Could you include the test cases you described in your email (& any others you think might be relevant) as actual test cases in your patch?
On Sat, Sep 15, 2012 at 3:02 AM, Dimitry Andric <[email protected]> wrote: > As reported in PR 13851, clang (trunk r163208) seems to ignore any -MF > option that is used in combination with -E -M or -E -MM. For example: > > $ clang -E -MM -MF .depend helloworld.c > clang: warning: argument unused during compilation: '-MF .depend' > > and no dependency file is generated, the dependency information is > dumped to standard output instead. > > It does work when also compiling, e.g if -MD or -MMD is used: > > $ clang -MMD -MF .depend -c helloworld.c > $ cat .depend > helloworld.o: helloworld.c > > Looking at Clang::AddPreprocessingOptions() in lib/Driver/Tools.cpp, it > seems the check for a -MF option is done too late: if dependency output > is requested in any way, the DepFile is always the default filename, > which is standard output if -E is used. > > I propose the attached patch to fix this. > > -Dimitry > > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits > _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
