The Perl 6 Summary for the week ending 2004-04-25
And we're back on a weekly schedule again (unless the Mayday bank
holiday knocks me for six next week). As I expected, the Apocalypse has
brought out a rash of prophets and prognosticators in perl6-language,
but perl6-internals is still ahead on number of messages per week.
Constant Strings
I confess I'm not sure I quite understand the constant strings patches
that Leo Tötsch and Jeff Clites were discussing. I understand the bottom
line though -- they make parrot a lot quicker when comparing constant
strings. Huzzah!
Then it turned into a discussion of Unicode (or at least, Parrot string
handling which is almost, but not quite, the same thing).
http://tinyurl.com/2mowl
Parrot m4 0.0.4
Bernhard Schmalhofer posted a patch to bring his parrot implementation
of m4 up to what he's calling version 0.0.4
http://tinyurl.com/34be5
SDL Parrot status
Portland Parrot stalwart, chromatic, posted a link to a SDL Parrot site
he's set up with current status, downloadable snapshots and other good
SDL Parrot related things.
http://tinyurl.com/2gcbs
Passing arrays of structs to NCI subs
That man chromatic again, this time he asked if there was a secret
syntax for passing and retrieving arrays of things to and from NCI. Leo
noted that, according to the docs, there is no secret syntax, it's all
documented but unimplemented.
http://tinyurl.com/2xsrc
PMC Constants
Last week, Leo asked for comments on his proposal for PMC constants.
This week, Dan replied. Essentially he's all for the idea, but wasn't
sure which implementation strategy was the best choice.
http://tinyurl.com/2ff8t
Assign and set
Leo announced some changes he'd made to the behaviour of the "set" and
"assign" opcodes. Jens Rieks pointed out a case that he'd missed.
http://tinyurl.com/3bl4u
"hyper" op - proof of concept
Leo also implemented what he described as a rather hackish and
incomplete new opcode called "hyper". Dan liked it enough to suggest
that we should go the whole hog and add a hyper vtable to PMCs, with
hyper versions of all the standard vtable entries. He and Dan had a long
discussion of this, with contributions from various luminaries including
Larry. There was some debate as to whether we really needed overridable
hyper ops, but Dan's adamant that whatever happens they'll be
implemented in a vtable to allow for potential optimizations in some
cases.
http://tinyurl.com/37wqy
Separating allocation and initialization of objects
Last week, Leo posted the latest object benchmarks, and things were
fast. But there was one test where Python was faster. Analysis of the
results and the code seemed to point the finger at Parrot's combined
allocation and initialization. This week Dan confessed that he was
leaning towards separating the two because that would allow for a
standard PCC call into the initialization code. He pointed out that
there were still a few issues, but that appears to be the way he's
thinking.
http://tinyurl.com/24h68
Another config task
Dan pointed out that the current config scheme relies rather heavily on
flags set in the current perl install, which isn't ideal. He suggested
that people might like to look into making Parrot's config step rather
less Perl dependent and pointed at things like 'metaconfig'.
http://tinyurl.com/3xn29
Problems with the Perl 6 Compiler
Allison Randal noted that languages/perl6/ was failing all its tests.
The issue arose because the Perl 6 test module inherits from
Parrot::Test, and Parrot::Test's behaviour got changed recently. She
wondered why the changes had been made. After some discussion, Allison
provided a patch to make things work with the new Test module.
http://tinyurl.com/2j3kp
IMCC temp optimizations...
Dan is possibly the only person using Parrot in a commercial (mission
critical?) environment, using a compiler from a legacy language to
Parrot. He's currently experiencing problems with IMCC taking a long
time to compile the PIR generated by his compiler. Apparently it's
because he's using a lot of ".local"s and temps. He and Leo discussed it
back and forth to try and find optimizations (both of IMCC and of Dan's
compiler) for the issue. (Dramatic) progress was made.
http://tinyurl.com/3f9gq
Korean character set information
Last week, Dan had wished he had access to anyone who knew anything
about Korean. This week kj provided some information. The ensuing
discussion of Unicode (again, maybe I'm going to have to extend my "I
don't cover the Unicode arguments" policy from perl6-language to the
internals list too) led Jarkko Hietaniemi to propose that Parrot's
standard character set be cuneiform, with Phaistos disc symbols for
variable names.
