RFC: proc-service -vs- multi-inferior

[Tom Tromey <tromey@redhat.com>]

I'd appreciate comments on this patch.

I have been trying multi-inferior debugging quite a bit lately.
I've noticed this message a few times from gdb:

    Cannot find new threads: capability not available

Today I tracked this down.

The problem is that we can end up in thread_db_find_new_threads_2 with a
PTID argument that is not the same as current_inferior.  Then,
libthread_db calls our ps_pglobal_lookup, which tries to look up a
symbol against the wrong inferior.

This fixes the problem by setting the current program space before
calling lookup_minimal_symbol.

I have no idea how to construct a test case for this.  I could reproduce
it reasonably reliably by 'gdb /bin/sh', putting gdb into multi-inferior
mode, and then running dejagnu on part of our test suite.

Tom

2012-03-06  Tom Tromey  <tromey@redhat.com>

	* proc-service.c (ps_pglobal_lookup): Set the current program
	space.

diff --git a/gdb/proc-service.c b/gdb/proc-service.c
index e317414..a1f90a8 100644
--- a/gdb/proc-service.c
+++ b/gdb/proc-service.c
@@ -201,14 +201,24 @@ ps_pglobal_lookup (gdb_ps_prochandle_t ph, const char *obj,
 		   const char *name, psaddr_t *sym_addr)
 {
   struct minimal_symbol *ms;
+  struct cleanup *old_chain = save_current_program_space ();
+  struct inferior *inf = find_inferior_pid (ptid_get_pid (ph->ptid));
+  ps_err_e result;
+
+  set_current_program_space (inf->pspace);
 
   /* FIXME: kettenis/2000-09-03: What should we do with OBJ?  */
   ms = lookup_minimal_symbol (name, NULL, NULL);
   if (ms == NULL)
-    return PS_NOSYM;
+    result = PS_NOSYM;
+  else
+    {
+      *sym_addr = core_addr_to_ps_addr (SYMBOL_VALUE_ADDRESS (ms));
+      result = PS_OK;
+    }
 
-  *sym_addr = core_addr_to_ps_addr (SYMBOL_VALUE_ADDRESS (ms));
-  return PS_OK;
+  do_cleanups (old_chain);
+  return result;
 }
 
 /* Read SIZE bytes from the target process PH at address ADDR and copy