[Bug c/48083] New: DEPENDENCIES_OUTPUT + -no-integrated-cpp produces bad dependencies

["cgd at google dot com" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48083

           Summary: DEPENDENCIES_OUTPUT + -no-integrated-cpp produces bad
                    dependencies
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: cgd@google.com


gcc version 4.6.0 20110311 (experimental) (GCC), but I reproduced this all the
way back to 4.2.x.


$ : > null.c

# Hey, we have an empty file!

$ DEPENDENCIES_OUTPUT=/dev/stdout ./gcc -c null.c
null.o: null.c

# Yeah!  Those are my deps all right!

$ DEPENDENCIES_OUTPUT=/dev/stdout ./gcc -c null.c -no-integrated-cpp
null.o: null.c
cccUba7s.o: /tmp/cccUba7s.i

# Huh?!?!  WHAT?!?!?!


also happens with -save-temps or -traditional-cpp instead of
-no-integrated-cpp.

also happens with SUNPRO_DEPENDENCIES instead of DEPENDENCIES_OUTPUT.


what's happening is that -no-integrate-cpp causes cc1 to be invoked twice:
once with the original source file,
and once with the .i file output by the first invocation.

While the second invocation would have such as -MMD stripped, it still have has
the DEPENDENCIES_OUTPUT or SUNPRO_DEPENDENCIES variables set in its
environment.

there's a horrible kludge that can be used to make this work as expected
(strategically insert -MF/dev/null into the second cc1 invocations in
default_compilers), that would be Wrong.

Probably a better solution would be to add a new internal option, to indicate
that the 'stage' of the compilation, and then use that to avoid processing the
env vars.

(I would suggest a new -M option to cancel dependency generation, and then pass
that internally ... but i think that could be really confusing because of the
way gcc handles (or doesn't handle) relative parameter position.)