Review Request: Replace thread usage with local event loop in kio/kio/hostinfo.cpp

[David Faure]

> I just tried a test calling that function with 0 as timeout (so it always
> timeouts) and sometimes it waits until 1 second so it is quite clear it is
> waiting for the thread.

Yes, Thiago explained to me the notion of cancellation points, and proved me 
wrong. And since anyway terminate is a bad idea, this needs to be redesigned.

Thomas: I can't see how locking a mutex can work in a single-threaded case 
(there are no threads in your code; you seem to rely on the fact that Qt uses 
a thread internally in QHostInfo? That seems dangerous...).

Thiago and I came up with a design like this:
* a single thread that makes async lookups. It's never blocked, it can always 
almost immediately read new requests from a queue and start them.
* each request is encapsulated in a class that has the QString
but also a QSemaphore.
* the main thread can block using a timeout (QSemaphore::tryAcquire(timeout))
* the request class is stored using a shared pointer, so that it's only 
deleted when neither the main thread nor the worker thread need it anymore
[otherwise the deletion is a problem, in the timeout case vs the normal case]

-- 
David Faure, faure at kde.org, http://www.davidfaure.fr
Sponsored by Nokia to work on KDE, incl. Konqueror (http://www.konqueror.org).