httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: apache/linux modules
Date Mon, 02 Feb 1998 22:40:07 GMT


On Mon, 2 Feb 1998, Cristian Gafton wrote:

> On Mon, 2 Feb 1998, Brian Behlendorf wrote:
> 
> > was suddenly an issue (such as the whole serialized-accept bizness) which
> > we needed to address per platform.  If you have experience with this and
> 
> As long as you can build a test case in what you are interested, the
> autoconf will handle it...

That's the problem, you can't necessarily build a test case.  As I
mentioned there are some bugs which show up only under heavy load, or only
when there are slow remote clients, and so on.  These are insidious kernel
bugs that we have to work around.

There are things like serializing primitives -- we have a half dozen ways
of doing it.  In order to decide which of many to use on a particular
system we generally do some benchmarking.  But it turns out that some
systems vary greatly depending on what options are installed, or if the
box is SMP or UP (IRIX comes to mind).  So we default to something that's
the least broken.  But that default is different for different
platforms...

So in the end I fear we'd end up using autoconf in a way it wasn't meant
to be used -- with dozens of architecture specific hacks.

*But* since I expect autoconf would eliminate the rather annoying porting
bug reports we get I'm mostly for it.  I just don't want to --have --to
--specify --sixteen --million --command --line --options --when --building
--a --static --server. 

Dean



Mime
View raw message