bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Dima Kogan]

Hi. I'm seeing this:

1. seq 10 > /tmp/file

2. emacs -Q /tmp/file

3. (highlight-lines-matching-regexp "3")

   This is bound to "M-s h l". I see the line containing "3" highlighted
   in yellow, as expected

4. M-x auto-revert-mode

5. Back in the shell: seq 20 > /tmp/file


auto-revert-mode kicks in, updating the buffer with the results of 'seq
20' (possibly updating to an empty buffer first, if we react immediately
to the file truncation). At this point I would expect either:

1. The buffer being fully reverted, with all the highlighting
   disappearing. This is what happens if we did M-x revert buffer

2. The buffer contents being reverted, but the highlighting being
   reapplied

In this auto-revert scenario, we get something in-between: after the
auto-revert hi-lock-interactive-patterns still contains the highlighting
regex, but no highlighting actually happens. It'd be really nice and
useful if the highlighting stayed.

Thanks!

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Eli Zaretskii]

> From: Dima Kogan <dima@secretsauce.net>
> Date: Thu, 01 Sep 2022 15:40:01 -0700
> 
> Hi. I'm seeing this:
> 
> 1. seq 10 > /tmp/file
> 
> 2. emacs -Q /tmp/file

This "file" is in Fundamental mode, yes?  If not, which major mode is
used, and does that major mode turns on Font Lock?

> 3. (highlight-lines-matching-regexp "3")
> 
>    This is bound to "M-s h l". I see the line containing "3" highlighted
>    in yellow, as expected
> 
> 4. M-x auto-revert-mode
> 
> 5. Back in the shell: seq 20 > /tmp/file
> 
> 
> auto-revert-mode kicks in, updating the buffer with the results of 'seq
> 20' (possibly updating to an empty buffer first, if we react immediately
> to the file truncation). At this point I would expect either:
> 
> 1. The buffer being fully reverted, with all the highlighting
>    disappearing. This is what happens if we did M-x revert buffer
> 
> 2. The buffer contents being reverted, but the highlighting being
>    reapplied
> 
> In this auto-revert scenario, we get something in-between: after the
> auto-revert hi-lock-interactive-patterns still contains the highlighting
> regex, but no highlighting actually happens. It'd be really nice and
> useful if the highlighting stayed.

The documentation of hi-lock-mode says:

  In buffers where Font Lock mode is enabled, patterns are
  highlighted using font lock.  In buffers where Font Lock mode is
  disabled, patterns are applied using overlays; in this case, the
  highlighting will not be updated as you type.  The Font Lock mode
  is considered \"enabled\" in a buffer if its `major-mode'
  causes `font-lock-specified-p' to return non-nil, which means
  the major mode specifies support for Font Lock.

If I use your recipe in a file visited with C Mode, which does use
Font Lock, the auto-revert doesn't lose the highlighted lines.

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Juri Linkov]

>> 3. (highlight-lines-matching-regexp "3")
>>
>>    This is bound to "M-s h l". I see the line containing "3" highlighted
>>    in yellow, as expected
>>
>> 4. M-x auto-revert-mode
>
> If I use your recipe in a file visited with C Mode, which does use
> Font Lock, the auto-revert doesn't lose the highlighted lines.

However, after manual revert with 'C-x x g' highlighting is still lost.
But this is bug#50431 where 'C-x x g' should be fixed to restore minor modes.

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Eli Zaretskii]

> From: Juri Linkov <juri@linkov.net>
> Cc: Dima Kogan <dima@secretsauce.net>,  57534@debbugs.gnu.org
> Date: Fri, 02 Sep 2022 10:04:25 +0300
> 
> >> 3. (highlight-lines-matching-regexp "3")
> >>
> >>    This is bound to "M-s h l". I see the line containing "3" highlighted
> >>    in yellow, as expected
> >>
> >> 4. M-x auto-revert-mode
> >
> > If I use your recipe in a file visited with C Mode, which does use
> > Font Lock, the auto-revert doesn't lose the highlighted lines.
> 
> However, after manual revert with 'C-x x g' highlighting is still lost.

That's not the same revert as in auto-revert-mode.

> But this is bug#50431 where 'C-x x g' should be fixed to restore minor modes.

I'm not sure this is one of the minor modes to be restored.  Maybe
optionally, but not unconditionally.  See the discussion in that bug.

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Drew Adams]

> > But this is bug#50431 where 'C-x x g' should be fixed
> > to restore minor modes.
> 
> I'm not sure this is one of the minor modes to be restored.  Maybe
> optionally, but not unconditionally.  See the discussion in that bug.

+2.

Reverting is not about restoring minor modes.
Certainly not blindly so.  Not at all.

Reverting is specific to the major mode.

The major mode gets to decide what reverting
means/does.  The major mode can decide to
restore whatever it likes, including this or
that minor mode.  Nothing should "restore the
minor modes" by default.

My crystal ball whispers that calls for
reverting to "restore minor modes" are trying
to use a sledge hammer where at most a pair
of tweezers might be needed in some uncommon
scenario.

