Ar an seachtú lá déag de mí Lúnasa, scríobh Aidan Kehoe: I still think this, but now I investigate some more I learn that GNU now have a #’make-temp-file, which does what a Unix mkstemp call does. For the sake of compatibility we should merge that, and for the sake of paranoia we should use it, here and elsewhere--it’s a better-designed API than #’make-temp-name or mktemp(3). Heh, not quite “later today” but the patch is a bit more comprehensive. No, it was due to Mike Sperber changing the load path for packages and not telling anyone. I’ve decided to use Skip-Test-Unless, with the not-particularly-robust assumption that if input-method-alist is greater than one in length (the first entry coming from canna, the #’register-input-method call for which is dumped), then all the package input methods are available. This reveals that the following input methods used in language environments are not available in packages: latin-8-prefix latin-10-prefix latin-5-prefix latin-4-prefix latin-4-prefix latin-4-prefix latin-4-prefix latin-4-prefix belarusian bulgarian-bds Using Known-Bug-Expect-Failure calls to say that we know this already is awkward, and the actual problem itself can be fixed with less work than that would involve. Hopefully I’ll get to that soon. I have added a Known-Bug-Expect-Failure for a separate bug, and documented what provokes it and what needs to be done to eliminate it. APPROVE COMMIT NOTE: This patch has been committed. tests/ChangeLog addition: 2007-08-21 Aidan Kehoe <kehoea(a)parhasard.net&gt; * automated/mule-tests.el (test-chars): There are now 2^21 possible characters under Mule, not 2^19. * automated/mule-tests.el (test-file-name): Call #'make-temp-name instead of writing the file needed for the rever-buffer test to the user's home directory. Wrap the following #'save-buffer in a Silence-Message call. * automated/mule-tests.el: In the language environment tests, check that a given input method exists before trying to activate it. If there are installed input methods, assert that every input method specified in any language environment exists. If the native-coding-system of the language environment is specified as a function, don't assert that it is a coding system. In the byte-compiler coding cookie tests, call #'read at runtime on strings with non-Latin-1 Unicode escapes. This avoids errors when make check is run under a non-Mule XEmacs, where unavailable Unicode escapes throw an error in the Lisp reader. Add a new Known-Bug-Expect-Failure call for a bug that occurred to me yesterday; if non-Latin-1 is generated in an eval-when-compile clause, that non-ASCII may be trashed and the compiled code incorrectly saved as no-conversion if no other non-Latin-1 exists in the input buffer. Add a comment giving more details of the bug and how one might fix it. lisp/ChangeLog addition: 2007-08-21 Aidan Kehoe <kehoea(a)parhasard.net&gt; * mule/vietnamese.el ("Vietnamese"): Remove vscii and viqr from the coding-system slot of the language environment for Vietnamese; they're no longer available as coding systems in XEmacs, and they were never widely used. XEmacs Trunk source patch: Diff command: cvs -q diff -u Files affected: lisp/mule/vietnamese.el =================================================================== RCS tests/automated/mule-tests.el =================================================================== RCS Index: tests/automated/mule-tests.el =================================================================== RCS file: /pack/xemacscvs/XEmacs/xemacs/tests/automated/mule-tests.el,v retrieving revision 1.19 diff -u -r1.19 mule-tests.el --- tests/automated/mule-tests.el 2007/06/22 16:37:41 1.19 +++ tests/automated/mule-tests.el 2007/08/21 12:34:15 ＠＠ -42,12 +42,12 ＠＠ (defun test-chars (&optional for-test-harness) "Insert all characters in a buffer, to see if XEmacs will crash. This is done by creating a string with all the legal characters -in [0, 2^19) range, inserting it into the buffer, and checking +in [0, 2^21) range, inserting it into the buffer, and checking that the buffer's contents are equivalent to the string. If FOR-TEST-HARNESS is specified, a temporary buffer is used, and the Assert macro checks for correctness." - (let ((max (expt 2 (if (featurep 'mule) 19 8))) + (let ((max (expt 2 (if (featurep 'mule) 21 8))) (list nil) (i 0)) (while (< i max) ＠＠ -118,14 +118,27 ＠＠ ;; Fixed 2007-06-22 <18043.2793.611745.734215(a)parhasard.net&gt;. ;;---------------------------------------------------------------- -;; #### need a temp file name but this will do for now -(let ((test-file-name (expand-file-name "~/test-revert-buffer-resets-modiff")) +(let ((test-file-name + ;; The Gnus people, when they call #'make-temp-name, then loop, + ;; checking if the corresponding file exists. Our #'make-temp-name + ;; already does this loop, and the Gnus approach doesn't bring + ;; anything; there remains a race condition if you can predict the + ;; path name. The path name in question depends on the process ID and + ;; a (weak) PRNG seeded with the seconds to the power of the + ;; milliseconds of some instant close to the startup time of this + ;; XEmacs; without being able to read the address space of this + ;; XEmacs, or monitor what stat() calls it does, it is not predictable. + ;; + ;; The really kosher way to do this is to merge GNU's make-temp-file + ;; and use that. It basically has the functionality of the Unix + ;; mkstemp. + (make-temp-name (expand-file-name "tXfXsKc" (temp-directory)))) revert-buffer-function kill-buffer-hook) ; paranoia (find-file