On Tuesday, April 15, 2003, at 02:33 PM, Mike Chamberlain wrote:
This is essentially what is being done currently. The only exception is that currently the system uses a push model rather than your proposed pull model. The choice is probably arbitrary
I think the differences in the implementation details mean it's a lot more than arbitary. The pull model makes a pipeline based cache implementation trivial.
The push model is pretty trivial too. In reality you have to query each stage to see if it's changed - so whether you start from the back and work forward, or vica-versa is arbitrary. If a cache point could determine if it's output was valid without consulting preceding stages, then a pull model would be more efficient. But since the output of any stage is dependent on the previous stages, it can't.
The pull model might be a little cleaner, but I don't think it's a big difference in this situation.
- although I think a pull model would be harder to implement due to dependancies being discovered during processing (rather than predetermined). I
I don't see why that is. The pipeline can still be predetermined, you just
can't modify it so easily once it's created. ie insert_last_stylesheet().
I was purely referring to the process of gathering dependancies and then storing them into a cache. If the cache was called first and then "pulls" from the source, it's going to have to handle the situation where the dependancies are unknown and then deal with storing them itself. Again, no big difference from whats there now (perhaps even a little more OO'ish) but I still think it's kind of arbitrary.
*shrug*
Regards, Chris
PGP.sig
Description: This is a digitally signed message part
