Overlay support broken (Re: [patch] [2/2] Discontiguous PSYMTABs (psymtabs->symtabs by addrmap))

["Ulrich Weigand" <uweigand@de.ibm.com>]

Jan Kratochvil wrote:

> 	* symtab.c: Include "addrmap.h"
> 	(find_pc_sect_psymtab): Support reading the field PSYMTABS_ADDRMAP.
> 	Move the psymtab locator into ...
> 	(find_pc_sect_psymtab_closer): ... a new function.

It appears this change broke overlay support.  In fact, it looks like the
main idea of this patch is fundamentally incompatible with overlays:

> +    /* Map addresses to the entries of PSYMTABS.  It would be more efficient to
> +       have a map per the whole process but ADDRMAP cannot selectively remove
> +       its items during FREE_OBJFILE.  This mapping is already present even for
> +       PARTIAL_SYMTABs which still have no corresponding full SYMTABs read.  */
> +
> +    struct addrmap *psymtabs_addrmap;

In the presence of overlays, a single address can map to *multiple* PSYMTABS
corresponding to multiple overlay segments optionally loaded at that address.

The old code chose between them using the "section" argument to 
find_pc_sect_psymtab; it seems the new code just ignores this argument:

  /* Try just the PSYMTABS_ADDRMAP mapping first as it has better granularity
     than the later used TEXTLOW/TEXTHIGH one.  */

  ALL_OBJFILES (objfile)
    if (objfile->psymtabs_addrmap != NULL)
      {
        struct partial_symtab *pst;

        pst = addrmap_find (objfile->psymtabs_addrmap, pc);
        if (pst != NULL)
          {
            /* We do not try to call FIND_PC_SECT_PSYMTAB_CLOSER as
               PSYMTABS_ADDRMAP we used has already the best 1-byte
               granularity and FIND_PC_SECT_PSYMTAB_CLOSER may mislead us into
               a worse chosen section due to the TEXTLOW/TEXTHIGH ranges
               overlap.  */

            return pst;
          }
      }


Any suggestions how to fix this?

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com