Precompiled perl...

[Eyck]

Hi,

Most not-completely-compiled languages provide a way of caching compilation - through the use of pre-compiled libraries/programs/stuff ( lisp, python, java etc use such solutions )

Why doesn't perl pre-compile often-used things?

UPDATE

The way I see it, "Bytecode" support in perl is just a toy, and a way to show off the flexibility of the language.

What is missing is in-perl support for parse trees, something that would bypass whole compilation phase, and that would include libraries support.

This would have quite remarkable effect on perl

• REALLY improved startup time for apps that use lot of libraries
• for sysadmins - reduced cost of perl in cron
• lower resources required for running perl, thus, perl on embeded machines starts making sense
• PAR - such bytecode support would turn ingenious PAR into relatively simple excercise. (OTOH, experiences with PAR suggest that it might not be that easy..)

What bugs me, is the fact that when perl was being created, such pre-compiled bytecode support was en-vogue, lisp did it etc.. And since perl internally uses bytecode/parse tree, it would be trivial to add. What is the reason for that?

[borisz]

It is possible to get this precompiled thing already. It is called bytecode.

# build a precompiled bytecode script named xyz
perl -MO=Bytecode,-H,-oxyz -e 'print "moin\n"' 

# the script can be called with this line
perl xyz

[Eyck]

Yeah, but what about pre-compiled libraries,

What I would like to achieve is to have all /usr/share/perl5 pre-compiled, this could be very nifty..

[diotalevi]

Not really. Other people have reported that it tends to be slightly more expensive to load a perl program from compiled form than from original source code. I wouldn't spend too much time worrying about this problem. Things like mod_perl exist so that the compilation process happens once and then the in-memory program is just re-used. That avoids the cost of *both* scenarios.

[PodMaster]

You are talking about "bytecode", and there exists ByteLoader/B::Bytecode, but it is HIGHLY EXPERIMENTAL. USE AT YOUR OWN RISK (note the caveats). Once it becomes stable perl might start caching bytecode, but chances are it will never happen with perl5 as we know it (but it probably will with perl5 under parrot, aka ponie). Why? I don't know, but my guess is as good as any. Take your pick
• didn't think of it early enough
• not enough funding
• not enough interest
• general perl5 maintenance cruft

[crenz]

Just for fun, I „compiled“ Wx.pm via B::Bytecode into a Wxbc.pm. I then timed execution of perl -MWx -e '' vs. perl -MWxbc -e '' to compare module load times.

The result: Both consistently take about half a second to start on my PowerBook, with the original Wx.pm being consistently very slightly faster.

[htoug]

It is part of the plans for Perl6/Parrot. The parrot byte code should be compact enough to be able to be loaded faster than the original program.

That is not the case in Perl5, where the byte code is just a serialization of the internal syntax tree and is tightly integgrated (in obscure ways) with the compiler. AFAICR is has been discussed several times and been given up as nearly impossible. Grahams Barr's attempts at a perl compiler has given us the B:: namespace, that contains many useful modules, but no compiler that speeds up loading.

You do have to load the entire perl compiler to be able to interpret eg evals, requires and uses, so the loadtime will be high whatever you do.

[Eyck]

The point of this excercise is not to get rid of loading perl compiler, but bypassing compilation of whole lot of code ( AFAIK "use Sth" compiles Sth, and then compiles everything that Sth depends on. If you've got your Sth already compiled into parse tree, then you just load your parse tree and go on with execution ).

[DrHyde]

it would be trivial to add

If it's trivial, how about you stop whining and DO IT.

[Eyck]

have you actually read what I wrote?

I said it would be trivial for the people who were creating perl AT THAT time. And that would probably be Larry Wall, and would require time travel.

And one should research such problems FIRST, and that's exactly what I'm doing here.