[gpsd-dev] Changing build systems

[Eric S. Raymond]

Bernd Zeimetz <address@hidden>:
> So at the end a proper function of libgps  (the shared one) is never ever 
> being
> tested?
> Also you are wasting a lot of memory, libgps21 is 119k. There are fore sure
> platforms where you want to avoid that.

That isn't the actual cost.  Only the modules actually referenced would be
linked, a much lower overhead.

> Again, the only proper way to fix this mess is to fix the build system.

You're welcome to propose a better one.  But "better" excludes anything 
two-stage, e.g. based on Makefile generation as an intermediate step.  
Having had way too much experience with such systems (autotools, Imake)
I do not believe it is possible for them to be a net improvement.  cmake
seems to me to be an admirably determined effort to overcome the problems
with two-stage designs, but I do not believe those problems are tractable.

> The only option I see is to take autotools or cmake,

*No.* That will *not happen*.  I consider both designs fatally flawed.
You will get no further by continuing to beat this drum - all it does
is irritate me, not persuade me in any way.

If the price of that not happening is that we lose your participation,
I will - very reluctantly - pay that price. You were a pleasure to
work with before you got this fixated idea that SCons is your enemy.
Now you are not rational on this subject. If you can't let go of that,
the project has to let go of you.

I will compromise this far: if you can choose a *one-stage* build
system (no Makefile generation) that you think is better than SCons,
it will get my most serious consideration, and probly adoption.  Waaf
might be mature enough at this point. I haven't been keeping track.

In the meantime, I'm going to implement the simplification I proposed.
Perhaps once you calm down you will see that it works.
-- 
                <a href="http://www.catb.org/~esr/";>Eric S. Raymond</a>