Re: [patch+7.3] gdbindex regression: stabs forgotten [Fix crash of gdb save-index on a STABS file]

[Tom Tromey <tromey@redhat.com>]

>>>>> "Jan" == Jan Kratochvil <jan.kratochvil@redhat.com> writes:

Tom> On the reader side, the code in elfread.c is written to read STABS
Tom> first -- and skip using the index if any are found.  This handles the
Tom> problem that an objfile can only have one set of quick functions.

Jan> I do not see it implemented.  This is a regression.

I must have dropped something somewhere along the line, oops.
Or maybe I just botched the logic when I added lazy psymtabs.

Jan>    if (dwarf2_has_info (objfile))
Jan>      {
Jan> -      if (dwarf2_initialize_objfile (objfile))
Jan> +      /* elf_sym_fns_gdb_index cannot handle simultaneous non-DWARF debug
Jan> +	 information present in OBJFILE.  If there is such debug info present
Jan> +	 never use .gdb_index.  */
Jan> +
Jan> +      if (!objfile_has_partial_symbols (objfile)
Jan> +	  && dwarf2_initialize_objfile (objfile))

This makes the case of "stabs, plus gnu index, plus -readnow" a little
worse.  

It is hard to imagine anybody caring about this kind of thing though.
I just thought it was funny.

Jan>    /* If we have not read psymbols, but we have a function capable of
Jan>       reading them, then that is an indication that they are in fact
Jan>       available.  */
Jan> -  if ((objfile->flags & OBJF_PSYMTABS_READ) == 0)
Jan> -    return objfile->sf->sym_read_psymbols != NULL;
Jan> +  if ((objfile->flags & OBJF_PSYMTABS_READ) == 0
Jan> +      && objfile->sf->sym_read_psymbols != NULL)
Jan> +    return 1;

I think this change requires an update to the comment and also maybe to
the corresponding comment in symfile.h.

Tom