July 16, 2004

Truth is stranger than fiction

Well, unless you're reading Phil Dick, at least.

Just sent Python::Bytecode 2.6 to CPAN (knowing, bluntly, that I broke the tests) because the translator now works, at least for simple programs. Woohoo!

My first victim program was simple--the test shipped with Python::Bytecode. So this:

a = 2

b = 3

if a:

  print a, b

  c = a + b

  print c
does, in fact, print
2 3

5
Go figure. I'm not sure which is more bizarre--the fact that it works, or the fact that I actually wrote correct Python code. (By the time this is all done I'm going to have to crank out a few pages of INTERCAL code to compensate.:)

I'm by no means done--assuming I've gotten it right, only 41 of the 110 ops are implemented, so I'd darned well better code faster, but...

Update 1: Make that 44 ops. Attribute get/set/delete is in, though we succeed in spots that Python throws errors.

Update 2: Make that 49 ops. Unary stuff and iterator getting, since it's easy. May have to check on setting an immutable true and false value, rather than a Boolean of the right truth value. We'll see if it matters.

Update 3: Make that 51 ops. PRINT_x_TO ops. And now it's past time for bed

Posted by Dan at July 16, 2004 12:28 AM | TrackBack (0)
Comments

I'd put on a cheerleader outfit and cheer, but that wouldn't help you much. It's tough to code if you've gouged out your eyes...

Hooray!

Posted by: Clinton Pierce at July 16, 2004 09:42 AM

Congratulations. You are doing a lot of impressive work here.

Posted by: Matt Fowles at July 16, 2004 11:15 AM

Sounds great.
Are you planning to implement binary module loading (.pyd) any time soon? I'm not sure if it's necessary for the pie-thon, but it looks like the python math module (for instance) would require it unless reimplemented.

Posted by: Andrew Whitehead at July 16, 2004 01:06 PM

awesome!

Posted by: timn at July 16, 2004 01:08 PM

On second thought I'm an idiot, because that would require implementing the entire CPython API.
Maybe the only choice is rewriting those modules for Parrot.

Posted by: Andrew Whitehead at July 16, 2004 01:14 PM

It's possible that at some point we'll build a thunking layer for CPython extension modules. It's probably possible for some portion of them, though it's always an open question whether 65% coverage means that 65% of the modules compile, or none of the modules compile because they all use that one function we can't implement.

The ponie folks are at work on a Perl 5 XS compatible layer. It's likely that, compared to that problem, the CPython API is relatively easy. Or at least less difficult, since it's unlikely to actually be easy.

Posted by: Dan at July 16, 2004 01:44 PM