Re: [PATCH] {yacc-work} coverage: more on 'yacc -d' and recovery from deleted headers

[Stefano Lattarini]

On Saturday 29 January 2011, Ralf Wildenhues wrote:
> * Stefano Lattarini wrote on Sat, Jan 29, 2011 at 01:48:16PM CET:
> > On Saturday 29 January 2011, Ralf Wildenhues wrote:
> > > FYI, I think the test has some overlap with one of the make -n tests
> > > that ensures the deleted header recovery doesn't build files in dry
> > > run mode.
> > >
> > Not exactly; it partly (and minimally) overlaps with `yacc7.test' (which
> > is an hodgepodge anyway), and with `yacc8.test' (which checks the
> > behaviour of the subdir-objects option).
> 
> Ah no, I meant yaccdry.  Sorry for not looking it up before.
> But the overlap is not a problem.
> 
> > > > +BUILT_SOURCES = parse1.h p2-parse2.h
> > > 
> > > Interestingly, the test also passes for me when I omit the BUILT_SOURCES
> > > line.  Bug in the test or just incidental?
> > >
> > Bug, sort of.  The test with BUILT_SOURCES removed passes for me too
> > if 'make' is not run in parallele mode, but fails with:
> > 
> >   $ MAKE='make -j2' ./yacc-deleted-headers.test
> > 
> > By changing the order in which files are listed in the p{1,2}_SOURCES
> > variables, I can make the test fail consistently if the BUILT_SOURCES
> > definition is removed.  See the squash-in below.
> > 
> > But there's more.  If I also make `main3.c' be the first entry in
> > p3_SOURCES, the test consistently fails because make tries to build
> > `main3.o' before the header `parse3.h' is created by yacc.  This
> > is fixed by the squash-in, too.
> 
> > I will push as soon as you ACK the squashed-in diffs.
> 
> I agree with everything except ...
> 
> >  BUILT_SOURCES = parse1.h p2-parse2.h
> > address@hidden@: parse3.h
> 
> ... this line shouldn't be there, other than to workaround
> a bug in automake;
>
Why not?  IMO it increases coverage by showing that, when we know
which files include a yacc-generated header, we can just declare
dependencies directly instead of relying on the $(BUILT_SOURCES)
hack, and still have things work correctly.

> and if it does, then it deserves a comment.
>
Yes, that could be useful.  What about the following squash-in?

-*-*-*-

diff --git a/tests/yacc-deleted-headers.test b/tests/yacc-deleted-headers.test
index ff215c6..6d6a0a1 100755
--- a/tests/yacc-deleted-headers.test
+++ b/tests/yacc-deleted-headers.test
@@ -39,8 +39,15 @@ p3_SOURCES = main3.c parse3.y parse3.h
 p4_SOURCES = parse4.y
 AM_YFLAGS = -d
 p2_YFLAGS = -d
+
 BUILT_SOURCES = parse1.h p2-parse2.h
+
+# When we know which files include a yacc-generated header, we
+# should be able to just declare dependencies directly instead
+# of relying on the BUILT_SOURCES hack, and things should still
+# work correctly.
 address@hidden@: parse3.h
+
 .PHONY: clean-p3 build-p3
 build-p3: p3$(EXEEXT)
 clean-p3:

-*-*-*-

Thanks,
  Stefano