httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <>
Date Sat, 25 Oct 1997 00:55:33 GMT
Ok lemme state my place on this and other portability issues more clearly. 

For FreeBSD, Linux, and Solaris the point is moot:  we can support these
architectures very easily since many of us use them.  Well, actually, we
can support the latest greatest of these very easily.  I certainly don't
have any legacy linux 1.x systems.  There are some specialty architectures
here which only one of us support, and they do quite well. 

For most everything else, and this includes HP/UX because it varies so
greatly across versions (especially those which we don't ourselves run),
we cannot hope to support them to their utmost *on our own*.  In these
situations *we should make them as easy to support as possible*.  In the
case of spencer regex that means we should always use spencer regex.  If
that's inefficient then some concerned individual using those
architectures will take an active interest. 

Another example, if I have to waste 30 minutes trying to figure out what
the stupid uname values are that GuessOS spits out and users report when
they give us a "porting report" just to generate a three line patch then
that is a waste of my time.  It's a waste of any of our time.  We could
improve GuessOS so that on failure it generates the appropriate
${SYSTEM}:${RELEASE}:${VERSION}:${MACHINE} line (I find the output it
generates now too hard to parse).  But even still, users give us reports
for one single version of their OS, and usually have no ideas about other
versions of their OS.  And even more bogus, they give us non-context or
non-unified diffs, which are a bitch to merge.  The cost of support here
is huge.

This is one of the reasons I really like autoconf -- instead of
complaining to us, they complain to GNU and GNU adds features to autoconf
and we inherit that work.  But I'm not about to touch that argument again. 

Our development/beta process is so stifling that I feel that if these
problems aren't forced out into the open we'd never add the two lines
necessary to support a new platform.  Look at the age of some of those PRs
in my recent patch ... there's got to be a more efficient way for us to do

In many ways I'd almost like to just ignore portability issues until the
system vendors themselves wake up and provide us with a complete solution. 
i.e. never. 

I suppose I could just stop caring about portability, and that's actually
the way I'm leaning.  It works for me on Linux, and you guys give it a
good spin on FreeBSD.  Ship it ;) 


On Fri, 24 Oct 1997, Dean Gaudet wrote: 

> Bloat?  Define bloat for me.  It's less than 16k of active code. 
> Dean
> On Fri, 24 Oct 1997, Brian Behlendorf wrote:
> > At 12:02 AM 10/24/97 -0700, Dean Gaudet wrote:
> > >+1 for eliminating the use of the OS regex everywhere.
> > 
> > -1.  No need to bloat apache when it's a part of the system.  I contend
> > that it actually works on some systems, like BSDI and FreeBSD.
> > 
> > 	Brian
> > 
> > 
> > --=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
> > "it's a big world, with lots of records to play." - sig
> > 

View raw message