Re: [PATCH] [GDB 18] gdb: resolve class name via DW_AT_signature in cooked index

[Tom Tromey <tom@tromey.com>]

>>>>> "Andrew" == Andrew Burgess <aburgess@redhat.com> writes:

Andrew> The problem is that the cooked index is unable to determine the name
Andrew> of the parent class `base1` in this case, and so decides not to index
Andrew> any of the member functions.

Andrew> 2. cooked-indexer.c (scan_attributes): Handle DW_AT_signature by
Andrew>    looking up the signatured_type via lookup_signatured_type and
Andrew>    constructing a section_and_offset origin pointing to the type DIE
Andrew>    in the type unit.  Restructure the is_declaration / origin-
Andrew>    following control flow: change the "else if (origin)" to a
Andrew>    standalone "if" so that class declaration stubs marked with
Andrew>    IS_TYPE_DECLARATION can still follow their origin to retrieve the
Andrew>    class name from the type unit.  Add origin.reset() in the other
Andrew>    declaration paths to preserve the original behaviour for non-class
Andrew>    declarations and Ada imports.

It's been a while since I was deep in the indexer, but looking at
done_reading makes me wonder if this approach is safe:

    void
    cooked_index_worker_debug_info::done_reading ()
    {
      /* This has to wait until we read the CUs, we need the list of DWOs.  */
      process_skeletonless_type_units (m_per_objfile, &m_index_storage);

That is, I think skeletonless type units aren't processed until all
other indexing is done.  So if the signatured type appears in one of
these, doesn't this mean the fix will fail?

I don't remember how to set one of these up.

Also my first thought when seeing this patch was that it might not be
thread-safe.  I'm still not completely sure.  Perhaps it's fine because
other type units seem to be (unfortunately) processed serially.

A typical fix for these kinds of issues is to defer some of the work to
the finalization step in the shard.  I guess here the idea would be to
store the signature and the relevant entry in some data structure, then
patch up the parentage or whatever when finalizing.

Tom