Re: [MacPerl] Byte Compiler for MacPerl

[Matthew Langford <langfml [at] eng [dot] auburn [dot] edu>]

On Wed, 12 Aug 1998, Matthias Ulrich Neeracher wrote:

> I still think the C/C++ backend option is overrated. What is missing in this
> debate is that the C/C++ backend still needs at least 80% of the Perl runtime,
> if not 100%, to run. Thus, you gain at most a minimal improvement in speed
> vs. byte code. Byte code should be a massive improvement in speed vs. source
> code and a significant improvement in size vs. C/C++.

You say 80% of the runtime is needed, but isn't the crucial question what 
execution-time fraction of your code depends on the runtime?  Perhaps
only a small proportion of code actually requires the full runtime.  This 
makes bloated executables, but you may still see great speedups if
the translated portion dominates running time.  (In other word, if a key 
loop wins, you may still win, and win big.)  At that point, if you are
still unhappy with speed, you might be happy in simply rewriting
the bit of Perl that required dragging along the whole interpreter.

And (if you know C/C++ pretty well, and if you can follow the generated
code--gulp) the translation may generate a vast amount of insight into how
to speed up the program (even in Perl).  On the other hand, the bytecode
will probably generate a fixed speedup; identifying and tackling
bottlenecks would be left to other tools. 

Is the bytecode more efficient than Java's VM stuff?  Is it
platform-dependent?  Should I be reading docs somewhere for these answers? 
(In Greek, you could tell that the presumed answer to this question was
yes.) 


--
MattLangford 



***** Want to unsubscribe from this list?
***** Send mail with body "unsubscribe" to mac-perl-request@iis.ee.ethz.ch