On Tue, Mar 18, 2014 at 09:44:43 -0600, Eric Blake wrote:
> On 03/18/2014 09:32 AM, Jiri Denemark wrote:
> > On Tue, Mar 18, 2014 at 06:27:48 -0600, Eric Blake wrote:
> >> On 03/18/2014 05:00 AM, Jiri Denemark wrote:
> >>
> >>>> Why don't we just avoid the whole issue by removing use of abs_srcdir
> >>>> and abs_builddir. Can this rule:
> >>>>
> >>>> $(abs_builddir)/cpu/cpu_map.xml:
> >>>> $(AM_V_GEN)ln -s $(abs_srcdir)/cpu/cpu_map.xml $@
> >>>>
> >>>> be just changed to
> >>>>
> >>>> cpu/cpu_map.xml:
> >>>> $(AM_V_GEN)ln -s $(srcdir)/cpu/cpu_map.xml $@
> >>>
> >>> That's what I tried first but it does not work at all. I don't
> >>> understand why but make thinks cpu/cpu_map.xml target is uptodate even
> >>> though the file does not exist in builddir.
> >>
> >> That would be VPATH rewriting at play. Does $(builddir)/cpu/cpu_map.xml
> >> fare any better?
> >
> > No, that doesn't work either.
> >
> > ...
> >> But maybe that's some food for thought - instead of having a rule that
> >> uses a direct file name, perhaps you can instead have a witness rule on
> >> a stamp file name, where we write the link ourselves when needed, and
> >> then have all dependencies be on the stamp (which will ALWAYS exist only
> >> in builddir):
> >>
> >> cpu/cpu_map.xml.stamp:
> >> $(AM_V_GEN)if test -f cpu/cpu_map.xml; then \
> >> :; else \
> >> ln -s `cd $(srcdir) && pwd`/cpu/cpu_map.xml \
> >> cpu/cpu_map.xml; \
> >> fi && touch $@
> >
> > OK, this seems to work. It's uglier and doesn't regenerate the link if
> > someone removes it (the *.stamp file would need to be removed too) but
> > if that's considered a better way compared to setting abs_*dir, I can
> > make a proper patch out of it after testing it in all situations.
>
> Everything we've tried is ugly. At this point, relying on the GNU make
> extension of $(shell), and using = instead of ?=, is probably the least
> bad solution. So feel free to keep testing the .stamp alternative if
> you want, but I'm okay if you check in what you've already tested
> instead of sinking more time into it.
>
> Oh, one other thing I thought of:
> Why not just name the git version cpu_map.xml.in, and use autoconf's
> normal mechanisms to always create the non-.in version in builddir.
> True, there's no @foo@ sequences being replaced, but that way, we don't
> have to worry about $(abs_srcdir) in the makefile at all; we also don't
> have to mess with a .stamp.
OK, I went ahead and pushed the minimal solution, i.e., setting
abs_{src,build}dir = $(shell ...)
After all, having the variables set may save us from similar build
issues should any of the abs variables be useful for other rules.
Jirka
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list