configure output summary

[Josip Rodin]

On Thu, Nov 18, 2010 at 08:48:38AM -0500, John Dennis wrote:
> On 11/18/2010 08:21 AM, Josip Rodin wrote:
>> I've actually been a bit confused by the notion of having separate autoconf
>> installations/invocation in multiple subdirectories. The point of that would
>> seem to be that if you just want to reconfigure and rebuild one particular
>> part, you can do it.
>>
>> But who ever does that?
>>
>> It seems to me that everyone only ever wants a single autoconf instance for
>> the whole tree, which can generate all the subdirectory makefiles.
>
> Yeah, I've never quite understood that either, but it works which at the  
> end of the day is what matters most even if it seems odd.
>
> If someone ever does decide to work on the build tools I wonder if it  
> might make sense to abandon autotools.

I personally have no problem with autoconf per se, configure.ac syntax in
general tends to be fairly clear to me. But having N copies where we only
seem to need 1? That sounds like a problem.

Also I think that this line of reasoning "it's apparent that it's odd, few
understand it, yet it works, so let's either keep it or attempt to replace
it completely" is problematic in and of itself. We must be able to fully
understand our own code, and the requirements that led to it, in order to be
able to both try to fix the problems in the existing solution *and* to be
able to attempt a successful replacement. If we don't, it's likely that
we're just going to end up repeating old problems and making things worse.

So it would be good if we could first get an authoritative opinion on
whether support for subdirectory reconfiguration is actually necessary,
or if perhaps it's a remnant of some other unrelated idea. Alan?

I've had some experience analyzing auto*-based build systems on other things
I've packaged, which seemed to result in them becoming less obfuscated, so
I could have a crack at this one if it's possible.

-- 
     2. That which causes joy or happiness.