how it's unstable. It's no different than any other lexer/parser works...
every lexer commits to the current token regardless of which production turns out longer 21:01
luqui if you insert "predictive, top-down" in there, you might be right
oh, so you're looking at this from a tokenizer standpoint
but the definition of what a token is is ad-hoc 21:02
TimToady it's defined by the rules you call.
luqui whereas when you're using flex/bison, you have defined exactly what your tokens are
and flex always picks the longest one
TimToady for your current lexer state
and lexer state is usually a proxy for a parser state. 21:03
luqui but where is the line between tokens and productions?
TimToady we're just attaching it directly to your parser state
luqui (is it the "token" keyword?)
TimToady no, token is whatever is constant on the front up to :: 21:04
pmichaud I think the line between tokens/productions is the set of initial constant strings :-)
TimToady or something non-constant like \b \s* etc
pmichaud in something like / a+ / -- would the initial constant string be 'a'?
luqui but that prohibits refactoring constants into data structures... 21:05
21:05 Aankhen`` joined
TimToady %hash keys are considered constants too 21:05
pmichaud: can argue that both ways
pmichaud TimToady: so can I , which is why I asked :-)
TimToady I'm inclined to say not, but don't have a really good reason 21:06
luqui TimToady, didn't you just lose your optimizations by saying that?
(%hash keys are considered constants
)
TimToady not if you know whether the hash has changed or not.
it's just another way for linkage to change
p5 would just have done generation++ 21:07
and recalculated the world
luqui so %hash keys don't count as constants if they are passed as parameters to a rule?
TimToady and been fast usually
we might need to be careful about terminators
pmichaud I think I prefer that / a+ / have '' as an initial constant string, if only to be able to deal with things like / a**{$m..$n} / later 21:08
TimToady maybe hardwire ) ] } as terminators and don't revise hash if the parameter is one of those terminators 21:09
luqui doesn't follow 21:10
pmichaud neither
TimToady most parameters turn out to be terminators, I suspect
whichever way a+ goes, we have a way to write it the other way
:: a+ / would force to ""
s/^/\// 21:11
pmichaud yes
TimToady and / a a* / of course
though that violates DRY
which tends to argue for "a" by default
just for optimizing purposes mainly
less for tokenizing
pmichaud well, if it changes semantics, is it really just optimization? 21:12
TimToady see tailcall optimization...
luqui pmichaud++ # my sentiment exactly
how does tailcall change semantics?
TimToady guarantees you don't blow your stack with recursive solutoins 21:13
luqui given enough ram, there is no change in semantics
i.e. on a turing machine, there is no difference
pmichaud in this case, the chose of returning 'a' versus '' could be very different
s/chose/choice/
unfortunately, I have to run off and make it to the post office to meet a mailing deadline 21:15
TimToady k
pmichaud if I come up with any other questions, I'll ping or write to the ml
(or tomorrow's call)
TimToady k
pmichaud thanks a bunch for the quick answers -- this is fun stuff to think about :-) 21:16
luqui so, TimToady, are you trying to emulate a predictive parser (like parsec) with these changes?
pmichaud luqui: this question came up on #parrot a few days ago -- someone was using an alternation of / .pcc_sub | .pcc_sub_begin / and was curious as to why .pcc_sub_begin never matched 21:17
it becomes odder when one writes token sub { .pcc_sub | .pcc_sub_begin }
TimToady um, we've been predictive ever since we introduced :ratchet 21:18
luqui note that both of those have an initial constant string of ""
pmichaud sorry, backwack the dots
luqui but if the dots weren't backwhacked...
then the problem remains
pmichaud it does also bring about the question of what to do with, say { $foo } :-) 21:19
s/say// # darn puns
I better get to the P.O. if I want to avoid a legal hassle -- bbiaw 21:20
luqui I just don't see the win... it seems like if I use | in a grammar I need to be maticulously knowledgeable about the exact details of how my rules are written to understand the behavior
and it's pretty easy to do a similar optimization on || 21:21
just look at constant strings on the left, and if one matches, jump there and try it
pmichaud luqui: that was true before. What's different now is that I need to know the details of my subrules to understand the behavior
luqui if not, try the right
luqui meant subrules
TimToady pmichaud: go away!!! :)
luqui dittos
:- 21:22
:-)
pmichaud I can't! I'm having too much fun! Taxes are not generally -Ofun!
(gone, for real this time)
ingy /kick pmichaud
TimToady I wonder if commutative should be specified with <=> in place of comma... 21:25
then you can have your first 2 args commutative without the 3rd...
luqui what if you want your 1st and 3rd commutative without the second? 21:26
luqui thinks "is commutative" is good enough
no more syntax for minor details please 21:27
use words for little things, so they're easy to look up, and presumably easy to read without having to look up 21:28
(<=> isn't obvious to me)
TimToady how 'bout multiple signatures for a single body?
