On Sat, Oct 12, 2019 at 3:08 PM Jouke Witteveen <j.wittev...@gmail.com> wrote: > > On Sat, Oct 12, 2019 at 2:50 PM Paul Smith <psm...@gnu.org> wrote: > > > > On Sat, 2019-10-12 at 13:11 +0200, Jouke Witteveen wrote: > > > Before, this was only expanded to $(*F) in prerequisites. > > > > Sorry but I need more information than this; I can't understand this > > change. > > > > The bug in Savannah, as I understand it, is that directory prefixes > > which should be present are missing during prerequisite expansion. > > > > How does switching from $* to $(*F) (which is explicitly a file only) > > help solve this problem? > > > > Cheers! > > > > Haha, yeah, this seems counterintuitive, doesn't it! > The true cause of the issue is not immediately visible in the patch: > stem_str is used as a temporary value of file->stem and this value > gets assigned to $* in the second expansion. To be consistent with the > value of $* in the recipe, we should thus prepend the path to stem_str > in some cases. This is what my patch does. > > However, as a sort of 'hack', the current code replaces % by $* on the > first expansion, relying on the second expansion to do the right > thing. This does not work after fixing $*, as directories have a > special treatment in implicit rules (they are prepended to > prerequisites separately). The reason this currently works, is because > $* is set incorrectly. By changing it to $(*F), everything is fine > again also with the fixed $* :-).
I realized this morning that my testing was too focused on one particular scenario and that whether % is replaced by $* or by $(*F) should depend on whether check_lastslash is set or not (i.e. whether add_dir is 1 or 0). I'll send a revised patch soon. - Jouke _______________________________________________ Bug-make mailing list Bug-make@gnu.org https://lists.gnu.org/mailman/listinfo/bug-make
