On 01/24/2013 04:38 PM, Amos Jeffries wrote:

> I was under the impression that we had
> proven the usefulness of add-and-remove model.

I do not think so. I provided an estimate for add-and-remove costs and
demonstrated how add-and-remove can be optimized, but those costs were
high and that optimization was not sufficient IMO.

Also, using eventAdd for I/O timeouts when the rest of Squid is using
Comm for I/O timeouts will move us in the wrong direction.


> So, with maintainer hat on please do the changes necessary to perform
> add-and-remove cleanly for timeout events. *If* possible using the
> event*() API instead of raw FD table access.

I will submit a patch that includes add-and-remove for timeouts.

If somebody wants to write a patch with clean removal of timeouts using
eventDelete(), please be warned that eventDelete() code currently does
not work in ConnOpener context. I will not waste your time further by
describing the boring details of why it does not.


Thank you,

Alex.

Reply via email to