Keep nxml prolog updated via syntax-propertize

Instead of using after-change-functions.  Also, stop consulting
nxml-prolog-regions during syntax-propertize.  It turns out the
problems fixed by using prolog information are actually due to using
the wrong syntax table during propertizing.  This was fixed in
2019-06-04 "* lisp/emacs-lisp/syntax.el: Use syntax-ppss-table for
syntax-propertize." so consulting the prolog data is no longer needed.
* lisp/nxml/nxml-rap.el (nxml-maybe-rescan-prolog): Remove.
* lisp/nxml/nxml-mode.el (nxml-mode): Stop using it.
(nxml-syntax-propertize): Don't use nxml-prolog-regions, just call
nxml-scan-prolog if needed before delegating to
sgml-syntax-propertize.
* test/lisp/nxml/nxml-mode-tests.el (nxml-mode-edit-prolog): New test.

lisp/nxml/nxml-mode.el

@@ -428,25 +428,10 @@ reference.")
 
 (defun nxml-syntax-propertize (start end)
   "Syntactic keywords for `nxml-mode'."
-  ;; Like `sgml-syntax-propertize', but handle `nxml-prolog-regions'.
+  ;; Like `sgml-syntax-propertize', but rescan prolog if needed.
   (when (< start nxml-prolog-end)
-    (catch 'done-prolog
-      (dolist (prolog-elem nxml-prolog-regions)
-        (let ((type (aref prolog-elem 0))
-              (pbeg (aref prolog-elem 1))
-              (pend (aref prolog-elem 2)))
-          (when (eq type 'comment)
-            (put-text-property pbeg (1+ pbeg)
-                               'syntax-table (string-to-syntax "< b"))
-            (put-text-property (1- pend) pend
-                               'syntax-table (string-to-syntax "> b")))
-          (when (> pend end)
-            (throw 'done-prolog t)))))
-    (setq start nxml-prolog-end))
-  (if (>= start end)
-      (goto-char end)
-    (goto-char start)
-    (sgml-syntax-propertize start end)))
+    (nxml-scan-prolog))
+  (sgml-syntax-propertize start end))
 
 (defvar tildify-space-string)
 (defvar tildify-foreach-region-function)
@@ -544,7 +529,6 @@ Many aspects this mode can be customized using
   (setq-local syntax-ppss-table sgml-tag-syntax-table)
   (setq-local syntax-propertize-function #'nxml-syntax-propertize)
   (add-hook 'change-major-mode-hook #'nxml-cleanup nil t)
-  (add-hook 'after-change-functions #'nxml-maybe-rescan-prolog nil t)
 
   ;; Emacs 23 handles the encoding attribute on the xml declaration
   ;; transparently to nxml-mode, so there is no longer a need for the below

lisp/nxml/nxml-rap.el

@@ -108,13 +108,6 @@ Return nil if the character at POS is not inside."
     (setq nxml-prolog-regions (xmltok-forward-prolog))
     (setq nxml-prolog-end (point))))
 
-(defun nxml-maybe-rescan-prolog (start _end _length)
-  "Reparse the prolog if START lies within it.
-`nxml-mode' adds this function on `after-change-functions'."
-  (when (<= start nxml-prolog-end)
-    (save-excursion
-      (nxml-scan-prolog))))
-
 ;;; Random access parsing
 
 (defun nxml-token-after ()

test/lisp/nxml/nxml-mode-tests.el

@@ -114,5 +114,23 @@
     (should (= 1 (- (car (syntax-ppss (1- (point-max))))
                     (car (syntax-ppss (point-max))))))))
 
+(ert-deftest nxml-mode-edit-prolog ()
+  "Test for Bug#23668."
+  (with-temp-buffer
+    (insert "
+ <t>
+ <sub/>
+</t>")
+    (nxml-mode)
+    ;; The leading "\n " before "<t>" is the prolog, indenting will
+    ;; delete the space hence changing the prolog size.  If that is
+    ;; not taken into account, then the <sub/> tag won't be indented
+    ;; correctly.
+    (indent-region (point-min) (point-max))
+    (should (equal (buffer-string) "
+<t>
+  <sub/>
+</t>"))))
+
 (provide 'nxml-mode-tests)
 ;;; nxml-mode-tests.el ends here