Perl 6 Design Minutes for 03 September 2008

brian_d_foy on 2008-09-25T19:29:00

The Perl 6 design team met by phone on 03 September 2008. Larry, Allison, Patrick, Jerry, Jesse, Nicholas, and chromatic attended.

Larry:

I was at the Lightweight Languages Conference in Tokyo last week
about a thousand people there
a large Ruby presence, as well as some of the other "scripting" languages
gave another variant on my talk
seemed well-received
mostly just vacationing and touring around otherwise
did some work getting STD and the test suite to agree with each other
now have symbol table checking in STD
a global scale, not lexical
can tell when you call an undefined function
found surprisingly few difficulties with the existing parsing
that makes me happy
hanging out on IRC and email, doing a little bit of spec work
the standard parser can parse itself in under 30 seconds
I have other ideas to make it faster
not sure I want to implement them yet
it might adversely impact other work with installing a real LTM
they remain ideas for the moment

Patrick:

most of you saw the announcement about Paula
we celebrated last Friday
mostly answering questions and debugging problems on Rakudo
expect to be back full-time on hacking tomorrow
Jonathan's out on vacation this week
Moritz and Carl and other people have been continuing to make progress
I've been responding to them
not a lot has happened beyond testing improvement
the primary coders are on vacation or busy with other things

Allison:

another nice quiet and productive week
we have a new, much cleaner dispatch route in Parrot
it uses signature objects, which are like Perl 6's Captures
everything you need to invoke a sub is all in one place now, not spread out over lots of calls and varargs and unpacking varargs and...

you realize that code has always been haunted, right?

Allison:

I'm ripping out some of the old code, but some of it will stick around
we can make another branch to clean it up
the branch has some problems in TGE
I probably haven't caught all old dispatch instances yet
we're ready to launch the new Parrot website
doing some work on that
almost ready to launch the new Parrot mailing lists too
just need to ask Ask or Robert for subscriber lists

started a couple of arguments online
need to work on the "How to Help the Patchmonster" document
worked on vtable cleanup in Parrot, had MMD problems, may have to wait a bit
otherwise applying patches
should have more time in the next couple of weeks

Nicholas:

I love parallel tests in blead, and I'm missing them in 5.10

Jesse:

I've been working and traveling for the past two weeks
also been trying to goad chromatic to post suggestions to p5p

Allison:

which seems to have worked

Jesse:

is there anything I can do to help you?

Patrick:

I just need time and energy
so maybe saying "Get back to work!"

Jesse:

I can do that

Jerry:

haven't had any commits recently
helping lots of people with implementation questions
talked to Evan Phoenix about the Rubinius VM
he had some questions about Parrot's architecture and design decisions
good chat
we started talking about bytecode and AST translation
they're lacking a well-defined AST now
just S-expressions
we'll chat more frequently
there are only so many people interested in implementing VMs
we should all get together
still talking to Microsoft
they want to know what an ideal testing environment would look like in their labs
I need to figure that out
it could be good for Perl 6, Parrot, and Perl 5

Patrick:

is that worth taking to the mailing list?
Allison mentioned something about old MMD and new MMD
what's the difference from a PIR perspective?

Allison:

absolutely none

Patrick:

there should be no difference in PGE and TGE then

Allison:

there are a few places internally which are C code
if there's a direct C call into the old MMD system, bam!
I've been replacing them with the new C-style calls

Patrick:

it's just that PGE is exposing the bugs
it's not a source

Allison:

that's why it's so perplexing
I haven't been able to trace it back to a particular call

Patrick:

I'll look at it too
I'm happy to see Parrot taking this new model

Allison:

we can avoid a lot of indirection

Larry:

there have been discussions of different levels of test harness support
we need to straighten out the namespace
"What does use Test; mean?"
it has overloaded meanings for many people
should that be the minimal thing, and everything else be larger
or should the minimal test thing be built into the language

Jesse:

that would be fascinating

Jerry:

and everything's multi and can be overridden?

Allison:

ok() and is() are easy to write
easy to include in the language

Larry:

just in the standard Prelude
it's like POD's play to bring documentation into the language to make it easy
it could be a module that Prelude brings in

Patrick:

just a module that implements these?
they print what they print now?
I'm happy with that
I'd like to define the minimal set of functions that the test suite can rely on

Larry:

or the fancy set should be the default set

Patrick:

I can argue it that way too
but that seems to be the wrong way
I'm happy for any clarity the Perl 6 designers throw in there

Jerry:

for a Perl 6 implementation to be a real implementation, it uses the core tests only
and builds on top of those
any other fancy test library is out of the core and builds on Perl 6?

Patrick:

the test suite functions are part of the official test suite

Nicholas:

that scared me slightly in how the perl 5 core bootstraps itself to testing, and trusting/testing its test system
the very early tests run inline
you can't be sure that require works
the next step uses a file named test.pl
doesn't use ++ for example
everything beyond the core ops tests are free to use Test::More
written in proper grown-up Perl
uses constructs which may have failed earlier

Larry:

we still have two of those things
the third level is potentially possible
the core definition of standard Perl 6 could have a fancier test module
the test prelude we're talking about is the equivalent of that test.pl

and if you're implementing this, you can hardcode these as special cases in the compiler
and then you can rewrite them in your Prelude when you have that working

Patrick:

we can get rid of the line use Test; in our tests now

Larry:

that's part of the motivation

Patrick:

we don't have to worry about getting our namespaces working first
I'm in favor of that
that would have sped up a lot of stuff in Rakudo development

Larry:

the only other major argument against it is the notion that the function names are Huffman-coded for testing
not necessarily general use in a language
a user might want an ok() function that does something else
that's okay, if we do the overriding properly
but there's a conflict with the current spec because there are two different meanings of is
I can change the meaning of is Dog;> within a class
it can be specific, so it only parses on is followed by a type name

Jerry:

or it's only a single parameter

Larry:

multidispatch could handle that, but it's declaration
declarators need compile-time resolution
it needs compile-time distinguishing
the other is has multiple arguments
this one never does

Patrick:

it hasn't been a problem in Rakudo yet

Larry:

it's doable
the question is if it's too much of a hack
or if there's a better way

Patrick:

that depends on what the meaning of "is" is
what kind of equivalence does it do?

Larry:

in the header, it's either a trait
or if it's a type name, it turns into an isa
there's some argument of some kind

Patrick:

what if I say is $a, 3?

Larry:

or we can do is_eq, is_eqv
or just ok

but you do lose diagnostic messages

Larry:

unless it's a macro
but that could be too clever

what if someone thinks is is infix?
equivalent to == or eq?

Nicholas:

or is can do smart match
you have smart match in Perl 6