Fix the problem described in PR #226221. (I think)

[mat]

Summary

The problem is that GH_SUBDIR are handled in a somewhat random manner. (The
truth is that in the end of things, they end up being sorted by the group name
being used in the GH_TUPLE.)

So if you have a submodule in bar/foo, and a sub-submodule in bar/foo/baz, it
may happen that foo/bar/baz is handled before foo/bar and then things are
messed up.

This makes it so the GH_SUBDIR target handling is sorted first by the
number of / in the path. (So, bar/foo is always handled before
bar/foo/baz.)

[mat]

• currently should not be used.

[ygy]

Looks good to me, though I am not a ports committer (yet), so it would be better to have someone else to review it :-)

[mat]

In D14532#304868, @ygy wrote:

Looks good to me, though I am not a ports committer (yet), so it would be better to have someone else to review it :-)

But does it fix the problem you are experiencing?

[ygy]

In D14532#305009, @mat wrote:

In D14532#304868, @ygy wrote:

Looks good to me, though I am not a ports committer (yet), so it would be better to have someone else to review it :-)

But does it fix the problem you are experiencing?

I did the test against my working port, and this patch worked beautifully. Thanks!