Re: add INVERT to re-search-{forward, backward} (was: occur API change proposal)

[Juri Linkov <juri@jurta.org>]

>>> I propose changing the occur API to allow an optional predicate
>>> function.  For example, this is useful to invert a regular expression so
>>> you don't have to run occur on the whole buffer matching everything,
>>> then limit the results.
>
> JL> The occur engine currently hard-codes the function call of
> JL> `re-search-forward'.  Instead of this, we could do the same
> JL> as is done already in isearch for customization of the search
> JL> function using a variable `isearch-search-fun-function', and
> JL> add a new analogous variable `occur-fun-function'.
>
> JL> Even though such a variable will be useful for other purposes,
> JL> no simple function will allow inverting a regular expression.
> JL> Implementing this feature requires significant modification in
> JL> the logic of the occur engine: either changing the algorithm to
> JL> call `looking-at' at every line (very inefficient), or recording
> JL> a position of the previous match and outputing lines between it
> JL> and the current position (not easy to customize).
>
> After thinking about this (yes, it sat in my TODO queue for a while) I
> realize the problem is fundamental: Emacs doesn't have a way to invert
> at the re-search-{forward,backward} API level.  There's no way to
> automatically invert a regular expression into another regular
> expression, so this change must happen at the API level to avoid
> unpleasant workarounds as Juri describes.
>
> I propose simply adding an optional INVERT parameter to those two
> functions.  The API cost is small with an optional parameter, though I
> don't know if performance will be affected by the extra check on every
> search cycle.
>
> Once this API is in place, making the change in occur-mode and other
> places to use the new API is very easy.

It seems such an API with a new INVERT parameter will be useless
because these two character-based search functions will find too much
matches. For example, calling it with the parameter "def" on the string
"abcdefghi" will match in it 6 places, i.e. everywhere except the
position under the character "d".  Thus this will provide no useful
information to occur.

Instead of this, I think the change should be done at the occur level,
since occur's algorithm is line-based and can skip lines that contain
a given regexp.  This could be similar to the algorithm used by the
command `flush-lines' in the same file.  `occur' currently works like the
command `keep-lines' that keeps lines containing the specified regexp.
But with a new INVERT argument, `occur' could work like `flush-lines' that
keep everything except lines containing the specified regexp.

-- 
Juri Linkov
http://www.jurta.org/emacs/