Re: Pushing the `gnus-range-*' functions down into the C layer

[Wojciech Meyer <wojciech.meyer@googlemail.com>]

Lars Magne Ingebrigtsen <larsi@gnus.org> writes:

> Wojciech Meyer <wojciech.meyer@googlemail.com> writes:
>
>> I think they are best to solve your performance problems.  The idea is
>> to keep just one array for all of the articles, means one per
>> group. That would be just the internal representation, you still
>> save-read from lists with article ranges.
>
> The performance issue is converting between FOO and ranges.
> Representing FOO as arrays instead of lists doesn't help with the
> conversion problem.

Yes, I understand that, but one could eliminate ranges completely, but I
am not an expert on nntp. Plus it has a linear complexity, as you it can
be computed with just simple state machine. How long it will take to
traverse 10000 articles consing pair of integers? (I bet not that
long). (maybe just reading the articles on the fly, then you have only
additional cost of skipping - proportional to number of them (0(n)).)
As I said, I am not entirely in the problem domain so I can't say with
100% sure, but I feel it should not be that complex to cause delays in
Gnus, and BTW: I am very happy user of Gnus, and I didn't notice this
problem :).