On 05 Feb 2004, storm@cua.dk wrote:

> Maybe a more generic approach could be used:
> 
> Let compile.el define a buffer-local compilation-next-error-function
> variable (default value nil) which can be bound in the *occur*
> buffer to #'occur-next (or whatever).
[...]
> With this approach, compile.el doesn't need to know anything about
> occur mode, and future similar modes can hook themselves into
> next-error without modifying compile.el any further.

I see, that is indeed a much better solution.  I've reworked the patch
to do what you suggest; only the universal-prefix is not handled by
occur-next-error.  It is much better from the perspective of
occur-mode, but I had to modify compilation-find-buffer and
compilation-buffer-p to avoid complicating next-error.  Anyhow, see
what you think.

multi-occur should be handled, I don't know if I need to do anything
there.  It works exactly like occur, according to the docs.

> IMO, it would be cleaner to add (provide 'occur) to replace.el and
> just test with (featurep 'occur).

OK, with the change you suggested above I don't need to do this
anymore, but I'll keep it in mind.  By the way, why doesn't
replace.el (provide 'occur)?  No one needed it?

> The autoload of occur-next should be done as an autoload cookie in
> occur.el But I don't see why do you need it at all?

I was calling occur-next explicitly, so I wanted to make sure there
were no compiler warnings.  I see it's unnecessary.

Thanks
Ted