bug#33887: 26.1; Emacs hangs for several seconds when going to the end of an XML file in nXML mode

[Noam Postavsky <npostavs@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 516 bytes --]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> I pushed a patch which should fix the "lone >" problem without
> introducing any undue extra cost.  It should also fix the "very long
> line" case.

Seems to pass my tests.  Not sure if you missed the alternate fix I
proposed in https://debbugs.gnu.org/33887#94 or not.  It does have the
disadvantage of leaving (car (syntax-ppss)) unreliable for any other
code which uses it.

Here's a patch against master that should cover the remaining cases
Vincent raised:


[-- Attachment #2: patch --]
[-- Type: text/plain, Size: 4011 bytes --]

From 2ffdab0e86161396e3d2606949d1fcf93c58b592 Mon Sep 17 00:00:00 2001
From: Noam Postavsky <npostavs@gmail.com>
Date: Sun, 26 May 2019 11:07:14 -0400
Subject: [PATCH 1/2] Fix some SGML syntax edge cases (Bug#33887)

* lisp/textmodes/sgml-mode.el (sgml-syntax-propertize-rules): Handle
single and double quotes symmetrically.  Don't skip quoted comment
enders.
* test/lisp/textmodes/sgml-mode-tests.el (sgml-tests--quotes-syntax):
Add more test cases.
(sgml-mode-quote-in-long-text): New test.
---
 lisp/textmodes/sgml-mode.el            |  5 +++-
 test/lisp/textmodes/sgml-mode-tests.el | 45 ++++++++++++++++++++++++++++------
 2 files changed, 42 insertions(+), 8 deletions(-)

diff --git a/lisp/textmodes/sgml-mode.el b/lisp/textmodes/sgml-mode.el
index 75f20722b0..1df7e78afc 100644
--- a/lisp/textmodes/sgml-mode.el
+++ b/lisp/textmodes/sgml-mode.el
@@ -363,9 +363,12 @@ (eval-and-compile
      ;; the resulting number of calls to syntax-ppss made it too slow
      ;; (bug#33887), so we're now careful to leave alone any pair
      ;; of quotes that doesn't hold a < or > char, which is the vast majority.
-     ("\\(?:\\(?1:\"\\)[^\"<>]*\\|\\(?1:'\\)[^'\"<>]*\\)"
+     ("\\([\"']\\)[^\"'<>]*"
       (1 (if (eq (char-after) (char-after (match-beginning 0)))
              (forward-char 1)
+           ;; Avoid skipping comment ender.
+           (when (eq (char-after) ?>)
+             (skip-chars-backward "-"))
            ;; Be careful to call `syntax-ppss' on a position before the one
            ;; we're going to change, so as not to need to flush the data we
            ;; just computed.
diff --git a/test/lisp/textmodes/sgml-mode-tests.el b/test/lisp/textmodes/sgml-mode-tests.el
index 1b8965e344..34d26480a4 100644
--- a/test/lisp/textmodes/sgml-mode-tests.el
+++ b/test/lisp/textmodes/sgml-mode-tests.el
@@ -161,15 +161,46 @@ (ert-deftest sgml-quote-works ()
       (should (string= "&&" (buffer-string))))))
 
 (ert-deftest sgml-tests--quotes-syntax ()
+  (dolist (str '("a\"b <t>c'd</t>"
+                 "a'b <t>c\"d</t>"
+                 "<t>\"a'</t>"
+                 "<t>'a\"</t>"
+                 "<t>\"a'\"</t>"
+                 "<t>'a\"'</t>"
+                 "a\"b <tag>c'd</tag>"
+                 "<tag>c>'d</tag>"
+                 "<t><!-- \" --></t>"
+                 "<t><!-- ' --></t>"
+                 ))
+   (with-temp-buffer
+     (sgml-mode)
+     (insert str)
+     (ert-info ((format "%S" str) :prefix "Test case: ")
+       ;; Check that last tag is parsed as a tag.
+       (should (= 1 (car (syntax-ppss (1- (point-max))))))
+       (should (= 0 (car (syntax-ppss (point-max)))))))))
+
+(ert-deftest sgml-mode-quote-in-long-text ()
   (with-temp-buffer
     (sgml-mode)
-    (insert "a\"b <tag>c'd</tag>")
-    (should (= 1 (car (syntax-ppss (1- (point-max))))))
-    (should (= 0 (car (syntax-ppss (point-max)))))
-    (erase-buffer)
-    (insert "<tag>c>d</tag>")
-    (should (= 1 (car (syntax-ppss (1- (point-max))))))
-    (should (= 0 (car (syntax-ppss (point-max)))))))
+    (insert "<t>"
+            ;; `syntax-propertize-wholelines' extends chunk size based
+            ;; on line length, so newlines are significant!
+            (make-string syntax-propertize-chunk-size ?a) "\n"
+            "'"
+            (make-string syntax-propertize-chunk-size ?a) "\n"
+            "</t>")
+    ;; If we just check (syntax-ppss (point-max)) immediately, then
+    ;; we'll end up propertizing the whole buffer in one chunk (so the
+    ;; test is useless).  Simulate something more like what happens
+    ;; when the buffer is viewed normally.
+    (cl-loop for pos from (point-min) to (point-max)
+             by syntax-propertize-chunk-size
+             do (syntax-ppss pos))
+    (syntax-ppss (point-max))
+    ;; Check that last tag is parsed as a tag.
+    (should (= 1 (- (car (syntax-ppss (1- (point-max))))
+                    (car (syntax-ppss (point-max))))))))
 
