[Scummvm-devel] ScummVM binary size comparison

Max Horn max at quendi.de
Sat Nov 8 19:48:58 CET 2008


Am 08.11.2008 um 16:02 schrieb Neil Millstone:

[...]

>>
>> It would still be a good thing to determine where it doesn't work,   
>> precisely, and to report this to upstream, to get it fixed.  
>> According  to Robin, the binary size goes down several hundred KB,  
>> which we'll  never be able to cut by optimizing our code manually.
>>
>>
> I'm going to do this, but tracking these random crashes with no  
> debugger takes a very long time.  Working out a minimal piece of  
> code which replicates the error takes even longer.   I get the  
> feeling that Tramboi or Robin would be better at this than me!

Quite possible. If you can give them at least a hint where to look,  
that would already help. I guess once you know what code roughly  
breaks, you can try to figure out what in there miscompiles, and use  
that to construct a minimal example.


>
> I think this is the wrong solution though.  I had a version of 0.9.1  
> that can run The Dig and Full Throttle.  I only managed this by  
> compiling with Thumb and turning optimistaions off.  If we need to  
> use Thumb just to keep games working that worked a few months ago,  
> there's going to be no chance of getting the binary down to the size  
> it was then in order to run The Dig/Full Throttle.
>
> And what do we do next time the binary grows?

If you read what I wrote, I specifically said that we should try to  
reduce code size actively -- but we will never be able to save 500KB.  
Thumbs mode will. Engine also grow, so, in my eyes it makes perfect  
sense to do both things to reduce code size: Comb the code, but also  
try to get thumbs mode working.



Bye,
Max




More information about the Scummvm-devel mailing list