Re: Optional Separate Programs for Interpreter Passes

2000-08-31 Thread Ken Fox

Fisher Mark wrote:
 The rest of us with our TVs, VCRs, and so on have only compiled
 code in our devices.

I'd buy a microwave that resets to 'JAPH' after a power failure.
Maybe. ;)

- Ken



Re: Optional Separate Programs for Interpreter Passes

2000-08-30 Thread Bradley M. Kuhn

Dan Sugalski wrote:
 At 12:58 PM 8/29/00 -0500, Fisher Mark wrote:
 Although Perl interpretation is divided into several passes (parser/lexer,
 optimizer, tree/bytecode runner), all these passes are grouped together in
 one binary.  Under some memory-constrained conditions, it could be better if
 each pass ran as its own program, passing the transformed data onto the next
 pass similarly to the way compilers usually work. 

 
 This is a good idea, and I've had fuzzy thoughts along this line (more or 
 less) myself. There's a proto-RFC in the works.

For the world of the JVM port, it's imperative that hooks be provided so
that the front-end can be run independently, and a different back-end can be
run (to emit bytecode of some sort).

This is done in perl5 by STOP blocks, and the mechanism is rather hokey.  I
don't care much myself *how* it is done here, but something non-hokey would
be good.  ;)


-- 
Bradley M. Kuhn  -  http://www.ebb.org/bkuhn

 PGP signature


RE: Optional Separate Programs for Interpreter Passes

2000-08-30 Thread Fisher Mark

Bradley M. Kuhn wrote:
 For the world of the JVM port, it's imperative that hooks be 
 provided so
 that the front-end can be run independently, and a different 
 back-end can be
 run (to emit bytecode of some sort).

All this also dovetails nicely with the mass-market world of embedded
devices, where almost all copies of the device have only the compiled code.
Only the developers ever have the source, the parser/lexer, etc. in a
device.  The rest of us with our TVs, VCRs, and so on have only compiled
code in our devices.  And, I expect to see more and more microcontrollers
come with a JVM (or have a JVM readily available).  So a perl that can
compile down to JVM bytecode would be a big win in the embedded world.

Mark Leighton FisherThomson Consumer Electronics
[EMAIL PROTECTED] Indianapolis, IN, USA
"Display some adaptability."  -- Doug Shaftoe, _Cryptonomicon_




Optional Separate Programs for Interpreter Passes

2000-08-29 Thread Fisher Mark

Although Perl interpretation is divided into several passes (parser/lexer,
optimizer, tree/bytecode runner), all these passes are grouped together in
one binary.  Under some memory-constrained conditions, it could be better if
each pass ran as its own program, passing the transformed data onto the next
pass similarly to the way compilers usually work.  This would be an
advantage in embedded systems where there might be a great deal of ROM
(perfect for storing pass programs) but not as much RAM (so you can't load
the whole interpreter into RAM at once).  This should be an option at perl
creation time, as most non-embedded systems would not benefit from splitting
the interpreter into separate programs.

Thoughts, anyone?

Mark Leighton FisherThomson Consumer Electronics
[EMAIL PROTECTED] Indianapolis, IN, USA
"Display some adaptability."  -- Doug Shaftoe, _Cryptonomicon_





Re: Optional Separate Programs for Interpreter Passes

2000-08-29 Thread Dan Sugalski

At 12:58 PM 8/29/00 -0500, Fisher Mark wrote:
Although Perl interpretation is divided into several passes (parser/lexer,
optimizer, tree/bytecode runner), all these passes are grouped together in
one binary.  Under some memory-constrained conditions, it could be better if
each pass ran as its own program, passing the transformed data onto the next
pass similarly to the way compilers usually work.  This would be an
advantage in embedded systems where there might be a great deal of ROM
(perfect for storing pass programs) but not as much RAM (so you can't load
the whole interpreter into RAM at once).  This should be an option at perl
creation time, as most non-embedded systems would not benefit from splitting
the interpreter into separate programs.

This is a good idea, and I've had fuzzy thoughts along this line (more or 
less) myself. There's a proto-RFC in the works.

Dan

--"it's like this"---
Dan Sugalski  even samurai
[EMAIL PROTECTED] have teddy bears and even
  teddy bears get drunk