Report from the Perl 6 Hackathon in Copenhagen

During and the "Open Source Days 2010" in Copenhagen there was a Perl 6 track, and also a hackathon that stretched out three more days after the conference.

I arrived on Saturday afternoon (which happend to be the last day of the conference) and thus missed all of the talks and the more public part of the hackathon.

Module Loading

On Sunday we discussed module loading extensively. Martin Berends had prepared that topic, and drove the effort by asking the right questions.

Discussions

Modules are complicated beasts in Perl 6. They not only have a name and a version as in Perl 5, but they can also have so-called authorities, which handle the case when different programmers write modules with the same name. Also unlike Perl 5 a module can be installed with different versions, and the programmer can either request a particular version or the highest version available.

Also they have Unicode and case sensitive names, which Perl 6 must support even if the underlying file system does not provide both features.

Although it doesn't really sound like it, this combination of requirements make it incredibly hard to implement a module loader that is both correct and efficient. Previously all discussions on the mailing lists and IRC channels ended in huge threads full of bikeshedding and even more feature ideas.

We were well aware of these requirements before the hackathon, and also aware of the fact that we have no chance of implementing all that in a reasonable time frame. So the first step was to decide what subset of features to implement, and how that could be achieved.

We decided not to implement Unicode emulation/mapping, and partially ignore authorities for now. Still we leave some room for wriggeling in the mapping from module name to file name: a module My::Module can live in the file My/Module.pm or in My/Module.ignored.pm. That way several modules with the same name but different versions or authorities can be stored in the same module repository.

This also implies that the Perl 6 compiler has to actually read and parse those files to find out which file to load. In a later stage the compiler will write a cache file to store the mapping from file name to module name(s), verisons, authorities, a timestamp and probably also dependencies plus timestamps (in order to know when to recompile a module).

Once this caching mechanism is implemented, it solves the Unicode problem mostly for free, because it will store the full Unicode module name and ASCII file name in the cache, and makes it available to the module locater.

After reaching an initial consensus, the discussion went on to other topics: can we use the existing CPAN infrastructure to actually distribute our modules? The conclusion was that we most likely can use (with very little patching) the PAUSE and mirroring infrastructure, but we likely have to write our own indexing and searching facilities and also a completely separate module installer.

Implementation

I wrote a very simplistic prototype of a module locater in Perl 5, which Jonathan mostly translated to NQP on Monday, and actually plugged it into Rakudo's module loading facility on Tuesday.

So now you can actually have multi modules with different versions, and load the one with the highest version - requesting a particular version is not yet implemented, but hopefully now a mere SMOP.

Other matters

Martin Berends wrote up the results of our module discussions, and spent the rest of his hacking time on a foreign function interface.

Not everybody worked full-time on module things; baest worked on some builtin functions and (most of the time) number handling. Carl Mäsak spent some time on tardis, his time travelling debugger. He also spent quite some effort on getting enums back into Rakudo, with occasional assistance from Jonathan.

Many of us fixed some bugs that popped up, and also found new bugs

RT Queue

The perl6 RT queue has grown to such a size that it was both scary and hard to use at all. On Sunday it peaked around 725 open tickets.

I did rough sweep over most of the tickets, identifying those that were either fixed (but not yet closed), superseded by changes to the specification or actually not bugs at all (spam and false reports), or could be closed by simple changes to Rakudo.

That way I closed about 80 tickets, and identified another 40 or so tickets that are actually fixed, but need test coverage before being closed.

Personal notes

This was the first Perlish event to which I have traveled, and it was just awesome. I met friends in "meat space" that I previously only (or mostly) knew over the Internet, and found them to be just as I had experienced them from the distance: friendly, witty, full of good jokes and bad puns, curious, relaxed and all in all a very enjoyable company.

We enjoyed the hospitality of Jonas Brømsø Nielsen and the OSD folks who did everything to make our stay pleasant: prepare the arrival with maps and tips about the public transport system, booking and funding (!) our accommodation, inviting us to lunch or dinner now and then, and being available whenever questions or problems arose. Thank you!. The March release of Rakudo will be named Copenhagen for very good reasons.