Re: PATCH: Support Windows in event-loop.c

["Eli Zaretskii" <eliz at gnu dot org>]

> Date: Thu, 21 Apr 2005 16:56:17 -0400
> From: Daniel Jacobowitz <drow@false.org>
> Cc: Mark Mitchell <mark@codesourcery.com>, gdb-patches@sources.redhat.com
> 
> > It isn't analogous: HAVE_POLL tests for a system-independent
> > functionality, not unlike HAVE_UNISTD_H, while USE_WIN32API tests for
> > an OS-specific misfeature and names an OS-specific macro.
> 
> I don't follow this objection:
> 
>   - I don't think that we get to call OS-specific interfaces
>     "misfeatures".

I'm now sorry that I used the word "misfeature" because it drew your
attention away from the real objection.  But since you asked: the
misfeature here is that `select' works only on sockets.

>   - I don't think that answering your objection with an autoconf test for
>     WaitForMultipleObjects would make you any happier.  Would it?

No.  The real issue is that WIN32API is an alias for _WIN32 and its
ilk, something we spent some time weeding out several months ago; and
that the code inside the ifdef calls WaitForMultipleObjects, which is
a Windows-specific function not standardized anywhere else.

> How does the fact that only Windows provides WaitForMultipleObjects
> make it conceptually different from the fact that a bunch of systems
> provide poll, and a bunch don't?

The difference is that `select' and `poll' are both Posix functions,
while WaitForMultipleObjects is MS-Windows specific.