On Mon, 10 Jan 2011 00:09:12 +0100
Pavel Sanda <[email protected]> wrote:
> [...]
> again i would like to keep this code simple and stupid if possible.
Well, in general I completely agree with this. In terms of stability
and maintainability, it is important to not to overcomplicate the code.
Unfortunately, in this case the stupid code would also be very slow,
and to speed things up it will be corrected anyway, which will render it
more and more complicated (with all that threads' priority setting,
checking whether the particular inset is visible or not and reordering
the generation on-the-fly, maybe even checking for the number of cores
available to adjust the limit of subprocesses). A fast method doesn't
need that, and the one I've mentioned should be fast enough.
The bisection isn't *that* smart anyway :). Just look at the
pseudocode provided: if you remove all the comments, only few simple
operations will be left, and the whole code generating the previews is
completely orthogonal to (independent of) it.
Enumerating IPs one-by-one means using the IP generation script
n times, instead of using it only log(n) times, per each opened
document. This is a huge boost for larger ones (like math tutorial
you've mentioned), so please, reconsider.
Regards,
Michał Skrzypek
