Re: [RFA/PATCH] PR/9711: quadratic slowdown for deep stack traces

[Doug Evans <dje@google.com>]

On Thu, Sep 3, 2009 at 11:36 AM, Joel Brobecker<brobecker@adacore.com> wrote:
> 2009-09-03  Joel Brobecker  <brobecker@adacore.com>
>
>        Avoid quadratic behavior when computing the value of a register.
>        * frame.c (frame_cache): New static constant.
>        (frame_cache_add, frame_cache_find, frame_cache_invalidate):
>        New functions.
>        (get_frame_id): Minor reformatting. Add the frame to the frame cache.
>        (frame_find_by_id): Search the frame cache first before walking all
>        frames starting from te current_frame.
>        (reinit_frame_cache): Add call to frame_cache_invalidate ();

Hi.

Nit: It seems like there's a redundant call to frame_id_eq in frame_find_by_id.

+/* Search the frame cache for an entry with the given frame ID.
+   If found, return that frame.  Otherwise return NULL.  */
+static struct frame_info *
+frame_cache_find (struct frame_id id)
+{
+  if (frame_cache && frame_id_eq (frame_cache->this_id.value, id))
+    return frame_cache;
+
+  return NULL;
+}

@@ -514,6 +547,11 @@ frame_find_by_id (struct frame_id id)
   if (!frame_id_p (id))
     return NULL;

+  /* Try using the frame cache first.  */
+  frame = frame_cache_find (id);
+  if (frame && frame_id_eq (frame->this_id.value, id))
+    return frame;
+
   for (frame = get_current_frame (); ; frame = prev_frame)
     {
       struct frame_id this = get_frame_id (frame);