In the spirit of “release early, release often”, last month’s Rakudo compiler release was the first one to come with very basic support for running on the JVM. It supported a small number of the features that Rakudo on Parrot does. Of note, it could pass the sanity tests, a small test suite we keep in the Rakudo repository. The sanity tests ensure that the compiler is functional enough to compile Test.pm, which is a prerequisite for running the specification tests. Essentially, they make sure that we can crawl our way to the starting line, where the real race begins. :-)
So, since that release, we’ve been working hard on getting Rakudo on JVM able to pass more of the specification test suite, gradually increasing its capabilities, hunting bugs, adding missing bits of infrastructure and guts code in order to get features working, and so forth. Progress has been rather swift, and today’s automated test run by Coke++ shows that Rakudo on JVM is now passing 92% of the tests that Rakudo on Parrot does!
So what does this number really mean? It means that Rakudo passes 92% of the individual test cases on the JVM that it does on Parrot. To put that into perspective, that’s more tests than either of Niecza (82%) or Pugs (36%), meaning that Rakudo on JVM is now the second most spectest-passing Perl 6 implementation. Not bad, to say around 8 months ago not a single line of implementation code related to Rakudo JVM targeting had been written.
However, the raw number should be taken with a pinch of salt. Here’s why. The test suite is not especially uniform in terms of tests it dedicates to each feature. For example, there are hundreds of passing tests dedicated to providing good coverage of Unicode character property matching in regexes, and trigonometry easily has many hundreds too. By contrast, the slurp function has more like dozens of tests – since you can comfortably cover it that way. However, these are failing still, and I’m pretty sure that more Perl 6 users depend on slurp that Unicode character properties and trig.
Anyway, my aim is that we’ll be some way past the 95% mark in time for the July compiler release. And yes, I’ll make sure slurp gets fixed before then! :-) I suspect the last couple of percent will be the most tedious, but it feels good to be reaching towards them.
In the coming weeks, I expect the focus to start shifting from the compiler to the ecosystem: getting Panda working with Rakudo on JVM, and starting to run the module tests and work through the issues we find there. Calling Java libraries and work on parallelism are also planned, not to mention digging in to optimization work, since this initial phase has very much been focused on “get things working”. All in all, plenty to be working on, and plenty to look forward to.
Finally, a week or so ago I spent a couple of hours being interviewed by Nikos Vaggalis for Josette Garcia’s blog. It was a very wide-ranging interview and I managed to spout sufficiently in response to the questions that it’s being published over three posts in the coming weeks. You can find the first part of the interview here; enjoy!