All the Perl that's Practical to Extract and Report

Day 91: YAML, PGE, fast arrays.

autrijus on 2005-05-02T18:12:51

We now have built-in YAML parsing support with eval_yaml($str). The amazing libsyck and Haskell FFI made it possible for me to code this in under an hour. Anyone wants to try writing the emitter as well? :-)

With my newly aquired FFI-fu, PGE (the Perl 6 Rules Grammar Engine) is now embedded within Pugs. Using rx// and s/// now emits some random PIR code. Rumour has it that Patrick will release his new PGE later today; I will promptly update our Text.PGE module to the newest code, so courageous lambdacamels can try hooking it up to Pugs.Rules while the rest of us continue to get the OO model done.

Sol Foster wrote in to say that his Erastothenes Sieve implementation, while indeed 10x faster than Pugs 6.2.1 at first, gradually slows down to a crawl. This is because array storage was using a List; I have quickly swapped it into using IntMap, allowing O(1) access for smaller arrays. Now the sieve program runs much faster.

shapr reports that Simon Peyton-Jones, father of GHC, grants us the permission to derive our subcontinuation code from the A Monadic Framework for Subcontinuation paper. I'm honoured at these words from SPJ:

Actually, we're rather impressed at the sophistication of the programming techniques that Autrijus and you and the other Pugs folk are deploying. Subcontinuations, GADTs, what next?
Hmm, Template Haskell and STM, of course -- Pugs already uses them profitably. :-)

Some other notable developments today:

  • I began converting my ~/bin/ to Perl 6, and hacked in system($cmd, @list) support, as well as exec($cmd, @list).
  • Stevan is thinking about :todo attributes; I pointed to him that sub ok(Bool $cond is lazy, +$todo) {} can delay evaluation for $cond, which may be what we want when the error is untrappable.
  • scook0 continued adding Haddock API documentations to Pugs modules. It is extremely helpful -- I fixed some naming inconsistencies by looking at his comments.
  • skew (aka brandon) checked in a small perl script -- likely his first -- to post-process profile outputs from make pugs.prof.
  • Limbic_Region added an infinite list generator to examples.
  • iblech updated svnlog2graph.pl to use a better scaling.
  • rootmj added tests for lvalue functions in for loops, e.g. for @list.values -> $num is rw.
  • gaal fixed concurrency issues with parallel make smoke; he also added more #line tests; I made them pass shortly after.
  • luqui noted that the for (@list_q): sub {...} form, as used in the current test, should not work; the colon should be a comma. I fixed the parse accordingly, too.

Since I am already past deadline for $work, this week I'll focus on it instead of Pugs. I trust the lambdacamels will keep it well-fed, though, so I'm not too worried. See you tomorrow!