Grizzl - A fuzzy-search utility for Emacs

The project is deprecated in favour of Ivy

Grizzl is a small utility library to be used in other Elisp code needing fuzzy search behaviour. It is optimized for large data sets, using a special type of lookup table and supporting incremental searches (searches where the result can be narrowed-down by only searching what is already matched).

Usage

Grizzl is still in development, though it is functional at this point, it just needs some work to make the UI more visually appealing and more informative, in addition to providing extension points in its internal minor-mode.

Using the completing-read

The main intended use-case for Grizzl is as a completing-read implementation, like ido-mode or helm. This is straightforward, but does require the preparation of a search index before use. It is assumed the search index will be re-used, though it need not be. For small data sets it may be better to just create an index on the fly.

;; define a search index
(defvar *search-index* (grizzl-make-index '("one" "two" "three" "four")))

;; prompt the use to pick from the index
(grizzl-completing-read "Number: " *search-index*)

The user is presented with the minibuffer and a list of matches, starting with all possible matches. As the user types, the list is reduced by repeatedly fuzzy searching in the index. The selection within the matched results can be changed by using the UP and DOWN arrow keys or C-p and C-n. The user hits ENTER to select the matching result.

If a match was successfully selected, grizzl-completing-read returns it as a string. If not, it returns nil.

Grizzl is case-insensitive by-default. To make it case-sensitive, specify :case-sensitive t when creating the index.

(grizzl-make-index '("One" "TWO" "three" "Four") :case-sensitive t)

No further settings are required for case-sensitivity; the index does the work.

Using the algorithm non-interactively

Grizzl aims to be small and focused, so that it can be used in other projects, without a huge codebase following it around. grizzl provides the functionality needed to incorporate fuzzy search logic into any Emacs Lisp code.

First an index must be created, as this is used in all other functions. The index is an optimization for large data sets, providing close to O(n+m) time lookup complexity, where n is the length of the search term and m is the number of possible matches. This proves to be extremely fast compared with regular expression matching and globbing.

Indexing

To define the index, pass a list of strings to grizzl-make-index.

(grizzl-make-index '("one" "two" "three"))

The returned data structure is used in other Grizzl functions.

If the data set is particularly large, the index may take a few seconds to build, during which time Emacs will appear non-responsive. You can observe the progress building the index by passing a callback function as the keyword argument :PROGRESS-FN. The callback receives two arguments, N and TOTAL, where N is the number of items processed so far and TOTAL is the number of items to be processed.

The following example shows the progress to the user as indexing is done.

(grizzl-make-index huge-list-of-strings
                   :progress-fn (lambda (n total)
                                  (message (format "Indexed %d/%d" n total))))

Of course, in a real-world implementation you'll probably only update the message area every 1000 or so items to avoid flooding the *Messages* buffer.

Indexes are case-insensitive by default, for the most-effective fuzzy-matching in most cases. If you need a case-sensitive index, specify a non-nil keyword argument :CASE-SENSITIVE.

(grizzl-make-index '("One" "TWO" "three" "Four") :case-sensitive t)

Searching

Given your index, ou may now search for something given a fuzzy search term, using grizzl-search.

(defvar *search-result* (grizzl-search "cntrl" *search-index*))

This function returns a new data structure representing the result of the search. You can read the strings from it with grizzl-result-strings.

If you need to run an interactive search, where the search term is changing as it is received from some external input, such as the minibuffer, you should pass each previous result back into grizzl-search as the third argument. While this is not strictly needed, it greatly improves performance on large data sets, since it allows the algorithm to focus only on what is already matched by the previous result.

If the new search term is, for whatever reason, entirely unrelated to the previous search term, there is little to no cost in passing the previous result in any case, since Grizzl will simply rewind its internal result tree and begin a fresh search, as needed.

(defvar *result-1* (grizzl-search "c"    *search-index*))
(defvar *result-2* (grizzl-search "cn"   *search-index* *result-1*))
(defvar *result-3* (grizzl-search "cnt"  *search-index* *result-2*))
(defvar *result-4* (grizzl-search "cntr" *search-index* *result-3*))

Passing nil as the previous result has the same effect as leaving the argument unspecified.

Result reading

Search results from grizzl-search are read with grizzl-result-strings.

(grizzl-result-strings *search-result* *search-index*)

The matching strings are returned ordered according to best match (first).

It is possible (and perhaps desirable, for performance reasons) to only read a subseq of the matched result. Just specify the keyword arguments :START and :END when reading the result strings.

;; returns at most the best 10 results
(grizzl-result-strings *search-result* *search-index*
                       :start 0
                       :end   10)

grizzl/grizzl