Re: shell mode syntax higlighting problem with here documents

[Stefan Monnier]

> #!/bin/bash
> cat << EOF | \
>     tee -a bug.txt

> This is a test case for a bug in bash shell mode text highlighting
> EOF

If people who hack on SH scripts with lots of heredoc thingies could try the
patch below, it would help.  It's relative to the Emacs-CVS code.
It seems "safe" and appears to do the right thing with the above example,
but I'd like to get some positive feedback before installing it,


        Stefan


--- orig/lisp/progmodes/sh-script.el
+++ mod/lisp/progmodes/sh-script.el
@@ -869,7 +869,18 @@
 (defconst sh-st-symbol (string-to-syntax "_"))
 (defconst sh-here-doc-syntax (string-to-syntax "|")) ;; generic string
 
-(defconst sh-here-doc-open-re 
"<<-?\\s-*\\\\?\\(\\(?:['\"][^'\"]+['\"]\\|\\sw\\)+\\).*\\(\n\\)")
+(defconst sh-escaped-line-re
+  (concat
+   "\\(?:"
+   ;; First the real regexp.
+   "\\(?:.*[^\\\n]\\)?\\(?:\\(?:\\\\\\\\\\)*\\\\\n\\(?:.*[^\\\n]\\)?\\)*"
+   ;; Then a failsafe fallback: in case the search boundary prevents us from
+   ;; finding the actual end of line.
+   "\\|.*\\)"))
+
+(defconst sh-here-doc-open-re
+  (concat "<<-?\\s-*\\\\?\\(\\(?:['\"][^'\"]+['\"]\\|\\sw\\)+\\)"
+          sh-escaped-line-re "\\(\n\\)"))
 
 (defvar sh-here-doc-markers nil)
 (make-variable-buffer-local 'sh-here-doc-markers)
@@ -883,7 +894,9 @@
          ;; A rough regexp that should find the opening <<EOF back.
         (sre (concat "<<\\(-?\\)\\s-*['\"\\]?"
                      ;; Use \s| to cheaply check it's an open-heredoc.
-                     eof-re "['\"]?\\([ \t|;&)<>].*\\)?\\s|"))
+                     eof-re "['\"]?\\([ \t|;&)<>]"
+                      sh-escaped-line-re
+                      "\\)?\\s|"))
         ;; A regexp that will find other EOFs.
         (ere (concat "^" (if indented "[ \t]*") eof-re "\n"))
         (start (save-excursion
@@ -922,7 +935,8 @@
 START is the position of <<.
 STRING is the actual word used as delimiter (f.ex. \"EOF\").
 INDENTED is non-nil if the here document's content (and the EOF mark) can
-be indented (i.e. a <<- was used rather than just <<)."
+be indented (i.e. a <<- was used rather than just <<).
+Point is at the beginning of the next line."
   (unless (or (memq (char-before start) '(?< ?>))
              (sh-in-comment-or-string start))
     ;; We're looking at <<STRING, so we add "^STRING$" to the syntactic
@@ -933,6 +947,20 @@
        (setq sh-here-doc-re
              (concat sh-here-doc-open-re "\\|^\\([ \t]*\\)"
                      (regexp-opt sh-here-doc-markers t) "\\(\n\\)"))))
+    (let ((ppss (save-excursion (syntax-ppss (1- (point))))))
+      (if (nth 4 ppss)
+          ;; The \n not only starts the heredoc but also closes a comment.
+          ;; Let's close the comment just before the \n.
+          (put-text-property (1- (point)) (point) 'syntax-table '(12))) ;">"
+      (if (or (> (count-lines start (point)) 1) (nth 5 ppss))
+          ;; If the sh-escaped-line-re part of sh-here-doc-re has matched
+          ;; several lines, make sure we refontify them together.
+          ;; Furthermore, if (nth 5 ppss) is non-nil (i.e. the \n is
+          ;; escaped), it means we haven't actually found the right \n.
+          ;; Don't bother fixing it now, but place a multiline property so
+          ;; that when jit-lock-context-* refontifies the rest of the
+          ;; buffer, it also refontifies the current line with it.
+          (put-text-property start (point) 'font-lock-multiline t)))
     sh-here-doc-syntax))
 
 (defun sh-font-lock-here-doc (limit)