() l...@gnu.org (Ludovic Courtès) () Wed, 17 Apr 2013 14:52:40 +0200 concrete cases of recurring short writes?
The original application was "ttn-do serve-debiso", which configures "ttn-do sizzweb" to serve debian ISOs (or rather, the various component .deb files and metadata that apt-get requests from the loopback-mounted ISO) over the home network. Specifically, proc ‘prep-file-transfer’ in module ‘(ttn-do sizzweb)’ calls ‘sendfile’ in a loop. Admittedly, that code could be replaced by ‘sendfile/never-fewer’, as the retval is not used in a particularly intelligent way, but anyway, the point is, at time of writing and early testing *sans looping*, apt-get on the other computer would stall and eventually die, complaining. That's when i learned to read and think about sendfile(2) more carefully than i had before. It would be nice to point to a ChangeLog entry that embodies this "ah ha!" moment, but as it turned out, the testing was done before commit. Oh well, life goes on... -- Thien-Thi Nguyen GPG key: 4C807502
pgpaljvkq1He6.pgp
Description: PGP signature