Flython

[Dave Benjamin]

In article <RZ-dnc4U_fMxCb_cRVn-uQ at powergate.ca>, Peter Hansen wrote:
> Dave Benjamin wrote:
>> I have been writing
>> ActionScript for two years now, and I am so sick and tired of its lack of
>> useful data structures (like sets and dictionaries), god-awful
>> error-checking (any bad message results in the "undefined" value, which
>> accpets any message and returns the "undefined" value, and so on, until your
>> error surfaces someplace completely different from where it was caused),
>> lack of exceptions, lack of any standard for modularity... I could just go
>> on and on. I'm currently maintaining a 10,000 line ActionScript program, and
>> I would be so happy to rewrite it in Python right now.
>> 
>> And Python *is* magic. =)
> 
> So Dan was right on both counts. :-)   ('''Either Dave thinks it just 
> can't be done, or he really, /really/ wants this compiler.''')

I'm sure it can be done. I just don't think *I* can do it. =) At least, I
have very little experience in writing compilers or anything like that...

> What version of the SWF format are you stuck writing to?  It
> appears that even SWF 5 supports Javascript-style "dictionaries"
> by using attributes on an object.  For example, "ActionEnumerate"
> will let you iterate over all properties on an object ala the
> Python "for k in dict".

Flash MX, which is SWF 6 I think. It is true that you can (ab)use objects as
dictionaries, but this has several unforunate consequences:

 - Keys must be strings. All objects will be toString()-ed.

 - Keys are case-insensitive (until SWF 7, I suppose).

 - Adding methods to the dictionary class.
   (to support .keys(), .values(), etc.) requires modification of the Object
   prototype, and thus the use of undocumented features (ASSetPropFlags) to
   hide these methods from "for ... in ..." loops.
   
 - There is no exception on a bad key, since there are no exceptions
   (pre MX 2004, anyway), so the only way to even get an error report is
   to use another undocumented feature, the "__resolve" method.
   
 - Implementing dictionaries as nested arrays, to get around the case
   insensitivity issue and string requirement for keys, has the most
   awful performance I've ever seen in any language.

> As for exceptions, they were added in SWF 7, so I'm guessing you're
> stuck somewhere prior to that.  Your life must really suck. ;-)

Well, I'm getting by. But there are so many times when I think "Wow...
Python really got this right". You don't realize how many ways things can go
wrong when you're spoiled by the good design decisions behind a language
like Python. I've had this feeling many times over.

> > Peter Hansen wrote:
>> Good point. I expect at least the full syntax as of Python 2.1, which means
>> list comps but no "yield" or metaclasses or any fancy stuff like that.
>> Perhaps one pie per Bagley's Shootout example, or maybe one pie per fully
>> exposed Flash API (MovieClip, LocalConnection, etc.). I'm open to
>> negotiation.
> 
> Gosh you set a high target.  Much as I like the taste of pie
> (Mmmmm... piiiiie), and that cold clammy feeling as it drips down
> inside my shirt, I don't think I'll be accepting the challenge any
> time soon.  Not, at least, until I learn a lot more about compilers
> than I currently know (hint, it involves the word "sweet" and "all").

See, I shouldn't have blown my cover. I could've at least gotten a "print"
statement out of you before I opened my big mouth.

> One thing that would help would be a little direction from
> someone in the know as to how Jython handles things that are
> simply not available in Java bytecode.  For example, as far
> as I know the main operators for basic primitive types in Java are
> implemented with different bytecode instructions, while in
> CPython things like "add" are implemented as a single bytecode
> which happens to special case integers for performance, but
> falls back to a dynamic lookup if the types being added are
> not integers.

Well, in ActionScript, and I'm guessing this applies even at the bytecode
level, things are very weakly typed. Numbers are transparently coerced to
Strings and back, all the time. There's a single "add" instruction that is
used for Numbers and Strings, performing addition, concatenation, and
coercion as necessary. At least, so it seems looking at the FLASM dump of my
program.

> That looks to me like a bit of a conundrum.  To turn the Python
> code into the appropriate Java bytecodes, wouldn't you have to
> know the types of the data during compilation (as in static
> data types)?  But Python doesn't have that (yet), so how is
> it done?  Are all "adds" sent to some library routine which
> is made available to all applications, and that routine then
> handles the Python-style dynamic add operation?

I like key lime, but not banana cream. Apple and rhubarb are good, too.

> In some ways, I think the SWF bytecode is even closer to Python
> than Java's bytecode is, but I'm pretty sure that I couldn't
> begin to imagine how one would target either of them...

I'm sure it is, since it was designed with a dynamically typed language in
mind. They're similar though; they're both stack-based, and they both
support some notion of objects at the bytecode level.

Let me know when it's ready. Thanks. ;)

Dave

-- 
 .:[ dave benjamin: ramen/[sp00] -:- spoomusic.com -:- ramenfest.com ]:.
        "talking about music is like dancing about architecture."