Hacker News Reader

Comment by kanatohodets | Hacker News Reader

Judging by the ample supply of version number rimshots, some clarification might be helpful: this isn't a Perl release, nor is it a new Perl version. It's the name of a general project in the Perl community to try to decouple Perl 5 the language from perl the runtime, and hopefully achieve interoperability between Perl 5 and Perl 6 code in the process. Perl 5 and 6 are distinct languages in the same family, much like (for example) Common Lisp vs Racket Scheme, which is why interop is an interesting and worthwhile goal.

There are some really cool things happening in the specific projects on the linked page, if you're into compilers and VMs.

View on HN

It would be nice if it actually said this on the page. I had to poke around for several minutes to figure out what this project actually was, and I'm a Perl person myself.

I'm not sure I get the motivation. Isn't Perl a Unix-y language? You can get "interop" by using pipes or other IPC.

This seems like a heavy, Java-esque solution. 50 lines of Perl does a lot, let alone 1000. Not sure I want to see 20,000 lines of mixed Perl 5 and Perl 6.

Well, the goal is to make sure that the plentiful modules on the CPAN (all in Perl 5) are usable for everybody involved. You're right, directly mixing Perl 5 and Perl 6 isn't likely to be a common usage - at least, doesn't seem that way to me.

The p2 goal is to use as much non-conflicting p6 syntax within p5, and mark conflicting parts in special lexically scoped syntax blocks. But data and methods should be shared, the ast, compiler, vm, threads and event model ditto.

Syntax blocks also allow easier ffi and sql interaction, { use syntax "C"; c declarations ... } being a nice FFI language, compared to "extern { c decls; ... }", which is also nice.

We want to use efficient signatures, methods, classes, tasks, coros, promises, hyper operators, types, lazy lists and so in perl, regardless if you call it p5, p6 or p2.

It's the name of a general project in the Perl community to try to decouple Perl 5 the language from perl the runtime...

This is literally impossible: http://www.perlmonks.org/?node_id=663393

Every few months this comes up:

http://www.modernperlbooks.com/mt/2009/08/on-parsing-perl-5....

I posted the link mostly in jest.

That being said, I still contend that a Perl 5 parser is still inextricably separable from the runtime. Any parser that wants to be fully compatible with Perl 5 will need to be able to also run Perl 5 code. That is, the parser will need to be coupled with the runtime.

A consequence of the parser needing to be able to execute arbitrary code is that the parser is undecidable. This follows from the halting problem, which is the point of the article I posted and is also pointed out in the one you posted.

As a sidenote, I find the distinction between "static" and "dynamic" parsing to be rather silly. The only people I've ever heard make that distinction are Perl apologetics. No need to be so defensive: parsing is allowed to be ambiguous, and it is allowed to be undecidable -- but it's not considered ideal.

Dynamic and static parsing is different to just calling eval within the parser.

"Dynamic parsing problems" such as solving the halting problem, by e.g. changing the prototypes or eval'ing code in BEGIN blocks are just small practical problems. In practice the parser only has to be GC-safe, which needs a few days work.

   | BEGIN b:block           { p2_eval(P, b) }

On the other hand the "dynamic vs static parsing" on perl11.org talks about compiling and extending parser rules at run-time, which needs a fast vm to do this. To be able to support macros. Not perl5, perl6 or nqp. Efficiency should be near the C level, but you shouldn't be forced to go the rakudo way with its insane bootstrapping and serialization overhead just to support BEGIN blocks in its stdlib and support run-time parsing. The JVM or .NET could do that, but I don't buy the overhead, esp. when you look at fast parser combinators in lua, lisp or look at maru, which is basically a jitting parser, bootstrapping itself.

The author

Of the two types of macros I'm familiar with, neither demands a "dynamic" parser. C-style macros (conceptually) work as a text pre-processor prior to the compiler's lexing phase. Lisp-style macros work on the syntax tree after the parse phase has finished.

In some Lisps, such as Common Lisp, macros are simply functions from syntax tree to syntax tree. Since these execute at compile time, you can still run into the same types of problems you get with Perl's BEGIN. Scheme's syntax-rules macros are more limited: they cannot execute arbitrary code, but they can do pattern matching and substitution on syntax trees. Despite this limitation, syntax-rules covers the vast majority of macro use cases. The term rewriting systems in Haskell and Cat are reasonably similar.

My point to all of this is that macros are no justification for "dynamic" parsing. The only distinction I can find of the two kinds of parsing is that a "dynamic" parser sometimes needs to execute bits of the program before it is able to progress. This is not necessary for macros!

As far as parser combinators go, I conjecture that they will not be powerful enough to deal completely with Perl's grammar. In general, they're limited to LL(k) grammars. Still, you may be able to jump through some hoops to get what you want out of them.

Reply on news.ycombinator.com