Also changed completion code to check registered status and
avoid completion logic if not registered, which would be the case if
_begin failed.

added to try to reduce computational overhead only works correctly for
HTTP and broke FTP and other stateful protocols. I took the caching back
out - maybe we can try to C-ify the method at some point as a better
alternative.

Fixed a find support bug which caused find permissions not to get correctly registered. Also fixed a find frame URL bug that became apparent after fixing the first bug.

Thanks Brian Hooper.

Common Long Format.

reload extenal method files every time.

The problem was that the entire os.stat() record was being compared,
and that includes the access time, which of course changes every time
the method is loaded - self fulfilling prophecy, and all.

Instead, we now just stash and compare the modification time
(ST_MTIME) field, and get rid of the extraneous (expensive) reloads.

ZTables)

that requires climbing the entire aq chain on every call (and it gets
called a *lot*). A little investigation revealed that it is called
over and over for the same objects during a request for complex pages,
so a cache was added that can cut the work down to about 10% of its
current level.

is legit HTML4. (C650)

  (in order to enable REMOTE_USER mode) also opens the monitor to any
  user connecting through localhost.

Fixed a bug which allowed unauthorized options to be displayed when adding a Folder. Now you only see
options to add an index_document and/or a user folder if you have adequate permissions.

NameError in manage_edit.

other strangosities.

argument is "self" to prevent things like hijacking external methods.

into themselves or their subobjects.

cloned requests do not leak.

duplicated code. The resolve_url of the REQUEST now correctly avoids aborting
the current transaction if an error occurs, so we just use that.

called without necessarily aborting the current transaction. This is
required by things like the resolve_url machinery, where there is a
need to call the exception() method to get a meaningful error. The
resolve_url method in HTTPRequest.Request has been changed to take
advantage of this. The existing semantics of RESPONSE.exception()
are unchanged - it will automatically abort just as it did before
unless you pass the keyword arg abort=0 when calling the method.

during resolve_url caused the exception method of the (cloned) response
object to be called. This mean that as soon as an error occurred during
resolve_url, the transaction would be aborted and things got extremely
weird since objects that had been uncataloged would suddenly be back
(due to the abort).