Perl 6 Design Minutes for 15 October 2008

grinder on 2008-10-17T17:02:00

The Perl 6 design team met by phone on 15 October 2008. Larry, Allison, Will, Jerry, Patrick, Nicholas, and chromatic attended.

Patrick:

presented at the Pittsburgh Perl Workshop
talked into being the keynote speaker about Perl 6
gave a more direct talk on contributing to Rakudo and the test suite
I don't think I even presented a single line of PIR at the conference

Allison:

I consider that a great success

depends on the line of PIR

Patrick:

I took the MMD failures out of the regression suite temporarily
didn't want to go into the weekend with test failures
we're still waiting for a fix there, but there's no particular rush
4409 tests passing now, up slightly from last week
up slightly from even yesterday (30 more)
worked with Jerry on getting is export to work in Rakudo
he's done all the work
I just give him guidelines
worked with Stephen Weeks to refactor PGE and PCT to use hierarchical names instead of string names
we merged that
but it broke TGE and Pheme and Tcl and another language
PGE expects to use strings to match the match object classes

Allison:

I'm pretty sure I know exactly which three lines of code to patch

I sent in a patch

Patrick:

I'll fix PCT to look in the right place for those
if you don't hear that I've done it, I haven't beaten you to it
cleaned up a lot of namespace issues in Rakduo
fixed some mis-implementations
trying to get nested namespaces to work so that Perl 6 can use them
a lot of "two steps forward, one step back"
updated the progress graph program
now it shows the total size of the official test suite
the graphs themselves will change a little bit
I'll make an explanation of that
still really want to get to lexical variables or Rakudo's container value model
I'll do that this weekend, when I have uninterrupted time

Allison:

cleaning up MMD problems from after the merge
started a branch for the calling conventions
Andrew Whitworth is working on that
wrote a merging and branching guide based on my good and bad experiences
thinking about things we can do to make merging cleaner
maybe a checklist of people to talk to before you do a merge
Patrick, that TODO test you added for the Complex PMC?
I have it not segfaulting anymore
it's always returning 0 now
it creates a new PMC that hasn't been set to the right values
I think it's a bug in the old proxy code
we probably need to remove that old code and it'll fix it

Will:

working on Partcl still
have another test file parsing
hoping to use the upcoming Parrot release as a new baseline
the TGE issue is the blocker for that
trying to close all of the easy tickets I can find in Parrot

Jerry:

went to PPW last weekend
it was very productive and too short
spoke at Bug Labs about porting Parrot and Rakudo to the Bug
they ported Perl 5 just before the conference
some people were playing with it
it's an ARM processor which runs OpenEmbedded Linux
if it gets ported, we can run on other devices that size

Allison:

that's definitely in our target range of uses

Jerry:

spoke to someone about NVIDIA about using their CUDA GPU API for hyperop support
that looks promising
also made the is export trait work
includes tag sets
committed failing tests for import()
working on passing those, may have that done by the end of the call
discussed branching and merging philosophy with Allison
put up a Wiki for the Parrot Developers Conference
has questions which need answering
I'll mail the list, requesting answers
that'll give us an idea of logistical questions, including budgeting

Larry:

on specifications, various simplifications
expressions in conditionals are not special with respect to blocks that might appear in argument lists
if you say for map { ... }, it parses the way the user expects
the block that the for expects must occur where an infix is expected
that simplification seems more DWIMmy
I nailed down the operator modifying adverbs
now applied on the basis of precedence
no longer a special mechanism

Patrick:

Larry:

