[distcc] Yet another compile failure

[Martin Pool]

On 17 Sep 2003 Ajay Agrawalla <aagrawalla at hyperchip.com> wrote:

> Martin Pool wrote:
> Yup. Agree.  Having different version of tool chain on the build machines
> can be nightmare while building one executable or lib.  May be this
> certification phase for the build process itself is not a good idea.

The only solution to compiler problems is for people to install the
right packages on every machine.  Hopefully they do that in some
systematic way by installing rpms or debs or something similar, rather
than unpacking compilers all over the place.  (I live in hope... :-)
In this case making sure that the versions are correct is trivial for
the user.  

If they have a mess of partial or mismatched installations, as can
happen, then detecting the problem is only a small step towards fixing
it.

A fair fraction of bugs reported these days and observed by me seem to
be, as in Russ's case, actually bugs in the user's Makefile.  Much as
I would like to be able to magic them away, I don't think there is
anything much we can do.

I think in many situations Make is also the performance-limiting
factor.  Moving to a new tool like SCons[1] or writing a better makefile
can help enormously, though it can be hard work.

> while we on this topic of pre-certification, have you thought abt it for
> other purpose like authenticated machines, qulified machine depending on
> their network thouroughput, etc.  but may be these are just things we as
> developers just don;t care to invest any time in it.

I have thought about it, and in particular about having the list
automatically sorted by speed.

At the moment I am spending a little time on trying to do a better
monitor program[0], and then I might try that.

> Good job at distcc though.

Thanks.

-- 
Martin 

[0] http://distcc.samba.org/news.html#2003-09-16
[1] http://lists.samba.org/pipermail/distcc/2003q1/000619.html