panels_export_display does not respect pane position and order

ChristianAdamski, this is an issue only when using Features Override to override a Panels display exported by another module, correct?

+++ b/panels.module
@@ -329,9 +329,7 @@ function panels_permission() {
 function panels_flush_caches() {
-  if (db_table_exists('cache_panels')) {
-    return array('cache_panels');
-  }
+  return array('cache_panels');
 }

Why did the table check get removed from flush cache hook?

This patch needed to be rerolled, but once applied it worked like a charm. No more mini panels always stuck in overridden mode.
Attached patch doesn't change any code compared to #6.

Needs more tests before committing.

@moonray: did it following occur for you in practice that positions were conflicting? I am referring to the situation that is handled in code near the comment // Determine if position already defined and not conflicting..

That piece of code lets a pane take over a position when there are conflicting position. I am just reading the patch and it seems to me that one conflict may lead to more: while iterating over the panes one conflicted pane may take over the position of the next pane, which then becomes conflicted too.

It seems panels lets the database decide what happens when multiple panes have the same position, see panels_load_displays(). Maybe it is better for this patch to sort on $pane->position and then on $pane->id if there are duplicates to match that behavior in a deterministic way. Perhaps the panels maintainers can be convinced to add pid to the list of columns to sort on. But then again, I don't know how often this occurs.

Also:

foreach ($output_panes as $position => $output_pane) {
  $output .= $output_pane;
}

can be simplified to

$output = implode('', $output_panes);

Please see the attached patch for my solution to this problem, which is, in my opinion, a cleaner and simpler execution of the same idea.

@ChristianAdamski : I am curious to hear your opinion on my approach.

Thanks for your quick response.

I have found a small bug in my patch as I broke the handling of $title_pid.