Re: 52-factored-install.patch

[Akim Demaille]

Tom Tromey <address@hidden> writes:

> Targets which are presented to the user must be defined regardless of
> whether they have any effect.  So for instance `install-exec' must
> always exist.

I understand this.  Then I'll have a list of required targets and
output them in any case.


> Akim> Also, I have not really understood why the -hooks had to be run
> Akim> explicitly by invoking make again instead of just listing them
> Akim> as dependencies.
> 
> If it is a dependency then it won't work properly with a parallel
> make.  That's because the -hook for must be run after all the other
> rules.

Why don't we use dependencies to say that -hook must be run last,
instead of using the actions to run them last?  Not that it matters,
it's just pure curiosity.  And `easier to implement' is definitely an
excellent reason!  Any advantage in one over the other?  Don't we
avoid forking another make which has to load Makefile again? (if we
use the dependencies to force the order I mean).


> Anyone who relies on the ordering has a broken Makefile.  If order
> matters you must either have an explicit dependency or you must
> reinvoke make.  If your patch changes only dependency orderings then
> it will only affect Makefiles which are already buggy.

Good to know, thanks!