On 4/26/07, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote:
It looks slick, but I'm not sure whether that isn't too much overhead
for the whole thing... after all the different strategies really only
differ in one algorithm that has the same input and output.
The overhead lies in the function poin
On Thu, Apr 26, 2007 at 10:23:33AM +0100, James Youngman wrote:
> I like your idea of separating the strategy (making decisions about
> whether to exec now and start again, or add this argument to the
> list) from the nuts and bolts of the implementation. Perhaps we couls
> separate the buildcmd.c
Obviously the runtime of that implementation of lin_suggest_argv_split
is linear in the size of the input and that's unfortunate. But the
constant factor would be small and I think we can fix the problem
(moving to O(1)) just by changing the interface slightly later. I
don't think it's a big dea
On 4/26/07, Leslie P. Polzer <[EMAIL PROTECTED]> wrote:
If there's sufficient time, the linear approach can be substituted
by a binary one then (while still doing it only on-demand, of course).
I like your idea of separating the strategy (making decisions about
whether to exec now and start a
On 4/26/07, Leslie P. Polzer <[EMAIL PROTECTED]> wrote:
It would be even more efficient to store the last known value that
worked somewhere. Perhaps /var/cache would be a good idea?
It certainly could be more efficient that way, but I have an
instinctive aversion to find modifying the filesyst
