Slash Boxes
NOTE: use Perl; is on undef hiatus. You can read content, but you can't post it. More info will be forthcoming forthcomingly.

All the Perl that's Practical to Extract and Report

use Perl Log In

Log In

[ Create a new account ]

scrottie (4167)


My email address is Spam me harder! *moan*

Journal of scrottie (4167)

Friday June 24, 2005
06:40 PM

Pugs, PONIE, Parrot, and the Path to Perl 6

[ #25350 ]
Peter Scott, the author of Perl Medic, has been reading my book, Perl 6 Now and commenting on it. He recently completed reading it and wrote the most informed review of Perl 6 Now I've seen. But it's not shallow praise. In fact, oddities about P6N got me to thinking about P6N came about, how P6N stemmed from predictions about how Perl 6 would come about, and how those predictions were almost correct, and how Pugs surprised me for reasons other than it probably surprised you.

Perl 6 and Parrot Essentials is staunchly about Perl 6's grammar and Perl 6 on the Parrot platform. Perl 6 Now, by contrast, is about Perl 6 using the Perl 5 virtual machine as a platform. (It's also less about syntax and more about the basic concepts.) This might sound like a gag, get this, I was convinced someone was going to haul off and write a Perl 6 grammar that didn't target Parrot, at least immediately.

It seems completely logical to target Parrot, as Parrot was designed by Perl people and it's designed for Perl 6. But the bytecode format is il-documented, register allocation in the virtual machine isn't finalized, portability is still immature, and it's less complete of an implementation than Perl 5. That's the magic thing -- I thought someone was going to haul off and implement a Perl 6 grammar in Perl 5, targeting Perl 5 (either through code translation or through bytecode generation). I thought we'd see a Perl 6 implementation overnight that used Perl6::Rules and Coro, among other things (most of them documented in P6N) as the actual basis of implementation. I was psyching myself up to do a lot of the work, too. Someone more familiar with the Perl 6 grammar than myself just had to jump in and start writing that parser. (Actually, there was a Perl 6 grammar in Perl 5 included with Parrot, but it was languishing, and I was just starting to try to pick it up.)

I tried to get into Parrot development briefly, but the more I got into it, the more I started to feel that the Perl 5 codebase wasn't so bad. Complexity of Parrot was running wild. The main point of rewritting the VM was to make the code accessible to developers to avoid the problem with Perl 5, where only a small handful of people understood the code. But fewer people understood Parrot than the Perl 5 VM. And from the beginning, it was emphasised that Perl 6 was not to be tied to any one implementation. When I wrote P6N, I was dubious of Parrot, and I'm still feeling cautions (I side with Joel on Software regarding his never rewrite software article).

I was kind of expecting the PONIE effort to overhaul Perl 5's guts to the point where they were clean and usable. PONIE has already factored out sv, av, hv, and gv usage and replaced them with Parrot PMCs. That's a lot of the mess in the Perl 5 interpreter, gone. Granted, the avs, hvs, etc, were just encapsulated in PMCs, but this paved the way to moving all of the datatype management and calls to sv_upgrade and all of that into nicely encapsulated objects. Then I was expecting someone would write a Perl 6-to-Perl 5 translator, or just a generic Perl 6 grammar in Perl 5, leaving me in a good position to jump in and do a Perl 5 "B" backend. And then a Parrot backend would be used for code generation, and the whole ball of wax could be turned into a binary, and then Perl 6 would become self-hosting. I expected that things like Perl6::Rules and Coro would actually become the de facto implementations of these Perl 6 features for several years, until Parrot really became competitive.

But no. Autrijus just had to write the grammar in Haskell and reimplement the whole virtual machine at the same time. The plan is still to some day rewrite the Haskell in Perl 6, create a Parrot back-end, generate a binary of itself, and make it self hosting. This plan reads just as I expected would happen except s/Haskell/Perl 5/.

Right now I'm throwing my initial stones at a Perl 5 "B" VM backend for Pugs. Or I hope that's what I'm throwing stones at. First step, learn Haskell (step zero, forget all of my false starts with Parrot, lang/perl5, etc). Step two, fix up B::Generate so it understands lexicals, so I can do proper Perl 5 code generation. Step three, get B::Generate running under Inline::Pugs so the Pugs universe and Perl 5 universe exist at the same point and time. Then learn how to navigate Pugs bytecode. The rest is glue.

In summary, Pugs is a good thing, Parrot is still useful for code generation and bootstrapping even though I'm glad people aren't waiting for it to be finished and use it as the primary VM, Perl 6 Now is still as relavent as I thought it would be half a year later though it took a lot longer getting any press than I had hoped thus wasting precious relavency, bootstrapping is a fun and interesting CS problem, and a language doesn't have to be self hosting to be cool and useful.

The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
More | Login | Reply
Loading... please wait.