Twostep compiler?

[Brian Yennie]

> My reply is: Self-modifying code should be forbidden by law, I really 
> though we got over that about 20 years ago ;-)
> And: Decoding bytecode is not necessary, if the byte code is 
> precompiled code (like in JAVA or PHP, which uses precompiled 
> "twostep" code to speed up things). Doing a full interpretation of 
> plain text really will be the slower solution compared to bytecode. I 
> wasn't only thinking of "encryption" here but also of safety, 
> performance etc.
> The more "data" has to be loaded the less CPU cache can be used to 
> actually EXECUTE the code. I was used to calculate CPU cycles to make 
> the best of my code, so again this may just be my (wrong) development 
> background :-)

Hah! Yes, self-modifying code is frightening in many respects. But it 
should be noted than in standalones, there is a serious limitation: the 
"do" statement and setting of script properties is subject to a 10-line 
limit. Also don't forget that most interpreted languages have the 
equivalent of "do", so they have the same potential hole.

Don't forget that without a 3rd party accelerator (or a "1st party" 
accelerator if you count those Zend weasels!), PHP is actually 
re-compiled for every execution. Granted, many people use an 
accelerator product.

In theory you are correct that re-interpreting source code must take 
extra time, but in practice I think you will be very pleased if you 
compare Transcript performance to that of Java. GUI runs much faster 
(and is at least as cross-platform, albeit outside the browser), string 
handling is light years faster, even graphics and math stack up pretty 
well, although you might be able to dig deep depending on your Java 
platform, skills, etc.

- Brian