Re: Improvement on parallel make

[Philip Guenther]

On 12/12/06, Paul Smith <address@hidden> wrote:
...
> Right, exactly.  However, this is a very common requirement in
> recursive builds; we can't get away from it.  Suppose we were to
> declare the REAL prerequisites in the top-level makefile; a.o
> depends on foo.h (which will be generated from one of the
> subdirectory builds):
<rehash of the initial problem description in the RMCH paper>

I guess my point wasn't clear.  I understand completely that declaring
full and correct dependency information is not practical in a
recursive make setup.  I'm not suggesting that.  The point of that
part of my comment was the .WAIT doesn't appear to be a good solution
in a non-recursive setup because in that case you can just declare the
right dependencies.

(Yes, I'm ignoring the 'ar' issue because .WAIT is a poor solution to it)

So, if we assume that you're already in a recursive make setup, is
there an acceptable solution to the problem solved by .WAIT?  It
appear to me that there is.  As I wrote in a previous message, this
makefile:

local_objects = a.o b.o c.o d.o
all: all-recursive .WAIT all-local
all-recursive:
     $(MAKE) -C subdir all
all-local: $(local_objects)

can currently be implemented like this:

local_objects = a.o b.o c.o d.o
all: all-recursive
    $(MAKE) all-local
all-recursive:
    $(MAKE) -C subdir all
all-local: $(local_objects)

For bonus points, change the 'all' rule to:
all: all-recursive
       $(MAKE) -s all-local
to suppress the extra "entering directory" message.

If the boat has already sailed on using recursive make, then I see no
reason not to use it in solving the problems it created.


Philip Guenther