I wrote:
> I looked into this, and find that the cause of the problem is that
> pgindent touched src/interfaces/ecpg/include/sqlca.h, which is copied
> verbatim into preprocessed files by the ecpg preprocessor, so the expected
> files had to change in tandem. This amounts to a dependency, but the make
> rules don't know about it. Should they? That particular file changes so
> seldom that it'd hardly be worth worrying about, but I'm not sure which
> other files can get copied similarly.
While I'm looking at it: there's no dependency forcing the test .c files
to get rebuilt after the ecpg preprocessor changes, either, and that
seems much more likely to be a routine problem.
Arguably, we need some more dependencies in this rule in
ecpg/test/Makefile.regress:
%.c: %.pgc ../regression.h$(ECPG) -o $@ -I$(srcdir) $<
I also notice that some of the subdirectory makefiles that include
Makefile.regress have custom build rules that seem mostly duplicative
of this one, except for passing different switches to ecpg. Those
would likewise need additions to their dependency lists, which suggests
that the "../regression.h" part ought to be wrapped up in some macro.
regards, tom lane