[bug] Sibling submakes: wait for other submakes before failing

[Alejandro Colomar (man-pages)]

Let's say there's a project with a submake for each module.

If a module depends on a file of another module, it's the other module
that will have the recipe for it.  For example, the shared library of a
module links to the shared library of another module.  Each module will
have the rules for building their shared libraries, and I can state that
the shared library of one module depends on the other, but I can't tell
it how to build it.  I can't even tell it to run $(MAKE) on the other
module, because if I run $(MAKE) on all modules from the top-level
Makefile, and then submakes also call make on other modules, then there
are duplicate makes, which if running `make -j` can fail, as I may be
compiling the same file at the same time from different sibling
submakes, which is something not wanted.

Also, the other solution would be to have a huge Makefile that controls
everything.  However, except for a bunch (5~10) of lines, all of the
modules build exactly the same (it was hard to generalize it, but I did
it), so I have a module.inc.mk that can build almost every module, and
then each Mkaefile for each module has a bunch of specific lines, and an
include of the generic Makefile.  Then the top-level Makefile calls the
same commands but on different dirs, to build each module.  If I had to
move all that to the top-level Makefile, I'd need to copy-paste a lot of
code, because that generic module.inc.mk can't work on a single invocation.

Anyway, here's the bug report/feature request:
See the following reduced example:



~/src/test$ find * -type f
a/Makefile
b/Makefile
Makefile


~/src/test$ cat Makefile
.PHONY: all aaa bbb

all: bbb aaa

aaa:
        $(MAKE) -C $(CURDIR)/a

bbb:
        $(MAKE) -C $(CURDIR)/b


~/src/test$ cat a/Makefile
.PHONY: all
all: $(CURDIR)/t

$(CURDIR)/t:
        sleep 10
        touch t


~/src/test$ cat b/Makefile
.PHONY: all
all: $(CURDIR)/s

$(CURDIR)/s: $(abspath $(CURDIR)/../a/t)
        touch s



module b depends on a file produced by module a, but given that it will
execute later (in a single-process make, because of the order or the
prerequisites of the main Makefile, and in parallel make, because of the
sleep), it won't have the file available, nor will it know how to
produce it, and therefore it will fail.

A solution for this would be that when a make doesn't know how to
produce a file, and it is part of a "super"make (i.e., it's a submake,
there are other make processes running), it could wait for those other
processes before failing, and maybe those other processes will resolve
the dependency.

Is this something that could be fixed in the current make
implementation?  Or the intercommunication would have to change too much?


Thanks,

Alex



-- 
Alejandro Colomar
Linux man-pages comaintainer; https://www.kernel.org/doc/man-pages/
http://www.alejandro-colomar.es/