On 09 Jan 2024 07:34, Tom Tromey wrote:
> >>>>> "Mike" == Mike Frysinger <vapier@gentoo.org> writes:
> 
> Mike> The lbasename function from libiberty provides the same API as this
> Mike> custom function.  The common/ code already made the switch, so make
> Mike> the same change to the ppc code to avoid target duplication.
> 
> When updating and rebuilding in an existing tree, some patch -- I guess
> probably this one, I didn't bisect -- caused a build failure in sim/ppc
> because filter_filename was missing.  Removing sim/ppc/*.o and
> rebuilding fixed it, but I suppose this means there is a missing
> dependency somewhere.

ppc/ used to generate a filter_host.c file in the output dir, and compile the
.o from that.  it switched to using the igen/filter_host.c in the source dir.
when i've migrated code like this, i haven't left behind clean rules to delete
files that only mattered to old builds since any new builds would never see
that stuff.  but it means bisect/incremental (re)builds across that timespan
might break.  i don't know what the general policy is in the tree for these
cases, but it seems like we just don't bother.

also a lot of sim rules rely on VPATH to locate sources.  e.g. the makefiles
write ppc/foo.c instead of $(srcdir)/ppc/foo.c.  the only time we really use
$(srcdir) is with generated files we commit to the tree.  i think this pattern
is standard in the binutils-gdb tree.  but it also contributes to incremental
(re)build failures like this.
-mike