Re: Running a plain ./config.status at top level

[Akim Demaille]

Alexandre Duret-Lutz said:

> Hi Akim,
>
>>>> "Akim" == Akim Demaille <address@hidden> writes:
>
> [...]
>
>  Akim> The idea is that we should probably run plain ./config.status at
> the
>  Akim> top level update of Makefile.
>
> Tom attempted to fix PR/46 (subdir5.test) with a similar patch last year.
>
> http://mail.gnu.org/archive/html/automake-patches/2002-07/msg00065.html

Arg, didn't know.  Thanks!

> The above thread lists a few drawbacks.  I think the most
> important one (besides not really fixing PR/46) is that always
> running a plain ./config.status when $(top_srcdir)/Makefile.in
> changes can take _minutes_ in a very large project, when only
> _one_ file really has to be rebuilt (the top-level Makefile).

My latest proposal took this into account.

> I think running a plain `./config.status' makes sense only when
> `config.status' has changed.  Perhaps we should start playing
> with `$?'.  If `config.status' appears in `$?', then run it
> without argument, otherwise run it only for the current
> directory.  Does it seem sensible?  I guess this scheme could be
> done in _all_ directories, not only the top-level.  Maybe this
> would also fix PR/46 (I haven't thought about that much).

I thought about $? too, but I have no idea how portable that is :(
That's why the latest patches did look at the most recent changed
dependency.

If you are concerned with speed, you might also consider adding
an option to automake that would let it decide what Makefile.am
are worth being reconsidered based on timestamps?  I don't know,
wild quick and dirty guess before running to the restaurant :)