This'll work if we remove the .sub files as discussed, with .sub files it
becomes more complicated.

On Tue, Nov 17, 2015 at 3:43 PM Rich Felker <dalias@libc.org> wrote:

> On Tue, Nov 17, 2015 at 06:27:11PM -0500, Rich Felker wrote:
> > Assuming .o files in arch dirs are ok, does the following work
> > (non-VPATH)?
> >
> > %.o: $(srcdir)/%.sub
> >       ...
> > %.o: $(srcdir)/%.s
> >       ...
> > %.o: $(srcdir)/%.c
> >       ...
> >
> > (with appropriate commands, of course). If so, I think I like it a
> > lot, because it factors the rules for file *types* separately from the
> > rules for which files to select (the "tricky" $(OBJS) setting). The
> > only big concern is whether this "tricky" setting is too costly in
> > runtime performance or complexity.
>
> Here's my idea for the 'tricky' setting of OBJS and it actually looks
> very simple and efficient:
>
> ARCH_OBJS = $(wildcard ...)
> BASE_OBJS = $(wildcard ...)
> OBJS = $(ARCH_OBJS) $(filter-out $(ARCH_OBJS:/$(ARCH)/=/),$(BASE_OBJS))
>
> If the above all works I think we have a pretty good path forward.
>
> It does change where the .o files for arch-specific sources live (one
> level deeper in the tree) and requires these dirs to be created in the
> output tree, but I think this is actually an improvement since
> removing an arch source file will result in the base-level .o file
> being built (right now the stale .o file based on the arch asm sticks
> around).
>
> Presumably we can also just do:
>
> LOBJS = $(OBJS:.o=.lo)
>
> although it's my intent to remove LOBJS soon anyway and instead have
> pic/non-pic objs (and only use pic ones by default). How to name them
> will be a fun bikeshed...
>
> Rich
>