The Perl 6 design team met by phone on 26 September 2007. Larry, Allison, Patrick, Jerry, Will, Jesse, and chromatic attended.
c:
- looking at fixing the parallel build problem with Parrot
Jerry:
- Jeff Horwitz fixed that as of five minutes ago
c:
- then I'm not working on that
- talked to David Maxwell of Coverity
- he, Paul Cochrane, and I should be able to get nightly builds scanned soon
- also talked to David Fetter about using the Pg buildfarm system with Parrot
Allison:
- hacking away on the object model
- getting very close
- need to add one feature for Patrick
- then I need to implement the full PMC PDD inheritance part
- shouldn't take very long
- just setting up the appropriate places to create a proxy PMC
- my goal is to merge the branch into the mainline around the end of the month
Jesse:
Allison:
- the events PDD and the exceptions system
Patrick:
- things are going well
- I had a class to teach last week and was busy all then
- working with Allison on the PGE changes to work with PDD 15
- it should be pretty straightforward
- there are a lot of changes to make all at once, but nothing appears to onerous
- working on the AST documentation for the Parrot compiler tools
- deadline of October 1 for that
- all of this indirectly ties into Perl 6
- we have to clean up things to progress
- I'll probably fix up and complete some things as part of the documentation
- hope to get that done within the week or so
- that'll finish up NQP, and then I'll be back on the Perl 6 compiler
- it's blocked waiting on NQP
- that's blocked waiting on the compiler tools
Larry:
- Wolfgang Laun, Damian, and I have corresponded over his review of the Synopses
- mostly still working on my auto-lexer to do longest token matching
- it works in prototype form with small grammars
- now trying to bootstrap it up to the complete standard grammar
- running into all sorts of places where my prototype is insufficient
- that's the way forward
- it only gathers the data it needs to generate a DFA match and passes it up the tree
- emulates that by looking at the constant strings it gets
- looks like it'll probably do what I expect
- already caching this information on a per-language, per-rule basis
- if you extend the grammar, it automatically generates a new lexer for you
- that's the "auto" part of the auto-lexer
- does that the first time you use a new rule
Patrick:
- I'm eagerly looking forward to the looking at it
Larry:
- should iron out some of the ideas
- even if the manifestation isn't how it ends up
- still using Perl 6 lazy lists to manage all the continuations
- you've never seen so many
- every time you see a sequence of tokens, you get a list of matches
- it gets pretty deep
- semantically it looks like it actually works
- eventually we can do the ratchet optimization at one-to-one correspondencese
- although a completely extensible grammar can't necessarily assume that its subrules will only ever return only one result
- even if they explicitly use ratcheting
- the user may eventually add more rules that do want to backtrack
- a full continuation model would allow that
- throwing information away in the continuation model might not be right
Jerry:
- we found a contact at Amazon web services
- I can forward it if you like
c:
- I think we should pursue the AWS question
- even if it's not exactly what we need right now, it might be useful to get test results back pretty quickly
Patrick:
- it should be cheap to experiment
Jesse:
- I'm happy to underwrite that
c:
- let's find someone on channel who can experiment with that
Jerry:
- Colin Kuskie is working with Eric Wilhelm with parallelizing the tests
Jesse:
- we've seen speedups of about 40% on a dual-core box with our stuff
Jerry:
- I can give Parrot committers access to my quad-core box if they want it
