Automatic line breaking sometimes results in an unpaired end of paragraph tag

In two specific cases line breaking results in an invalid markup. Both cases arise from the filter not removing all empty paragraphs, and then removing unneeded paragraph tags (supposedly enclosing block elements) one by one.

1. An (almost) empty paragraph in a block element: <div><p></p></div>
2. Nonprocessed content (pre, script, style) in a blockquote: <blockquote><pre>aaa</pre></blockquote>

_filter_autop in filter.module

      ...
      $chunk = preg_replace('|<p>\s*</p>\n|', '', $chunk);
      ...
      $chunk = preg_replace('!<p>\s*(</?'. $block .'[^>]*>)!', "$1", $chunk);
      $chunk = preg_replace('!(</?'. $block .'[^>]*>)\s*</p>!', "$1", $chunk);

First line fails to remove an empty paragraph not followed by a newline, the second line removes the paragraph start (p is also a block tag), and the last fails to remove the end tag. Under certain conditions the filter itself makes empty paragraphs not followed by a newline.

This can be fixed by making the newline after empty paragraph optional (adding a '?' to the regular expression), and moving the empty paragraphs removing three lines down (to solve the blockquote problem).

Attaching a patch to perform those changes. You can find it running here: http://www.wrwrwr.org/node/10. Same code seems to be present in 5.x and 6.x branches, however I've only tested with 5.6. The second patch does the same and adds some comments — it's not so easy to quickly read all those regular expressions ;)

Part of this issue has been reported already: http://drupal.org/node/177139.