Re: [Alephmodular-devel] A note on films

[Woody Z. I. <woo...@sb...>]

On Saturday, January 18, 2003, at 03:14  PM, Mark Levin wrote:

> On Saturday, January 18, 2003, at 12:40  PM, Woody Zenfell, III wrote:
>
>> Might I suggest that it may be more practical to have a facility for 
>> dumping and another for comparing against an existing "checkfile", so 
>> that the game can automatically detect and warn you about 
>> inconsistencies?
>
> Heh... Might as well let the game record the playing film into a 
> brand-new film and diff them :P

No no no no no this is **totally** different from what I'm saying.  I 
agree that what *you* said is useless.  :)  (Unless by "playing film" 
you mean the M2 filmfile and by "record into a brand-new film" you mean 
"render into a QuickTime file" or something... which I doubt.)

Remember a film file only has the "inputs", i.e. the action_flags 
stream.  The checkfile would be a representation of (part of) the game's 
*state* (i.e. something gotten by extensive computation within the 
engine, based on the inputs but very definitely distinct from them).

The problem we're trying to detect is when applying the same input 
stream produces a different "state stream" - indicating that that 
extensive computation within the engine is no longer coming up with the 
same results it once did.  And the checkfile/filmfile combo would be 
*quite* helpful in this regard.

(I assume it's obvious that the "checkfile" would be generated from the 
filmfile as early as possible... then occasionally in the game's 
development lifetime, that original filmfile and checkfile would be used 
to make sure everything's still working the same way internally.)

I guess I didn't get my point across very well - must not have used 
enough words.  ;)

OHH, or maybe there's a concern that AM *already* diverges from M2 
behavior.  Well, right, obviously the checkfile scheme can't detect that 
since M2 doesn't write checkfiles.  (I guess you could carefully hack 
the binary or something to add this capability in, but...)

A checkfile alternative could be to use a saved-game at the end of the 
film, the idea being that if the states diverge at some tick, they will 
be different for all subsequent ticks, and thus will show up in the 
final state.  And of course a saved-game captures all relevant 
game-state, and so *might* be able to catch things that merely the value 
of get_random_seed() would not.  (Though I rather doubt it, in 
practice.)  And, automatically comparing two whole game states is not 
nearly as easy as comparing two 16-bit values.  (Yeah you could actually 
write the two saved-games out and diff them or something, but it's more 
painful to automate.)

Woody