I wrote:
> I looked for rules with this bug by looking for comments that
> mention parser/Makefile. There may well be some more, but I have
> no good idea how to find them --- any thoughts?
I realized that grepping for line-ending semicolons in makefiles
would be a pretty efficient way to check this. Doing so reveals
these additional trouble spots:
src/Makefile.shlib:324:$(stlib): $(shlib) ;
src/Makefile.shlib:361:$(stlib): $(shlib) ;
src/backend/Makefile:79:libpostgres.a: postgres ;
src/backend/Makefile:89:libpostgres.a: postgres ;
src/test/isolation/Makefile:46:specparse.h: specparse.c ;
src/interfaces/ecpg/preproc/Makefile:42:preproc.h: preproc.c ;
I'm not too excited about the libpostgres.a cases, but the last
two are definitely hazards for VPATH builds, and the two cases in
Makefile.shlib might be worth fixing too.
Also, I trolled the archives for possible reports of actual problems
of this ilk. I found this:
https://www.postgresql.org/message-id/569C22C0.70404%40lucee.org
which certainly looks exactly like the sort of behavior I'd expect,
if there'd been a timestamp problem in the 9.5.0 tarball. But the
rules around *dll.def files don't look like there's any such bug.
regards, tom lane