On Mon, Sep 14, 2009 at 10:35 AM, Joel Brobecker wrote: >> I've just checked in both patches. Hopefully this thread will end here, >> it's been long enough :-) > > There is still the OBJF_NOT_MAPPED patch. Given this subthread: http://sourceware.org/ml/gdb-patches/2009-09/msg00307.html I don't believe OBJF_NOT_MAPPED is the right way to go. I agree with Tristan that we should separate the 'read objfile for symbols, sections, etc.' from read an object file just for its debug bits. This would allow: - Darwin "dwarf debug is found in the separate *.o files", - Solaris cc/CC without '-xs' (dwarf or stabs debug is found in original *.o files -- very similar to Darwin), - We could then use the same on Linux and drop the separate_debug_objfile/separate_debug_objfile_backlink objfile members, as well as filter_debuginfo_sections(). Note that current checked in code on Darwin does not fail any asserts, it just doesn't work correctly and issues bazillion of complaints. Given above, I don't think committing/testing the OBJF_NOT_MAPPED patch is the right way to go -- it's more of a hack then a solution. However, if someone needs a hack, attached is the current state of the patch. Christian Thalinger tested essentially this version on his Darwin/Java build, and reported success. Cheers, -- Paul Pluzhnikov 2009-09-14 Paul Pluzhnikov * objfiles.h (OBJF_NOT_MAPPED): New define. * objfiles.c (update_section_map): Ignore not mapped objfiles. * machoread.c (macho_oso_symfile, macho_oso_symfile): Adjust.