[Bug gdb/26828] SIGSEGV in follow_die_offset dwarf2/read.c:22950

[simark at simark dot ca]

https://sourceware.org/bugzilla/show_bug.cgi?id=26828

--- Comment #9 from Simon Marchi <simark at simark dot ca> ---
Nils, could you try the following patch?

The issue lies with the fact that we have a DIE in a CU A referring to a DIE in
a CU B (using DW_AT_abstract_origin).  The DIEs for CU B are not loaded yet (or
rather, have been unloaded due to DWARF CU aging stuff), and we try to load
them using:

      /* If necessary, add it to the queue and load its DIEs.  */
      if (maybe_queue_comp_unit (cu, per_cu, per_objfile, cu->language))
        load_full_comp_unit (per_cu, per_objfile, per_objfile->get_cu (per_cu),
                             false, cu->language);

Since the CU is already queued for expansion, maybe_queue_comp_unit returns
false and we don't call load_full_comp_unit.  I think maybe_queue_comp_unit is
the wrong thing to call here, because "queuing for symtab expansion" is
unrelated to "loading the DIEs in memory".  We should rather check: "are the
DIEs for per_cu loaded into memory yet?  if not do it now", which is what the
patch below implements.


@@ -22937,12 +22939,15 @@ follow_die_offset (sect_offset sect_off, int
offset_in_dwz,
       per_cu = dwarf2_find_containing_comp_unit (sect_off, offset_in_dwz,
                                                 per_objfile);

-      /* If necessary, add it to the queue and load its DIEs.  */
-      if (maybe_queue_comp_unit (cu, per_cu, per_objfile, cu->language))
-       load_full_comp_unit (per_cu, per_objfile, per_objfile->get_cu (per_cu),
-                            false, cu->language);
-
       target_cu = per_objfile->get_cu (per_cu);
+      if (target_cu == nullptr)
+       {
+         load_full_comp_unit (per_cu, per_objfile, per_objfile->get_cu
(per_cu),
+                              false, cu->language);
+         target_cu = per_objfile->get_cu (per_cu);
+       }
+
+      gdb_assert (target_cu != nullptr);
     }
   else if (cu->dies == NULL)
     {

-- 
You are receiving this mail because:
You are on the CC list for the bug.