Haven't really thought about it ... not really sure if nsis 2.0 should exist...
I mean, from the exehead side, there are many directions you could go, but all of them are just different.. do you go:
- smaller but less functionality?
- larger and more functionality? is this necessary?
- dynamic exeheads that are generated on the fly? is this necessary?
Or do we focus on something like using bz2, but compressing the entire datablock as one (this is necessary to really make bz2 outperform zlib in most cases). The problems with doing that are:
- bz2 has more runtime memory requirements
- bz2 would make us have to decompress all the data, do we do that
to a temp file or something? it'd make the installer slower....
- bz2 decompressors may be larger than gzip, but that's probably not much of an issue...
Finally, there is the issue of the compiler end. Basically the existing scripting language has evolved into its current state. Which makes it kind of gross and nasty. But it works, and I kinda like it. Would all the work of making some new fancy parser that is more C like be worthwhile? What good does that do?
I don't know. I think I'd like to focus on taking NSIS 1.x as far as it can go -- i.e. as flexible and small and efficient as possible. Thinking about experimenting with bz2 more, but not sure if it is worth the tradeoffs.
I'd like to cut down the existing exehead from 36k to 32k. Just for cluster size effect. Heh.