6macros: discussing the finer points of Perl 6 macros, Qtrees, and how to stay sane | irclog: irclog.perlgeek.de/6macros/today
Set by moderator on 12 July 2015.
01:25 vendethiel joined 04:25 vendethiel joined 05:57 vendethiel joined 07:10 vendethiel joined 07:39 vendethiel joined 08:09 vendethiel joined 08:36 vendethiel joined 09:33 vendethiel joined 10:56 Ven joined 14:14 vendethiel joined 14:39 vendethiel joined
masak so, I've been thinking. 17:52
(it's great that this channel exists, because these are only half-formed thoughts so far, and I don't want to have everyone reviewing them just yet)
hooks per grammatical category 17:53
something like "whenever we parse a function (and get a Qtree), also call this user-provided function/macro (which can inspect and manipulate the Qtree)" 17:54
among all the things we know we want to do, this wish seems the most straightforward.
my use case is from 007: someone wants to extend `my foo;` declarations to be able to handle type declarations: `my Int foo;` 17:55
oh, actually, that's an interesting but slightly too advanced use case to start with. 17:56
better to start with [Coke]'s idea of having checked exceptions from the other day. for that to work, some function (like above) would have to inspect each parsed sub to see if it throws a checked exception -- and if it does, make sure that the sub is marked up with an appropriate trait. (or die otherwise) 17:57
vendethiel interesting idea. 18:01
that'd be lexically installed?
masak yes, that makes sense to me. like operators. 18:02
...but exportable.
(disclaimer: 007 doesn't do exports in any form yet. nor does it have modules.) 18:03
vendethiel first thought: is there a way to opt-out from that, say, to generate subs from macros that will not respect this rule?
well, I guess you can generate { no x; our sub ... }
masak hm. 18:04
I'm not sure I see the need. feels like it'd be better if the installed hook regulated what it applied to.
the downstream client code, in a sense, isn't supposed to care how it's being analyzed/mogrified. 18:05
vendethiel hence why I said "for generating a sub inside a macro" 18:06
but let's drop that idea for now
masak ah, ok 18:13
don't understand, but will keep in mind that there may be a use case
certainly there are interesting timing issues involved.
for example (like with macros), I think that there should be no retroactive firing of hooks on already parsed things. too messy. 18:14
vendethiel right 18:15
still feels like action-at-a-distance somewhat :P
reminds me of jnthn++'s OO:: modules
which used macros to check for the correct state declared in the "class"'s "is state(<...>)" 18:16
masak I'll have to re-check those and compare. 18:18
oh, it's definitely action-at-a-distance. so are source filters, though -- and I'd rather use what I'm proposing. 18:19
vendethiel no question here
your "my T $a" seems like a really complicated use-case though 18:34
the non-homoiconic syntax bites here
masak I was thinking about that one a bit more since I suggested it 18:52
in the best case, I'd simply like to be able to install a `rule statement:my-typed`, and it gets to compete in LTM with `rule statement:my`. 18:54
actually, the bothersome detail there -- and something I haven't quite figured out yet how to solve -- is that such a new rule would generate an AST, but what the rest of the compiler needs is a Qtree. 18:56
provisionally, the answer to that is "action methods should generate Qtrees".
oh, and coincidentally 007 does! neat :)
vendethiel the "AST" should really be the QTee 18:58
r*
masak yeah. 19:00
well, the Perl 6 compiler is still allowed to generate something like QAST for later stages. but it shouldn't be user-exposed in the same way. 19:01
vendethiel right.
I think the problem with typed-my is far deeper than that, though.
you need to "install" stuff to handle "=" and all that
masak yes, I saw that from looking at the current `my`'s action method 19:02
basically, it's really hard to do this without incurring quite a bit of code duplication
I've sort of come to accept that
writing it down to the fact that we have no "lore" for how to extend grammars and action classes in this way yet
hopefully we'll get better at it
unlike with classes, there seems to be a much bigger temptation in grammars to open them up and mess around "in the middle". 19:03
vendethiel i'm not happy of where grammars are :) 19:06
masak Perl 6 grammars allow subclassing to extend into new behaviors. but I can't recall ever seeing a successful real example of that.
vendethiel they're both too far and too close
masak yeah
vendethiel example: your generation thingie
I think the same about MMD
masak MMD I really like. 19:07
vendethiel Oh, I sure do as well
masak I can see a point where, for the sake of user-induced extensibility, I'd have to let go of the Perl 6 grammar in 007's parser.
vendethiel but I strongly think it should be first-class.
they*
well no, it. dumb me. it's still MMD i'm talking about
masak how would MMD look if it were first-class. 19:08
?
you could do a dispatch, get an object from that, and then later use the object to actually call the method found? 19:10
vendethiel something like "match" 19:11
I need to code something like "newtype" soon-ish 19:31
I'm afraid of string programming etc taking over. 19:38
and it seems like Perl6 is bound to see that happen
masak heaven forbid. 19:39
vendethiel well, that's how it's gonna happen 19:45
maybe I should start from another angle? 19:46
a context-sensitive template engine 19:47
where you can not interpolate strings
a "SafeString"-like thing, where you're sure you can't get XSS'd
masak good luck. 19:48
vendethiel it's not really about luckj 19:50
-j
it's about stopping complaining about how things are, and moving them where they could be, I guess
masak indeed.
well, good that, then. :P 19:51
vendethiel tough choice. 20:05
how could I emulate typeclasses in perl6?
I also don't want people to be scared if I write some kind of "scalaz-perl6", aka "siz" 20:07
masak please be more specific with "typeclasses". are roles with `...` methods enough? 20:14
vendethiel how do you give an implementation "from outside"? 20:19
masak oh, I see. 20:20
vendethiel Haskell has global instances (which leads to many issues)
scala has "import" 20:21
masak only clumsily, I suspect. Adapter pattern and the like.
vendethiel we can't specialize a role after it's been defined in a module
masak no; roles are immutable after definition. 20:24
vendethiel exactly. sooo. :( 20:25
masak feels almost like subset types in Perl 6 are closer to what you want. 20:27
except they are not very static.
vendethiel yrzh 20:29
yeah*
i'm pretty aware of what subsets can do :P
I've thought about it
masak just hopes he can get 007 into a demo-able state in the next few weeks 20:32
vendethiel checks if his talk has been accepted 20:36
"Status\t \tAccepted" \\o/
masak \\o/ 20:37
vendethiel now to write it...
I think I sent you my draft already 20:38
masak I think you did.
23:02 Ven joined

Powered by Cro and the Raku® Programming Language.

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

Thank you!