I think he was joking.
http://tinyurl.com/ysd9z
Containers and Values
The difference between containers and values cropped up again (for the
first time in a while actually). It got kicked off by a discussion of
what
my Joe $foo;
$foo = 12;
meant. It turns out that the "my Joe $foo;" part sets up a lexical
variable (called $foo), which points to a container (in this case a Perl
Scalar), which is constrained to contain either undef or a Joe objects.
I believe it's somewhat open as to whether "$foo = 12" is legal and, if
it isn't, when the decision about the legality of the assignment would
be made. Simon Cozens asked if the type declaration information went
with the variable or the value and the answer was 'neither', it goes
with the container that the variable points to. (Actually, that's not
quite true, variables also know what type the containers they point to
should be constrained to holding, otherwise
my Dog $spot;
my $feline = RSPCA.get_stray(Cat);
$spot := $feline
wouldn't throw an exception.) Dan noted that, right now, there is a
'muddled split' between values and containers in Parrot, which may well
come back to bite us in the future.
http://tinyurl.com/365k3
Dan Does Unicode
After struggling manfully against the slings and arrows of outrageous
text representations, Dan finally choked down the Unicode bolus and
declared Unicode the "One True Character String Encoding Set Stuff
Thingie". Not long after this he posted the shiny new, all Unicode, all
the time (ahem) strings design document.
http://tinyurl.com/22nxa
http://tinyurl.com/3dfas
Meanwhile, in perl6-language
Backticks
Last week's thread rumbled on. Larry eventually made a ruling in favour
of *not* using "`" as an operator. He mentioned that the current
backticks behaviour 'probably needs to be completely rethought anyway',
and promised to cover it in a future Apocalypse. This didn't stop Austin
Hastings speculating, but a p6l without Austin speculating wouldn't be
the p6l we all know and love.
http://tinyurl.com/ywyle
Spaces in method calls
Late last week, Abhijit A Mahabal had wondered about one of the examples
of using method calls without brackets given in Apocalypse 12. In
particular, he wondered why
$obj.method ($x + $y) + $z
was equivalent to
$obj.method(($x + $y) + $z)
rather than
$obj.method($x + $y) + $z
Larry was forthcoming with an explanation. He hoped it would be possible
to unambiguously define what is ambiguous.
http://tinyurl.com/39qvw
Returning from Rules
Luke Palmer noticed a common pattern in his grammar writing, where he
did a great deal of assigning a value to $0. He proposed that "<{
some_thing() }>" be redefined to assign the result from the block to $0.
Warnock applies.
http://tinyurl.com/26bc6
Hyper mutating methods
Matthew Walton wondered if the new "@things».method()" hyperized method
syntax also worked with mutating methods ("@things».=method()"). Answer:
Yes.
http://tinyurl.com/2kuqn
Placeholder attachment
Trey Harris asked for an explanation of placeholder attachment. Being
the gentleman he so obviously is, he even came up with a few thorny
examples. Larry answered that the rule of placeholders is simple:
Placeholders bind to the most closely surrounding closure. That is all.
If you want anything more complicated then arrow blocks ("-> $a, $b, $c
{ ... }") are what you should be using.
http://tinyurl.com/yq424
Lvalue methods
John Siracusa is after a neat way of overriding the assignment side of
an "is rw" attribute. Various proposals were batted about, including
using a different metaclass. Larry insisted that the "$obj.foo($value)"
was the wrong way of doing what "$obj.foo = $value" does in A12 and
repeated his reasoning for this. It looks like John was reasonably happy
with the 'different metaclass' approach though.
http://tinyurl.com/33tcm
Required Named Parameters Strike Back
Way back in the mists of time, John Siracusa had argued forcefully for
required named parameters; required arguments to a function (or method)
that must be supplied as pairs rather than positionally. Apocalypse 12
gave him some grist for this particular mill and he reopened the
discussion with a long recap and extension of his argument.
