Ligatures are single-character replacements of strings. Examples of ligatures: replacing "alpha" with the alpha symbol and "!=" with the a slashed equal sign. See Coding with Mathematical Notation for details and pictures.
There is a serious flaw with ligatures - either the indentation you see with ligatures or without ligatures is correct, not both. So if someone that does not use ligatures works on your code, your indentation's will not match. An example:
;; True indentation, what you want others to see (alpha b c) ;; Emacs indentation, what you want to see when working (a b c)
This problem significantly hampers ligature adoption.
I do not believe any editor implements a solution to ligatures such that you see the indentation you want to see, while the true indentation remains correct.
I present a proof-of-concept solution to ligature spacing,
How Emacs displays text
text-properties with strings. A property can be anything.
Some property names are special and tell Emacs to handle the text in a
particular way, like
face for how a text is highlighted.
overlay has associated text-properties but is buffer-local. So when we move
that text to another buffer, if that overlay had a face, then that face would
not be carried over.
Properties to be aware of:
display: How Emacs displays that region, can be any string.
invisible: Whether the text should be displayed.
modification-hooks: When text in the overlay is edited, run these hooks.
evaporate(overlays) : Once the overlay is "done-with", delete the overlay.
compose-region is similar to
display in that the composed region
is displayed as (possibly many) characters. Current implementations of ligatures
all leverage compose-region by searching the buffer for say alpha and composing
from alphas beginning to end point the Unicode symbol for alpha.
There are several important distinctions between
- Indentation uses the composed character for indenting while the text-property display indents with the true, original string.
- Composition cannot be set for overlays. The internal
unlike all other properties, cannot be put manually.
- Editing within a composed region will undo the composition while one must delete the whole region with the display property to undo the display.
Working through a solution
To compose or display the ligature?
Because composition adjusts the underlying indentation, it cannot be used for a ligature spacing solution. Indentation cannot be adjusted in a major-mode agnostic manner. Indentation always considers the true number of characters preceding the text on the line, so dynamically adding invisible spaces will not work.
But how to make editing a display behave like a composition?
It is a serious issue to have to delete the whole text for the ligature to disappear.
The solution is the
(defun lig-mod-hook (overlay post-mod? start end &optional _) (when post-mod? (overlay-put overlay 'display nil) (overlay-put overlay 'modification-hooks nil))) ; force evaporation (overlay-put lig-overlay 'modification-hooks '(lig-mod-hook))
Now editing text with the display property will behave as desired.
So how to visually collapse the indentation?
We could set
invisible on the first 5 spaces of the line to collapse the
visual indentation by 5. But the invisible property will modify subsequent
line's indentation by 5 fewer (if necessary), an issue that cannot be resolved
as we cannot determine in general the "if necessary" part.
The trick is to make the 5 first spaces display as one space. Because display doesn't modify indentation, subsequent lines will be indented properly.
(overlay-put space-overlay 'display " ")
How do we determine the indentation we want to see then?
We let Emacs do the work - we create a mirror buffer where the ligatures are actually composed and compare the differences in indentation.
Overlays are not just buffer-local, they also do not transfer to indirect
buffers. Ideally we would have a hidden indirect buffer where we keep ligatures
composed instead. Unfortunately, since the
composition text property is
special, it can only be set with
compose-region which does not work for
Further, calculating indentation always adjusts the indentation. The significance is that whenever we indent the indirect buffer, all the text will move back-and-forth. So indirect buffers are out.
Instead we create temporary buffers for the composition and retrieve an alist of lines and their composed indentations.
A working example
The current ligature snippets floating around hack
font-locks to perform the
ligature substitutions. I recently became familiar with context-sensitive syntax
highlighting via the
syntax-propertize-function in my work on
I develop a minimal major-mode
lig-mode that uses the syntax function to
First we setup a basic major-mode for testing.
(provide 'lig-mode) (add-to-list 'auto-mode-alist '("\\.lig\\'" . lig-mode)) (define-derived-mode lig-mode fundamental-mode "Lig" (setq-local indent-line-function 'lisp-indent-line) (setq-local syntax-propertize-function 'lig-syntax-propertize-function))
This is a proof-of-concept - we implement spacing for a single ligature for now. Lets replace "hello" with a smiley face.
(defun lig--match-lig (limit) (re-search-forward (rx word-start "hello" word-end) limit t)) (setq lig-char #x263a) (setq lig-str "☺")
Determining the indents we want to see
We copy the buffer contents to a temporary buffer, search and compose the symbols, indent the buffer, and copy the indentation for each line.
(defvar lig-diff-indents nil) (defun lig-get-diff-indents () (setq lig-diff-indents nil) (save-excursion ;; Compose the ligatures (goto-char (point-min)) (while (re-search-forward (rx word-start "hello" word-end) nil t) (compose-region (match-beginning 0) (match-end 0) lig-char)) ;; Change indent to match the composed symbol (indent-region (point-min) (point-max)) ;; Build an alist of line and indention column (goto-char (point-min)) (setq line 1) (while (< (point) (point-max)) (push (cons line (current-indentation)) lig-diff-indents) (forward-line) (setq line (1+ line))))) (defun run-lig-get-diff-indents () (let ((true-buffer (current-buffer))) (with-temp-buffer (fundamental-mode) (setq-local indent-line-function 'lisp-indent-line) (insert-buffer-substring-no-properties true-buffer) (lig-get-diff-indents))))
Bringing it together
For details on how
syntax-propertize-function works, check this post.
Whenever we edit the buffer this hook will run, recalculating and visually collapsing all the leading spaces as needed.
(defun lig-syntax-propertize-function (start-limit end-limit) ;; Make sure visual indentations are current (run-lig-get-diff-indents) (save-excursion (goto-char (point-min)) (while (lig--match-lig end-limit) (let ((start (match-beginning 0)) (end (match-end 0))) (unless (-contains? (overlays-at start) lig-overlay) ;; Create and set the lig overlays if not already set (setq lig-overlay (make-overlay start end)) (overlay-put lig-overlay 'display lig-str) (overlay-put lig-overlay 'evaporate t) (overlay-put lig-overlay 'modification-hooks '(lig-mod-hook))))) ;; Remove all spacing overlays from buffer (remove-overlays nil nil 'invis-spaces t) ;; Recalcualte and add all spacing overlays (goto-char (point-min)) (setq line 1) (while (< (point) (point-max)) ;; Don't add the spacing overlay until we indent (unless (> (+ (current-indentation) (point)) (point-max)) (let* ((vis-indent (alist-get line lig-diff-indents)) (num-spaces (- (current-indentation) vis-indent)) (start (point)) (end (+ num-spaces (point)))) ;; only add invisible spaces if the indentations differ (unless (<= num-spaces 1) (setq space-overlay (make-overlay start end)) (overlay-put space-overlay 'invis-spaces t) (overlay-put space-overlay 'display " ") (overlay-put space-overlay 'evaporate t)) (setq line (1+ line)) (forward-line))))))
lig-mode to see:
;; The true text (hello how are you (hello hi again)) ;; What we see (☺ how are you (☺ hi again))
The indentation we see is not the true indentation anymore!
The full and current code is hosted here.
The missing space on the second hello is a bug. There are many issues with this implementation - this is a proof of concept. I suspect a completely correct solution to be still some time and effort away, if only because this approach is incredibly inefficient.
This post shows that we maybe can have our cake and eat it too in regards to ligatures.