Comment 13 for bug 413278

gdb/dwarf: disable per-BFD resource sharing for -readnow objfiles

    New in v2:

      - Disable sharing only for -readnow objfiles, not all objfiles.

    As described in PR 27541, we hit an internal error when loading a binary
    the standard way and then loading it with the -readnow option:

        $ ./gdb -nx -q --data-directory=data-directory ~/a.out -ex "set confirm off" -ex "file -readnow ~/a.out"
        Reading symbols from /home/simark/a.out...
        Reading symbols from ~/a.out...
        /home/simark/src/binutils-gdb/gdb/dwarf2/read.c:8098: internal-error: void create_all_comp_units(dwarf2_per_objfile*): Assertion `per_objfile->per_bfd->all_comp_units.empty ()' failed.

    This is a recurring problem that exposes a design issue in the DWARF
    per-BFD sharing feature. Things work well when loading a binary with
    the same method (with/without index, with/without readnow) twice in a
    row. But they don't work so well when loading a binary with different
    methods. See this previous fix, for example:

        efb763a5ea35 ("gdb: check for partial symtab presence in dwarf2_initialize_objfile")

    That one handled the case where the first load is normal (uses partial
    symbols) and the second load uses an index.

    The problem is that when loading an objfile with a method A, we create a
    dwarf2_per_bfd and some dwarf2_per_cu_data and initialize them with the
    data belonging to that method. When loading another obfile sharing the
    same BFD but with a different method B, it's not clear how to re-use the
    dwarf2_per_bfd/dwarf2_per_cu_data previously created, because they
    contain the data specific to method A. https://komiya-dental.com/

    I think the most sensible fix would be to not share a dwarf2_per_bfd
    between two objfiles loaded with different methods. That means that two
    objfiles sharing the same BFD and loaded the same way would share a
    dwarf2_per_bfd. Two objfiles sharing the same BFD but loaded with
    different methods would use two different dwarf2_per_bfd structures. http://www.iu-bloomington.com/

    However, this isn't a trivial change. So to fix the known issue quickly
    (including in the gdb 10 branch), this patch just disables all
    dwarf2_per_bfd sharing for objfiles using READNOW.

    Generalize the gdb.base/index-cache-load-twice.exp test to test all
    the possible combinations of loading a file with partial symtabs, index
    and readnow. Move it to gdb.dwarf2, since it really exercises features
    of the DWARF
gdb/dwarf: disable per-BFD resource sharing for -readnow objfiles
     https://www.webb-dev.co.uk/
    New in v2:

      - Disable sharing only for -readnow objfiles, not all objfiles.

    As described in PR 27541, we hit an internal error when loading a binary
    the standard way and then loading it with the -readnow option:
    https://waytowhatsnext.com/
        $ ./gdb -nx -q --data-directory=data-directory ~/a.out -ex "set confirm off" -ex "file -readnow ~/a.out"
        Reading symbols from /home/simark/a.out...
        Reading symbols from ~/a.out...
        /home/simark/src/binutils-gdb/gdb/dwarf2/read.c:8098: internal-error: void create_all_comp_units(dwarf2_per_objfile*): Assertion `per_objfile->per_bfd->all_comp_units.empty ()' failed.
     http://www.acpirateradio.co.uk/
    This is a recurring problem that exposes a design issue in the DWARF
    per-BFD sharing feature. Things work well when loading a binary with
    the same method (with/without index, with/without readnow) twice in a
    row. But they don't work so well when loading a binary with different
    methods. See this previous fix, for example: http://www.logoarts.co.uk/

        efb763a5ea35 ("gdb: check for partial symtab presence in dwarf2_initialize_objfile")

    That one handled the case where the first load is normal (uses partial
    symbols) and the second load uses an index. http://www.slipstone.co.uk/

    The problem is that when loading an objfile with a method A, we create a
    dwarf2_per_bfd and some dwarf2_per_cu_data and initialize them with the
    data belonging to that method. When loading another obfile sharing the
    same BFD but with a different method B, it's not clear how to re-use the
    dwarf2_per_bfd/dwarf2_per_cu_data previously created, because they
    contain the data specific to method A. http://embermanchester.uk/

    I think the most sensible fix would be to not share a dwarf2_per_bfd
    between two objfiles loaded with different methods. http://connstr.net/ That means that two
    objfiles sharing the same BFD and loaded the same way would share a
    dwarf2_per_bfd. Two objfiles http://joerg.li/ sharing the same BFD but loaded with
    different methods would use two different dwarf2_per_bfd structures. http://www.jopspeech.com/

    However, this isn't a trivial change. So to fix the known issue quickly
    (including in the gdb 10 branch), this patch just disables all http://www.wearelondonmade.com/
    dwarf2_per_bfd sharing for objfiles using READNOW.

    Generalize the gdb.base/index-cache-load-twice.exp test to test all http://www.compilatori.com/
    the possible combinations of loading a file with partial symtabs, index
    and readnow. Move it to gdb.dwarf2, since it really exercises features
    of the DWARF http://www-look-4.com/