there will be a few FAQs from that
by and large it's a win though
going back and forth on the subject of unpacking arrays and hashes
someone noticed that the standard parser didn't actually handle those
I went to add them
discovered that the hash unpacking syntax using curlies was terribly ambiguous, especially inside a pointy block (arrow lambda)
tried to rescue that, but in the long run decided that hashes have to be unpacked with signatures, rather than curly notation
left in the square bracked notation for unpacking lists
it contributes to clarity
decided that enumeration types ought to define a .pick operator
you can do Bool.pick and it'll return true or false
thinking a lot about IRC and on the mailing lists
how do you deal nicely about serialized token lists in the order that the tokens occurred in the original?
how do those bind to signatures usefully?
given that we want match objects to behave like captures
did a lot of hacking on STD.pm
somewhat driven by adding options to viv
it has several options now
a match option attaches the old match objects to a position
a positional option
the main option is a p6 option which attempts to spit out the exact Perl 6 as originally parsed
discovered various places where I was throwing away information
that leads to trying to keep track of tokens and match interstices
also several other people are trying to use the STD output to try to produce other things
they're discovering missing information that I've been hacking in
it threw away all of the information on elsifs and elses
that made it hard
removing cyclic references which prevent analysis of the tree as a DAG
discovered a fundamental bug in the operator precedence parser
reducing if there were any terms left
it should have reduced if there were any operators left
it didn't reduce any unary operators at the end as it should have

I fixed some bugs
started looking at the DAG, but couldn't find much documentation
do you have a getting started guide, or do you say "Look at this code and think really hard?"

Larry:

it's the latter right now
but I could at least put a help message on it

a list of node types and their attributes should suffice

Larry:

that's the kind of documentation that quickly becomes a lie
right now, it generates classes on the fly as it sees nodes
it could at least tell you what it's doing as it's doing it

looked like it could be fun to play with for a weekend

Larry:

in my copious free time

Patrick:

right now, I can look up a namespace by an array, by a key, or by a string
I can create a class by a namespace, a key, or a string

Allison:

but not an array?
I thought we fixed that

Patrick:

we work around that in P6Object
get look up the namespace by the array
and then we look up the class by the namespace

Allison:

we should be symmetrical about this
your workaround should move into the Class PMC

Patrick:

you have to do that for subclass, isa

get_class

Patrick:

that specifically does not work
that's what Stephen Weeks and I tried

Allison:

submit a ticket
there's no reason they shouldn't take arrays

Patrick:

it's not a high priority at all
just a discontinuity that stunned me when I ran into it

Jerry:

there's a list of tagsets without commas in is export
STD.pm allows it
but it seems to open shift-reduce problems with multiple terms in a row

Larry:

two adverbs in a row are always considered one term
just a sequence of adverbs
whenever you parse one adverb, you look for one more

Patrick:

they're not adverbs; they're terms

Larry:

it works for terms too
whenever you see a colon pair

Patrick:

if I write 3 + :a :b
the :a :b is a term
what kind of thing is that?

Larry:

a list of pairs
or a mapping maybe

Patrick:

that's the same as saying 3 + 2
the :b would not be an adverb on the plus
inside an argument list, :a :b should act like two arguments
not a list of two pairs

Larry:

if it's coming in as a list of named parameters, it doesn't care
if it's in a listy area, it gets interpolated as a list
if you bring them in as a scalar parameter... you can't

Patrick:

is it the came as :a, :b?

Larry:

in list context, yes
when you're looking for named args, yes

Patrick:

if I had a subroutine that took a positional parameter and a named parameter of a, and I call it with :a :b

Larry:

it acts like there's a comma between them

Patrick:

is there a way to see them as a single term?

Larry:

if you put them in parentheses

Patrick:

they solve everything
if I see :a :b in an argument list, I have to translate as though they were separate arguments

Larry:

I guess
that's the way it works currently

Patrick:

if we have a term that's a list of colon pairs
we have to do some special unpacking stuff when we turn that into an argument list

Jerry:

you call .list on it?

Patrick:

this is at compile time
they have to be interpolated into the capture properly

Larry:

now that we've changed how adverbs work, they stack up nicely in any case without any special rule
if we removed the special rule, they'd work with the comma

Patrick:

getting rid of the quantifiers with the colon pair in noun?

Larry:

1 + :a :b would apply the :b to the +

Patrick:

it gets weird when dealing with it inside the argument work
we can make it work, but...

Larry:

the phrase "make it work" means there's something suboptimal
let me think about it a while longer
the way the adverbials parse now would apply them to the same operator
that was the main motivation for stacking them like that, to have multiple of them in the adverbial slot

Patrick:

I thought part of that was for the filetest operators

Larry:

that's another one
let me think about that one

Patrick:

we'll assume commas there for the time being

Larry:

sounds good

Jerry:

I changed the tests to use commas