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

[Mark Mitchell <mark at codesourcery dot com>]

Mark Kettenis wrote:

> The problem here is that traditional gdb programmers are familliar
> with POSIX and therefore certainly with select, and probably with
> poll.  They might very well not be familliar with Windows and the
> magic of WaitForMultipleObjects.  If we accept the patch as-is, the
> number of people that can hack on the event loop code decreases since
> the number of people who are familliar with both Windows and POSIX is
> smaller than the number of people who know about POSIX.  This is why I
> think it would be preferable if an emulation for select or poll would
> be used here.

The flip side of this argument is that unless we can provide something 
that really does behave like select, people will *think* they understand 
what's going on in this code, but won't.  For example, if we end up with 
a select that can't handle waiting for "write" events on file 
descriptors, then at some point someone might add such a thing, and then 
Windows will break.

This is a somewhat different situation from the sockets situation, where 
the Windows API is very nearly identical to the UNIX API, modulo 
spelling, for the purposes of GDB.  Intuition about UNIX sockets applies 
to Windows sockets, but intuition about UNIX file descriptors and select 
doesn't apply all that well to Windows.

I have a feeling, though, that I'm shouting into the wind.  I understand 
that the GDB community is POSIX-centric and of course GNU is Not UNIX, 
rather than Not Windows.  I understand that, and it's much more 
important to me that the functionality be provided than that any 
particular form be used.

I think my current intent is to wait for Chris' comments re. 
WaitForMultipleObjects and then try to make something that works like a 
limited form of select.

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304