luqui woots
*that* is a pretty good idea I think 21:29
TimToady every now and then...
luqui could take a bit of pain out of multi disambiguation 21:30
you still have a combinatorical explosion to deal with now and then...
TimToady but at least you don't have all these {...} scattered all over 21:31
well, need to glare at that one for a while...
luqui (Foo $a, Bar $b | Bar $b, Foo $a) # match the longest initial object... 21:32
21:32 jferrero joined
TimToady hmm, current subsig syntax already allows it, almost, except it would bind both subsigs to the first arg only... 21:33
ah, no, it would be anded not ored 21:34
so subsigs wouldn't work in any case 21:35
I think people would read | as binding tighter than ,
luqui especially since it does a lot of the time (Foo | Bar $x)
but || is available... (might be too much punning) 21:36
TimToady I was just thinking (Foo $a, Bar $b) (Bar $b, Foo $a)
theorbtwo (Foo $a, Bar $b) | (Bar $b, Foo $a) might be clearer.
TimToady or (Foo $a, Bar $b)|(Bar $b, Foo $a) maybe 21:37
theorbtwo grins.
TimToady and rightfully so.
TimToady wonders if all the alternatives must declare the same parameters or fail compilation... 21:39
or if unbound ones just become undef 21:40
luqui I like failing as long as it's easy to specify defaulting to undef
but it's not the same as $param
TimToady former behavior is more conservative, and we could relax to latter later
luqui er
$param? = undef
(i.e. $param? ;-)
TimToady I like the double check, since it would be easy to mix up $a and $b 21:41
though if you did get them backwards, it might have some spectacularly hard to diagnose failure modes
assuming the type inferencer doesn't complain loudly 21:42
which it can certainly do if it decides $a.WHAT === None 21:45
but as soon as people start throwing Any in there we have to pretty much trust them to mean what they said. 21:46
21:49 kanru_ joined
luqui but this is mainly for mmd, right? 21:50
so Any will not be very likely
TimToady ya
gives you an asymmetric default case
luqui because it has no information
TimToady Num/Any
vs Num/Num
luqui elaborate
TimToady it would dispatch a Num/NotQuiteaNum to Num/Any, presumably 21:51
if your mmd coerced its right arg to Num, for instance. 21:52
luqui oh, I see the problem.. if you had (Num $a, Any $b) | (Any $a, Num $b)
TimToady /mmd/mm/
and forget to swap $a and $b
luqui Num|Any = Any, so both $a and $b would get typed as Any
TimToady and how would you detect the error? 21:53
well, you'd have the same problem if you cut-n-paste the body...
luqui aye 21:54
TimToady reminds himself not to cut-n-paste any bodies in public)
one thing the design of Ada taught me is that eventually you just give up and say "that's just erroneous, and we can't do diddly about it." 21:55
luqui I mean, if you have strong type inference on (because weak inference isn't going to catch anything; it's just going to assume you're right), then Any will die on many things, because you can't do anything to it
TimToady saved by the pragma. :)
no stupid;
luqui haha
luqui considers making it so at least the types of the same parameter in different signatures have to match 21:56
but that seems to take away some power
TimToady hmm leave out the params on 2nd and following, and have them autoassociate with the type? 21:57
luqui uh no 21:58
a little too dwimmy for my taste
TimToady prolly
luqui (Any, Num) | (Num, Any) ($a, $b)
oh wait... that's the erroneous case! 21:59
(Any $a, Num $b) | ViceVersa. (I guess that's spelled "is commutative")
TimToady you'd probably want (Any, Num) | (Num, Any) -> Num $a, Any $b {...}
luqui yikes 22:00
theorbtwo Why not just spell that (Any $a, Num $b) | ($b, $a).
Allos explicit types on the second, but do not require it.
luqui that's interesting
TimToady Don't see that it's much different than (Any $a, Num $b) | (Num, Any) 22:01
though I guess that flips the order too 22:02
theorbtwo For one thing, what does (Num $a, Num $b, Str $c) | (Num, Str, Num) mean? 22:03
TimToady also ($b, $a) is a valid sig in it's own right
which always means (Any $b, Any $a)
so you can't also make it default, I think 22:04
luqui ... true ...
TimToady probably best to stay explicit for now
and make the IDE flip entire paramlets for you... :) 22:05
presumably when you hit the "commute" button.
sec & 22:06
22:06 ruoso joined
TimToady (Num $a, Any $b) || (Num $a, Num $b) # desugared ; 22:09
luqui Num, Num? 22:14
TimToady desugared Num; Num 22:16
luqui ?
oh
TimToady note || instead of |
luqui semicolon
TimToady (Any $a, Num $b) || (Num $a, Num $b) # backwards semicolon 22:18
I'm probably all dingled in the head though 22:19
ingy /topic <TimToady> I'm probably all dingled in the head though 22:21
22:25 dduncan joined 22:32 gnuvince joined 22:36 Limbic_Region joined, weinig|bbl is now known as weinig 22:40 ludan joined 22:43 trym_ joined, [particle] joined
TimToady n 22:53
not to mention y
23:07 ludan joined
luqui is beginning to agree with TimToady's assessment about his head :-) 23:12
TimToady I prefer it when you're disagreeable. :) 23:15
23:15 Psyche^ joined 23:24 xinming joined 23:29 Psyche^ is now known as Patterner 23:46 neonse left, ludan joined 23:50 Ziggy6 joined

Powered by Cro and the RakuĀ® Programming Language.

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

Thank you!