PushIOvHelper errors

[Russ Allbery]

Marco d'Itri <md at Linux.IT> writes:
> On Sep 11, Russ Allbery <rra at stanford.edu> wrote:

>> It's pretty straightforward on the storage API end; just add an
>> SMsendfile routine that takes a token and a file descriptor.  Overview
>> will be trickier and isn't as obvious of a win.

> It's useless, sendfile(2) is a win when you have to copy something
> straight from disk to the network.

Which we do, for at least tradindexed and I believe buffindexed.

>> The internal implementation of SMsendfile for tradspool will need to
>> account for non-wireformat articles, though.

> In these situations you just use write(2) or writev(2), sendfile(2) may
> actually be slower.

Right.  My point is more that nnrpd shouldn't have to care about that; it
should just call SMsendfile and the implementation should do the right
thing.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.