On Jul 20, 2012, at 9:02 AM CDT, Dave Goodell wrote: > On Jul 12, 2012, at 3:43 PM CDT, Stefano Lattarini wrote: > >> In particular, I propose the attached patch. Can you give it a try to check >> whether it works? > > I'm currently in the process of running the test suite with the given patch, > but it almost certainly doesn't work with the PGI 10.x compilers that I have > handy. I just tried a manual test of the compiler with "-MD,asdf.d" and it > didn't complain but it also did not generate the expected dependency file.
Surprisingly, the test suite did pass. But I can't figure out which tests I should even expect to fail here. A good test for this case needs to use some per-target flags to build two objects with different names from the same source. Ideally it would set a -DINCLUDES1 or -DINCLUDES2 depending on which object is being built, and then have the source file include different header files. Stefano, can you point me towards the correct tests to beef up here? I can't promise when I'll get to it though. In the meantime, I propose pushing the original patch (plus your formatting fixup) and then revisiting the "-MD,asdf.d" approach and tests afterwards. Does that sound reasonable? -Dave