bug#20774: auto-fill doesn't work properly when first-line prefix differs in adaptive-fill-mode

[Samuel Freilich <sfreilich@google.com>]

[-- Attachment #1.1: Type: text/plain, Size: 3191 bytes --]

You're right about string-width, since I'm counting number of columns.
Fixed that. (Which required me to deal with the fact that fill-prefix can
be nil.)

Your understanding is correct. My fix avoids do-auto-fill breaking a line
only to create a subsequent line that's as long or longer.

The previous version avoided that in most cases by refusing to break a line
during (or immediately after) the fill-prefix. But that failed to avoid
that problem when:
* It's breaking the first line of a paragraph
* The prefix of that first line doesn't match the fill-prefix for the rest
of the paragraph

I've attached the fixed version of the patch with a hopefully-improved
commit message.



On Wed, Aug 23, 2017 at 10:16 PM, <npostavs@users.sourceforge.net> wrote:

> Samuel Freilich <sfreilich@google.com> writes:
>
> > tab-width = 4 seems consistent with how the rest of the file is
> formatted.
> > I didn't want to change the spacing in lines not touched by my diff.
>
> Hmm, somehow the code that landed in my buffer looked wrongly indented.
> Something got messed up somewhere (possibly by the email system,
> possibly during patch application, I've been experimenting with various
> forms of automation for that), but since the indentation in your next
> patch looks fine there's not really much point in worrying about it.
>
> > I added tests and a change description, and formatted the patch according
> > to the current instructions.
>
> Thanks!
>
> > I made the patch a bit more minimal. I pruned the part that dealt with
> > adaptive-fill-first-line-regexp didn't work as well as expected, since
> > that didn't work as well as expected (it didn't deal with all the
> > complexity possible with adaptive-fill-function). The updated version
> > at least handles cases where the fill-prefix isn't shorter than the
> > first-line prefix. That allowed me to simplify the code quite a bit,
> > since that makes the previous logic for skipping the exact fill-prefix
> > redundant, fill-move-to-break-point already handles the logic of
> > trying to skip at least one word after the start position passed to
> > it. Please take another look.
>
> Your commit message is a bit too focused on the problems of what was
> there previously rather than how your new code is correct.  This
> description in your email is better, but I'm still struggling a bit (I'm
> realizing I was probably a bit too superficial previously, there are a
> lot of interacting options which makes things tricky).
>
> If I understand correctly, the previous code would take the fill-prefix
> as non-breakable if the current line started with the fill-prefix.  Your
> new code takes the first N characters of the line as non-breakable
> (where N is the length of fill-prefix, should this be based on
> string-width instead?).  Is that correct (for both adaptive-fill-mode
> and otherwise)?  If so, please add an explanation of why it's correct to
> the commit message.  If not, I hope we can elaborate the commit message
> until even I can understand what's happening ;)
>
> (Minor commit message format nitpick: the part explaining the changes to
> the function should start with "* lisp/simple.el (do-auto-fill):".)
>

[-- Attachment #1.2: Type: text/html, Size: 3997 bytes --]

[-- Attachment #2: 0002-Do-not-split-line-before-width-of-fill-prefix.patch --]
[-- Type: text/x-patch, Size: 3529 bytes --]

From 42da1d80367b7519b0047328a37d845e78aaa157 Mon Sep 17 00:00:00 2001
From: Samuel Freilich <sfreilich@google.com>
Date: Wed, 23 Aug 2017 13:40:45 -0400
Subject: [PATCH] Do not split line before width of fill-prefix

When auto-filling a paragraph, don't split a line before the width of the
fill-prefix, creating a subsequent line that is as long or longer (Bug#20774).
* lisp/simple.el (do-auto-fill): Only consider break-points that are later in
the line than the width of the fill-prefix. This is a more general solution
than the previous logic, which only skipped over the exact fill-prefix. The
fill-prefix doesn't necessarily match the prefix of the first line of a
paragraph in adaptive-fill-mode.
---
 lisp/simple.el            | 29 ++++++++++++++---------------
 test/lisp/simple-tests.el | 11 +++++++++++
 2 files changed, 25 insertions(+), 15 deletions(-)

diff --git a/lisp/simple.el b/lisp/simple.el
index 072723cd64..15c7f1f935 100644
--- a/lisp/simple.el
+++ b/lisp/simple.el
@@ -7148,18 +7148,20 @@ do-auto-fill
 	       (setq fill-prefix prefix))))
 
       (while (and (not give-up) (> (current-column) fc))
-	;; Determine where to split the line.
-	(let* (after-prefix
-	       (fill-point
-		(save-excursion
-		  (beginning-of-line)
-		  (setq after-prefix (point))
-		  (and fill-prefix
-		       (looking-at (regexp-quote fill-prefix))
-		       (setq after-prefix (match-end 0)))
-		  (move-to-column (1+ fc))
-		  (fill-move-to-break-point after-prefix)
-		  (point))))
+        ;; Determine where to split the line.
+        (let ((fill-point
+               (save-excursion
+                 (beginning-of-line)
+                 (let ((line-start (point)))
+                    (move-to-column (1+ fc))
+                    ;; Don't split earlier in the line than the length of the
+                    ;; fill prefix, since the resulting line would be longer.
+                    (fill-move-to-break-point
+                     (if fill-prefix
+                         (min (line-end-position)
+                              (+ line-start (string-width fill-prefix)))
+                       line-start))
+                    (point)))))
 
 	  ;; See whether the place we found is any good.
 	  (if (save-excursion
@@ -7167,9 +7169,6 @@ do-auto-fill
 		(or (bolp)
 		    ;; There is no use breaking at end of line.
 		    (save-excursion (skip-chars-forward " ") (eolp))
-		    ;; It is futile to split at the end of the prefix
-		    ;; since we would just insert the prefix again.
-		    (and after-prefix (<= (point) after-prefix))
 		    ;; Don't split right after a comment starter
 		    ;; since we would just make another comment starter.
 		    (and comment-start-skip
diff --git a/test/lisp/simple-tests.el b/test/lisp/simple-tests.el
index ad7aee1db1..c7330e4034 100644
--- a/test/lisp/simple-tests.el
+++ b/test/lisp/simple-tests.el
@@ -497,5 +497,16 @@ simple-test-undo-with-switched-buffer
       (should (equal (line-number-at-pos 5) 3))
       (should (equal (line-number-at-pos 7) 4)))))
 
+(ert-deftest auto-fill-mode-no-break-before-length-of-fill-prefix ()
+  (with-temp-buffer
+    (setq-local fill-prefix "   ")
+    (set-fill-column 5)
+    ;; Shouldn't break after 'foo' (3 characters) when the next
+    ;; line is indented >= to that, that woudln't result in shorter
+    ;; lines.
+    (insert "foo bar")
+    (do-auto-fill)
+    (should (string-equal (buffer-string) "foo bar"))))
+
 (provide 'simple-test)
 ;;; simple-test.el ends here
-- 
2.14.1.342.g6490525c54-goog