The Perl 6 design team met by phone on 14 October 2009. Larry, Allison, Patrick, and chromatic attended.
Patrick:
- have a question
- I'm not happy with the
:lang attribute in regexes
- I'll explore in my own direction
- Larry can complain about that when he gets back
Larry:
- I wouldn't expect anything less
Patrick:
- worked more on nqp_rx
- PAST- and POST-based
- going exceedingly well
- I have almost every feature that PGE provides
- what's left is minor
- either we don't need them and/or they're easy to add when necessary
- the new engine passes about 3/4 of PGE's test suite
- the rest is minor
- specifying characters by number or Unicode name, for example,
\x and \c
- today's task is to make the new engine self-hosting
- the regex parser should be able to parse itself
- that shouldn't be hard to do at all
- that'll also speed things up quite a bit
- it's currently two or three times slower than PGE when running the tests
- that's because the regex parser is slower
- the code that does the matches is much faster
- that's a nice tradeoff; we generally parse regexes once and run them a lot
- adding features to the new engine has been incredibly easy
- many of them tend to be five or six lines at most
- it's a bunch of short code
- I'll continue working on that for the next week
- I expect to write the new version of NQP in the next couple of days
Allison:
- working on the pcc_reapply branch of Parrot
- lots of fixes there
- went down from 82 failing tests on the corevm target to 18 failing tests on the full Parrot test suite
- PGE now compiles and runs just fine on the branch
- still working on the last few failures
- started doing language testing as well
- Cardinal and another language pass on the branch with no problems
- have CallSignature optimizations ready to merge in
- waiting for one feature from chromatic on that
- also, quantum computation is a lot of fun
c:
- refactored CallSignature PMC to optimize it
- one last change to make
- where did the 12,000 passing Rakudo tests come from?
Patrick:
- mostly trig, transcendental, and math functions
- they're automatically generated for
Int, Num, and Complex
- it's a nice jump on the graph
- we need to put out a nice message about what happened
- I'll talk more about the percentage of the suite passing than the number of passing tests
- the argument "You can add thousands of passing tests and not make progress" makes some sense
- that's why I'll talk about percentage of the suite from now on
- I'll report for Jonathan too
- he's redoing parameter binding in a branch
- currently, that's fairly slow in Rakudo
- after we call a subroutine, we perform the type checks and throw exceptions if necessary
- we do binding from incoming parameters to lexicals there too
- that all required a lot of method calls
- every subroutine call in Rakudo caused at least four or five Parrot method calls
- he's redone parameter binding
- primarily in C
- he's created a set of dynops to make it much quicker
- Rakudo's Subroutine PMC type can do the type checking before invocation now
- he's seeing a significant performance improvement there
- the pcc_reapply branch in Parrot will also simplify what he's doing there
- he expects to merge before the release next week
- it's been outstanding work
- do we have an ETA for the branch landing?
Allison:
- very shortly after 1.7
- within minutes or hours
Patrick:
- that fits our planning well
- Parrot releases 1.7 next Tuesday
- Jonathan will probably merge his branch in then
- his branch will use Parrot's existing calling conventions
- if someone tries to build Rakudo today or Jonathan's branch against pcc_reapply, Rakudo will likely fail
- we're fine with that
- don't consider it a blocker for Parrot
- we'll adapt to whatever you do
- I expect that Rakudo's trunk will run against the 1.7 release until 1.8
- we have two major projects
- one is converting our argument handling to Parrot's new PCC
- the other is completely replacing Rakudo's grammar to use the new NQP
- almost all of our work will take place in branches, not trunk
- we'll merge back to trunk when we have everything working
