[Swfdec] compiling vs interpreting

[Eric Anholt]

On Sat, 2007-01-13 at 00:36 +0100, Benjamin Otte wrote:
> So I thought I'd write this down, in case anyone has useful input to it or
> wants to read up on the decisions later.
> 
> Currnetly swfdec compiles actions from a flash file into JSScript objects
> and executes them using the standard Spidermonkey API. While this allows
> all the niceties that are available with SpiderMonkey (most
> important: a whole debugging framework with stack inspection etc - see
> player/swfdebug to get an idea about this), I don't like this approach.
> 
> I'm probably going to rewrite this into an interpreter that duplicates a
> lot of the functionality of jsinterp.c and interpretes the original
> Actionscript from the file. So in short: I'm going to end up where Eric
> already was. Here's a list of the reasons for this decision:
> 
> - security
> SpiderMonkey checks security of its scripts at compile-time. This is nice
> for SpiderMonkey since it can use opcodes tailored directly at its
> compilation and do all necessary checks while it compiles. However,
> Actionscript bytecodes are stack-based and I'd need to do runtime checking
> to make sure it does never pop off more value than it had pushed before.
> 
> - incompatible bytecodes
> the bytecode for a function call (JSOP_CALL) includes the number of
> arguments to call the function with. However, the Actionscript code pops
> it off the stack. While being security sensitive (see above) there is also
> no bytecode for it.
> Another such example is the GetVariable action that can request a whole
> old-style path such as "/root/movie/video:width". Good luck trying that
> with JSOP_GETELEM. I'd need to hack this functionality into all these
> opcodes or add new ones, but that doesn't make it simpler.
> 
> - one layer less
> since I don't convert to (probably slower because of various workarounds)
> Spidermonkey bytecodes, there's one area less that can fuck up, namely
> jsinterp.c - which is hell to debug, because 90% of the code in that file
> is inside macros, and you tell me why ELEMENT_OP (2, -1, CACHED_GET (-1))
> or whatever crashes. So I hope to get the crashes into my code - I can
> easier work around them then without affecting other working code.
> Oh, and it's probably faster, too.
> 
> - version differences
> Flash executes the same bytecode different depending on version. So I have
> to add lots of workarounds generating stupid bytecodes (like converting
> NULL to 0 for some versions) in the code that I could solve far easier in
> my own interpreter.
> 
> - better housekeeping
> Since a JSScript is bound to a JSContext, I have to carry the JSContext
> everywhere to all my parser objects (aka SwfdecCharacter) that can contain
> scripts. If I'd just have a SwfdecBuffer there, it'd be far easier to
> maintain.
> 
> - debugging
> Since when reverse enginering you want to step in ActionScript bytecode
> cunks, you need a way to do this. It works quite ok (see
> swfdec_debugger.c), but it has its issues. And it's hard to catch internal
> bugs this way. You end up in the hell that is jsinterp.c again. :)
> 
> I think I'm versed well enough in SpiderMonkey now to make JSFunction cope
> with 3 types of functions; JSScript, native and SwfdecBuffer. :)
> I'm probably gonna create a new branch (named interpreter) and try it out
> there.

In the conversation on irc, I didn't realize that you had replaced all
of the stuff I wrote.  I haven't been following the new development
really at all (now that we've got commit mail working hopefully, I
should be paying closer attention).

I kind of liked what I had before, of reusing spidermonkey code as-is as
much as possible (getting a good chunk of class library automatically,
in particular), while retaining actionscript bytecode interpreting in a
specifically actionscript interpreter.  It sounds like this is roughly
where you're going, which is fine with me.

I'm not trying to endorse spidermonkey in any way -- if we come up with
something else that works, great.  Trying to understand how that
spidermonkey worked broke my mind many times.  I just wanted a library
that got us a good chunk of the way there without having to write it all
ourselves.

-- 
Eric Anholt                             anholt at FreeBSD.org
eric at anholt.net                         eric.anholt at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: This is a digitally signed message part
Url : http://lists.freedesktop.org/archives/swfdec/attachments/20070112/b73fefdf/attachment.pgp