[RFA] Fix leaks in macro definitions.

[Tom Tromey]

>>>>> "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:

Simon> Ok.  Fixing this in the splay tree code would be a quite long task
Simon> (reviewing all usages in binutils-gdb and gcc, I don't think this code
Simon> is available externally?), so I am ok with this patch which fixes the
Simon> issue in the mean time.  Tom?

Actually, I just misunderstood the patch and/or the splay-tree API.

For some of the patch, the splay tree is doing the right thing.
This applies to the macro_define_* patches.


This though:

@@ -841,8 +850,10 @@ macro_undef (struct macro_source_file *source, int line,
          arguments like '-DFOO -UFOO -DFOO=2'.  */
       if (source == key->start_file
           && line == key->start_line)
-        splay_tree_remove (source->table->definitions, n->key);
-
+	{
+	  splay_tree_remove (source->table->definitions, n->key);
+	  macro_tree_delete_key (key);
+	}
       else
         {
           /* This function is the only place a macro's end-of-scope


This one seems like it is definitely a splay-tree bug.  The issue is
that it deletes a node but not the node's key.

I think it would be best by far to fix this in splay_tree_remove.
I agree it's hard, but working around this seems worse to me.

Tom