On Fri, Sep 15, 2000 at 12:53:29PM -0400, Dan Sugalski wrote:
> The only reason I can see nice winning over fast is if nice brings in whole
> new concepts to the language. (Like, say, matrix ops or Damian's currying stuf)
Well, taking the idea of writing the parser in Perl, let's have a look
at how practical it actually is.
We need to be able to parse that Perl, so we do have a bootstrapping problem.
We have a number of choices, and I've only just seen the third and I like it.
A lot.
1) Write a simple Perl parser in C which can parse the main parser written in
Perl. Now, parsing Perl in Perl is *hard*. I know, I've tried it. There
*isn't* a way to write a Perl parser so that it can be parsed by a simpler
parser than one which parses the whole of Perl. (Can you follow that?)
2) Distribute the parser in bytecode. Sounds nice, but where do you get the
bytecode from? Hand-written? Rather you than me, pal.
3) Think emacs. Write a "safe" parser in C, which gets linked into miniperl.
That reads in the complete Perl parser and spits out bytecode. The bytecode
gets read by perl6. You have the option of using it to create a modified
version of itself. OK, you're writing an entire Perl parser in C, but you've
still got the flexibility of a Perl version. Hmm. Nice. And bytecode will be
fast, won't it?
> That, I think, is a bit of excessive pessimism, though I'm rather glad it's
> here.
Dan, any time you want me to be a miserable, pessimistic bastard, just shout.
How are we doing on deadlines, again?
--
God damn it, it is not acceptable, it is NOT FUCKING ACCEPTABLE for a
*single* post that is from a *person* talking to other *people* to be deleted,
to be dropped on the uncaring floor to make room for machine generated spew.
Period. - Russ Allbery, http://www.slacker.com/rant.html