 (provide 'sgml-mode-tests)
 ;;; sgml-mode-tests.el ends here
-- 
2.11.0


[-- Attachment #3: Type: text/plain, Size: 134 bytes --]


And about the highlighting of quoted text outside tags, we can just
disable fontification, while leaving the syntax code untouched:


[-- Attachment #4: patch --]
[-- Type: text/plain, Size: 4141 bytes --]

From a4a6008d96011e2517939cb8cb51624802a8c31e Mon Sep 17 00:00:00 2001
From: Noam Postavsky <npostavs@gmail.com>
Date: Sun, 26 May 2019 17:41:22 -0400
Subject: [PATCH 2/2] Don't fontiy text outside of SGML/XML tags (Bug#33887)

* lisp/font-lock.el (font-lock-syntactic-face-function-default): New
function.
(font-lock-syntactic-face-function): Use it as default value.
* lisp/textmodes/sgml-mode.el (sgml-font-lock-syntactic-face): New
function.
(sgml-mode):
* lisp/nxml/nxml-mode.el (nxml-mode): Use it as
font-lock-syntactic-face-function value.
---
 lisp/font-lock.el           |  7 +++++--
 lisp/nxml/nxml-mode.el      |  4 +++-
 lisp/textmodes/sgml-mode.el | 11 +++++++++--
 3 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/lisp/font-lock.el b/lisp/font-lock.el
index 3991a4ee8e..ddf1cbdb9f 100644
--- a/lisp/font-lock.el
+++ b/lisp/font-lock.el
@@ -527,9 +527,12 @@ (defvar font-lock-syntactically-fontified 0
 sometimes be slightly incorrect.")
 (make-variable-buffer-local 'font-lock-syntactically-fontified)
 
+(defun font-lock-syntactic-face-function-default (state)
+  "Default value for `font-lock-syntactic-face-function'."
+  (if (nth 3 state) font-lock-string-face font-lock-comment-face))
+
 (defvar font-lock-syntactic-face-function
-  (lambda (state)
-    (if (nth 3 state) font-lock-string-face font-lock-comment-face))
+  #'font-lock-syntactic-face-function-default
   "Function to determine which face to use when fontifying syntactically.
 The function is called with a single parameter (the state as returned by
 `parse-partial-sexp' at the beginning of the region to highlight) and
diff --git a/lisp/nxml/nxml-mode.el b/lisp/nxml/nxml-mode.el
index da01b2a342..05044d66df 100644
--- a/lisp/nxml/nxml-mode.el
+++ b/lisp/nxml/nxml-mode.el
@@ -551,7 +551,9 @@ (define-derived-mode nxml-mode text-mode "nXML"
           nil  ; no special syntax table
           (font-lock-extend-region-functions . (nxml-extend-region))
           (jit-lock-contextually . t)
-          (font-lock-unfontify-region-function . nxml-unfontify-region)))
+          (font-lock-unfontify-region-function . nxml-unfontify-region)
+          (font-lock-syntactic-face-function
+           . sgml-font-lock-syntactic-face)))
 
   (with-demoted-errors (rng-nxml-mode-init)))
 
diff --git a/lisp/textmodes/sgml-mode.el b/lisp/textmodes/sgml-mode.el
index 1df7e78afc..225fe72a01 100644
--- a/lisp/textmodes/sgml-mode.el
+++ b/lisp/textmodes/sgml-mode.el
@@ -329,6 +329,11 @@ (defconst sgml-font-lock-keywords-2
 (defvar sgml-font-lock-keywords sgml-font-lock-keywords-1
   "Rules for highlighting SGML code.  See also `sgml-tag-face-alist'.")
 
+(defun sgml-font-lock-syntactic-face (state)
+  "`font-lock-syntactic-face-function' for `sgml-mode'."
+  (and (nth 9 state) ;; Only use faces within tags.
+       (font-lock-syntactic-face-function-default state)))
+
 (defvar-local sgml--syntax-propertize-ppss nil)
 
 (defun sgml--syntax-propertize-ppss (pos)
@@ -573,7 +578,7 @@ (define-derived-mode sgml-mode text-mode '(sgml-xml-mode "XML" "SGML")
   ;; This is desirable because SGML discards a newline that appears
   ;; immediately after a start tag or immediately before an end tag.
   (setq-local paragraph-start (concat "[ \t]*$\\|\
-[ \t]*</?\\(" sgml-name-re sgml-attrs-re "\\)?>"))
+\[ \t]*</?\\(" sgml-name-re sgml-attrs-re "\\)?>"))
   (setq-local paragraph-separate (concat paragraph-start "$"))
   (setq-local adaptive-fill-regexp "[ \t]*")
   (add-hook 'fill-nobreak-predicate 'sgml-fill-nobreak nil t)
@@ -591,7 +596,9 @@ (define-derived-mode sgml-mode text-mode '(sgml-xml-mode "XML" "SGML")
   (setq font-lock-defaults '((sgml-font-lock-keywords
 			      sgml-font-lock-keywords-1
 			      sgml-font-lock-keywords-2)
-			     nil t))
+                             nil t nil
+                             (font-lock-syntactic-face-function
+                              . sgml-font-lock-syntactic-face)))
   (setq-local syntax-propertize-function #'sgml-syntax-propertize)
   (setq-local facemenu-add-face-function 'sgml-mode-facemenu-add-face-function)
   (setq-local sgml-xml-mode (sgml-xml-guess))
-- 
2.11.0