(For the reader: this is mostly about moving code around to ensure that everything is in Makefile and that Makefile.nt is always trivial; most of the sharing was already done before.)

Would it be possible to further factorize the post-processing of .depend (to produce .depend.nt) in the common in otherlibs/Makefile?

For the Windows-specific directories, I'd suggest ensuring that calling make from Unix fails explicitly.

A more ambitious approach would be to ensure that the Makefile in otherlibs/unix and otherlibs/graph could be used even under Windows and in that case automatically delegate to otherlibs/win32unix and otherlibs/win32graph to produce the required artifacts. These win32-specific directories could then be moved e.g. to otherlibs/unix/win32 and otherlibs/graph/win32 to make it clear that they are really alternative implementations of the same library. This might allow some more sharing between the build instructions for the Unix and Windows variants.

Alain Frisch (2016/08/30 15:30 -0700):

Would it be possible to further factorize the post-processing of
.depend (to produce .depend.nt) in the common in otherlibs/Makefile?

Sure. It has been done. The depend target has also been factorized and
the whole branch has been rebased on the latest trunk.

For the Windows-specific directories, I'd suggest ensuring that calling make from Unix fails explicitly.
A more ambitious approach would be to ensure that the Makefile in
otherlibs/unix and otherlibs/graph could be used even under Windows
and in that case automatically delegate to otherlibs/win32unix and
otherlibs/win32graph to produce the required artifacts. These
win32-specific directories could then be moved e.g. to
otherlibs/unix/win32 and otherlibs/graph/win32 to make it clear that
they are really alternative implementations of the same library. This
might allow some more sharing between the build instructions for the
Unix and Windows variants.

Could these be implemented in separate commits?
Would it be okay to just add a TODO comment in the makefiles?

A make -f Makefile.nt clean (before any build) now fails with

make[1]: Entering directory '/home/frisch/ocaml/trunk/otherlibs/win32graph'
../Makefile:146: .depend.nt: No such file or directory

Would it be okay to just add a TODO comment in the makefiles?

Yes, of course.

Alain Frisch (2016/08/31 00:45 -0700):

A make -f Makefile.nt clean (before any build) now fails with

make[1]: Entering directory '/home/frisch/ocaml/trunk/otherlibs/win32graph'
../Makefile:146: .depend.nt: No such file or directory

Ah indeed, sorry.

It would actually be possible to make sure .depend is not included when
cleaning, but that would not be enough because in its present state
win32graph does not use any .depend file.

So my suggestion would be to renounce to the factorizatio commit for
now, because this would lead further than was initially intended.

What do you think?

I don't understand the specific issue with win32graph, but would like to say:

• This refactoring of makefiles is great and something we should have done a long time ago, so please don't give up easily, and let us all try to help @shindere complete it.
• otherlibs/win32graph, like otherlibs/graph, should eventually be removed from the core distribution and live and be distributed as a separate project. Maybe it is time to do that.

What do you think?

Yes, ok.

Alain Frisch (2016/08/31 03:14 -0700):

What do you think?

Yes, ok.

So this factorization commit is now gone.

I added a note in otherlibs/win32unix suming up our discussion but I
left the Makefile in otherlibs/win32graph unchanged for now.

Xavier Leroy (2016/08/31 02:07 -0700):

• otherlibs/win32graph, like otherlibs/graph, should eventually be
  removed from the core distribution and live and be distributed as a
  separate project. Maybe it is time to do that.

I can take care of that if you like. I'd just need to make sure what it
involves technically, in addition to removing the directories and all
the references to them in other files, or perhaps just leaving a
directory with one file saying it has been removed, as is done for
ocamlbuild I think.

Alain Frisch (2016/08/31 00:48 -0700):

Would it be okay to just add a TODO comment in the makefiles?

Yes, of course.

Done in win32unix. I didn't touch win32graph yet, in case it is decided
to remove it from the distribution.