Larry's response is a masterpiece of conciseness:
Well, actually, we saved you last summer when we decided to make +
mean that the parameter must be named.
Discussion continued after this of course, but it was mostly concerned
with making sure things worked as expected.
http://tinyurl.com/2xdpt
Conflicting Attributes in Roles
Jonathan Lang appeared to be shooting for the most cryptic question of
the week award when he simply posted a chunk of code. Austin thought he
understood it and essayed an answer. The resulting thread generated all
sorts of interesting stuff. Which is odd, because on rereading
Jonathan's original post, I don't think the code he wrote does what he
thinks it does.
http://tinyurl.com/2fx7d
Syntax to call attributes.
Jonathan Lang also wondered what the syntax was for accessing a list
attribute from a scalar object. Answer: The same was as you would access
any other attribute of that object; by calling a sigil free method.
(Which seems to imply that you either can't have all of "$.foo", "@.foo"
and "%.foo" as attributes in the same class. Which in turn looks like
plain common sense to me.)
http://tinyurl.com/3b63e
Lower case magic methods?
Aldo Calpini worried that 'implicit' methods like "meta", "dispatch",
etc seem to break the rule that implicit things get UPPER CASE (and some
of the implicit methods have names that a programmer might want for
something else -- especially "dispatch"). Larry confessed that the names
were chosen for the very strong reason that he "hadn't thought about the
issue yet". He went on to discuss things that go into the Apocalypses as
'placeholders' -- he knows that more design is needed, and there wants
to be *something* there, but he hasn't necessarily got the name right
yet. He went on to make like a disabled octopus, offering six hands
worth of reasons for choosing different rules for naming the various
structural methods. No decision yet, but now we know that Larry's
thinking about it...
http://tinyurl.com/ytpex
Subtypes that lack methods or roles
Jonathan Lang wanted to know how to declare a "subtype of a class which
messes with the dispatching mechanism to exclude certain methods and/or
roles from it". I want to know whether the class he's subtyping is the
one that messes with the dispatching, of if the new class is the one
that does the messing. I appear not to have been the only one confused
by the question: Larry asked Jonathan for some sample code, which he
did. Only to have the underlying design criticised by Dov Wasserman and
chromatic, who argued that what Jonathan was asking how to do was
exactly the thing that Roles were designed to avoid in the first place
by pulling orthogonal behaviour out into roles and then composing them
in classes.
http://tinyurl.com/2pvay
Delegation using arrays
Austin Hastings voiced his misgivings about the A12 magic for delegating
to an array attribute. After a short discussion with Larry, a light went
on over Austin's head.
http://tinyurl.com/2bbsh
Typed undef
Austin Hastings had a few things to say about typed undefs (the things
that get made when you do "my Dog $spot"), he liked the idea of being
able to call class methods on a typed undef for instance. The thread
went on to discuss other tricks with flavoured undefs, like undefs that
contain unthrown exceptions so that, when someone finally checks their
return value, they can examine the exception to find out what it was
that failed in the first place.
http://tinyurl.com/37hoz
Universal Exports
Aaron Sherman had questions about the new export syntax discussed at the
end of A12. In particular, he wondered how he'd go about reexporting a
symbol that a module imports from somewhere else. Discussion and sketchy
design occurred in the ensuing thread.
http://tinyurl.com/2424p
MethodMaker techniques in Perl 6
John Siracusa wondered how he'd go about writing a Class::MethodMaker
equivalent in Perl6. (Personally, I'd roll my own subclass of Class and
MetaClass). Discussion followed, but I don't think they came up with any
implementation techniques yet.
http://tinyurl.com/3x8xq
Announcements, Acknowledgements, Apologies
I'm sorry, but there's no announcements this week.
If you find these summaries useful or enjoyable, please consider
contributing to the Perl Foundation to help support the development of
Perl. You might also like to send me feedback at
<mailto:[EMAIL PROTECTED]
http://donate.perl-foundation.org/ -- The Perl Foundation
http://dev.perl.org/perl6/ -- Perl 6 Development site