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Juri Linkov]

[-- Attachment #1: Type: text/plain, Size: 1275 bytes --]

> 1. seq 10 > /tmp/file
>
> 2. emacs -Q /tmp/file
>
> 3. (highlight-lines-matching-regexp "3")
>
>    This is bound to "M-s h l". I see the line containing "3" highlighted
>    in yellow, as expected
>
> 4. M-x auto-revert-mode
>
> 5. Back in the shell: seq 20 > /tmp/file
>
> auto-revert-mode kicks in, updating the buffer with the results of 'seq
> 20' (possibly updating to an empty buffer first, if we react immediately
> to the file truncation). At this point I would expect either:
>
> 1. The buffer being fully reverted, with all the highlighting
>    disappearing. This is what happens if we did M-x revert buffer
>
> 2. The buffer contents being reverted, but the highlighting being
>    reapplied
>
> In this auto-revert scenario, we get something in-between: after the
> auto-revert hi-lock-interactive-patterns still contains the highlighting
> regex, but no highlighting actually happens. It'd be really nice and
> useful if the highlighting stayed.

Now re-highlighting after revert is implemented in Emacs 30.

However, I noticed one problem that makes this feature unusable:
when a file contains file patterns, then after every revert it asks
again and again.  This is because hi-lock-file-patterns-policy
missed the value `always', now added in this patch:


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: hi-lock-file-patterns-policy.patch --]
[-- Type: text/x-diff, Size: 2918 bytes --]

diff --git a/lisp/hi-lock.el b/lisp/hi-lock.el
index 64b84cdf859..de4384054ef 100644
--- a/lisp/hi-lock.el
+++ b/lisp/hi-lock.el
@@ -122,9 +122,10 @@ hi-lock-file-patterns-policy
   "Specify when hi-lock should use patterns found in file.
 If `ask', prompt when patterns found in buffer; if bound to a function,
 use patterns when function returns t (function is called with patterns
-as first argument); if nil or `never' or anything else, don't use file
-patterns."
+as first argument); if `always', use file patterns without prompt;
+if nil or `never' or anything else, don't use file patterns."
   :type '(choice (const :tag "Do not use file patterns" never)
+                 (const :tag "Always use file patterns" always)
                  (const :tag "Ask about file patterns" ask)
                  (function :tag "Function to check file patterns"))
   :group 'hi-lock
@@ -334,8 +335,8 @@ hi-lock-mode
   (See `font-lock-keywords'.)  They may be edited and re-loaded with \\[hi-lock-find-patterns],
   any valid `font-lock-keywords' form is acceptable.  When a file is
   loaded the patterns are read if `hi-lock-file-patterns-policy' is
-  `ask' and the user responds y to the prompt, or if
-  `hi-lock-file-patterns-policy' is bound to a function and that
+  `always' or if it's `ask' and the user responds y to the prompt,
+  or if `hi-lock-file-patterns-policy' is bound to a function and that
   function returns t.
 
 \\[hi-lock-find-patterns]
@@ -852,6 +853,7 @@ hi-lock-find-patterns
                    (funcall hi-lock-file-patterns-policy all-patterns))
                   ((eq hi-lock-file-patterns-policy 'ask)
                    (y-or-n-p "Add patterns from this buffer to hi-lock? "))
+                  ((eq hi-lock-file-patterns-policy 'always) t)
                   (t nil)))
         (hi-lock-set-file-patterns all-patterns)
         (if (called-interactively-p 'interactive)
@@ -866,12 +868,16 @@ hi-lock-font-lock-hook
 (defun hi-lock-revert-buffer-rehighlight ()
   "Rehighlight hi-lock patterns after `revert-buffer'.
 Apply the previous patterns after reverting the buffer."
-  (when-let ((patterns hi-lock-interactive-lighters))
-    (lambda ()
-      (when hi-lock-interactive-lighters
-        (hi-lock-unface-buffer t))
-      (dolist (pattern (reverse patterns))
-        (highlight-regexp (car pattern) (cadr (nth 1 (caddr pattern))))))))
+  (let ((policy (if hi-lock-file-patterns 'always 'never))
+        (patterns hi-lock-interactive-lighters))
+    (when patterns
+      (lambda ()
+        (unless hi-lock-mode
+          (let ((hi-lock-file-patterns-policy policy))
+            (when hi-lock-interactive-lighters
+              (hi-lock-unface-buffer t))
+            (dolist (pattern (reverse patterns))
+              (highlight-regexp (car pattern) (cadr (nth 1 (caddr pattern)))))))))))
 
 (defvar hi-lock--hashcons-hash
   (make-hash-table :test 'equal :weakness t)

bug#57534: 29.0.50; Highlighting lost after auto-revert-mode triggers

[Juri Linkov]

close 57534 30.0.60
thanks

>> It'd be really nice and useful if the highlighting stayed.
>
> Now re-highlighting after revert is implemented in Emacs 30.
>
> However, I noticed one problem that makes this feature unusable:
> when a file contains file patterns, then after every revert it asks
> again and again.  This is because hi-lock-file-patterns-policy
> missed the value `always', now added in this patch:

This is fixed now as well.  So this request can be closed now.
Thanks for the request.