test-file-name) (erase-buffer) (insert "a string\n") - (save-buffer 0) + (Silence-Message (save-buffer 0)) (insert "more text\n") (revert-buffer t t) ;; Just "find-file" with autodetect coding didn't fail for me, but it does ＠＠ -491,7 +504,7 ＠＠ ;; The sort is to make the algorithm of charsets-in-region ;; irrelevant. (sort (charsets-in-region (point-min) (point-max)) - 'string<) + #'string<) '(arabic-1-column arabic-2-column ascii chinese-big5-1 chinese-gb2312 cyrillic-iso8859-5 ethiopic greek-iso8859-7 hebrew-iso8859-8 japanese-jisx0208 japanese-jisx0212 ＠＠ -500,26 +513,54 ＠＠ (Assert (equal (sort (charsets-in-string (buffer-substring (point-min) (point-max))) - 'string<) + #'string<) '(arabic-1-column arabic-2-column ascii chinese-big5-1 chinese-gb2312 cyrillic-iso8859-5 ethiopic greek-iso8859-7 hebrew-iso8859-8 japanese-jisx0208 japanese-jisx0212 katakana-jisx0201 korean-ksc5601 latin-iso8859-1 latin-iso8859-2 thai-xtis vietnamese-viscii-lower)))) - ;; Language environments. - (dolist (language (mapcar 'car language-info-alist)) - (set-language-environment language) + ;;--------------------------------------------------------------- + ;; Language environments, and whether the specified values are sane. + ;;--------------------------------------------------------------- + (loop + for language in (mapcar #'car language-info-alist) + with language-input-method = nil + do + ;; s-l-e can call #'require, which says "Loading ..." + (Silence-Message (set-language-environment language)) (Assert (equal language current-language-environment)) - (set-input-method (get-language-info language 'input-method)) - (Assert (equal (get-language-info language 'input-method) - current-input-method)) + + (setq language-input-method + (get-language-info language 'input-method)) + (when (and language-input-method + ;; #### Not robust, if more input methods besides canna are + ;; in core. The intention of this is that if *any* of the + ;; packages' input methods are available, we check that *all* + ;; of the language environments' input methods actually + ;; exist, which goes against the spirit of non-monolithic + ;; packages. But I don't have a better approach to this. + (> (length input-method-alist) 1)) + (Assert (assoc language-input-method input-method-alist)) + (Skip-Test-Unless + (assoc language-input-method input-method-alist) + "input method unavailable" + (format "check that IM %s can be activated" language-input-method) + ;; s-i-m can load files. + (Silence-Message + (set-input-method language-input-method)) + (Assert (equal language-input-method current-input-method)))) + (dolist (charset (get-language-info language 'charset)) (Assert (charsetp (find-charset charset)))) (dolist (coding-system (get-language-info language 'coding-system)) (Assert (coding-system-p (find-coding-system coding-system)))) - (dolist (coding-system (get-language-info language 'coding-system)) - (Assert (coding-system-p (find-coding-system coding-system))))) + (dolist (coding-system (get-language-info language + 'native-coding-system)) + ;; We don't have the appropriate POSIX locales to test with a + ;; native-coding-system that is a function. + (unless (functionp coding-system) + (Assert (coding-system-p (find-coding-system coding-system)))))) (with-temp-buffer (flet ＠＠ -535,9 +576,9 ＠＠ temporary-file-name (current-buffer) byte-compile-result) - (Assert (string-match "^;;;###coding system: escape-quoted" - (buffer-substring nil nil - byte-compile-result)))))) + (Assert (string-match + "^;;;###coding system: escape-quoted" + (buffer-substring nil nil byte-compile-result)))))) (Assert-elc-has-no-specified-encoding () "Assert the current buffer has no coding cookie if compiled." (save-excursion ＠＠ -552,17 +593,21 ＠＠ byte-compile-result) (Assert (not (string-match ";;;###coding system:" - (buffer-substring nil nil byte-compile-result)))))))) + (buffer-substring nil nil + byte-compile-result)))))))) (insert - ;; Create a buffer creating the Unicode escapes. - #r" (defvar testing-mule-compilation-handling - (string ?\u371E ;; kDefinition beautiful; pretty, used + ;; Create a buffer with Unicode escapes. The #'read call is at + ;; runtime, because this file may be compiled and read in a non-Mule + ;; XEmacs. (But it won't be run.) + (read + "#r\" (defvar testing-mule-compilation-handling + (string ?\\u371E ;; kDefinition beautiful; pretty, used ;; in girl's name - ?\U0002A6A9 ;; kDefinition (Cant.) sound of shouting - ?\U0002A65B ;; kDefinition (Cant.) decayed teeth; + ?\\U0002A6A9 ;; kDefinition (Cant.) sound of shouting + ?\\U0002A65B ;; kDefinition (Cant.) decayed teeth; ;; tongue-tied - ?\U00010400 ;; DESERET CAPITAL LETTER LONG I - ?\u3263)) ;; CIRCLED HANGUL RIEUL ") + ?\\U00010400 ;; DESERET CAPITAL LETTER LONG I + ?\\u3263)) ;; CIRCLED HANGUL RIEUL \"")) (Assert-elc-is-escape-quoted) (delete-region (point-min) (point-max)) ＠＠ -570,55 +615,83 ＠＠ (insert ;; This time, the buffer will contain the actual characters, because of ;; u flag to the #r. - #ru" (defvar testing-mule-compilation-handling - (string ?\u371E ;; kDefinition beautiful; pretty, used + (read + "#ru\" (defvar testing-mule-compilation-handling + (string ?\\u371E ;; kDefinition beautiful; pretty, used ;; in girl's name - ?\U0002A6A9 ;; kDefinition (Cant.) sound of shouting - ?\U0002A65B ;; kDefinition (Cant.) decayed teeth; + ?\\U0002A6A9 ;; kDefinition (Cant.) sound of shouting + ?\\U0002A65B ;; kDefinition (Cant.) decayed teeth; ;; tongue-tied - ?\U00010400 ;; DESERET CAPITAL LETTER LONG I - ?\u3263)) ;; CIRCLED HANGUL RIEUL ") + ?\\U00010400 ;; DESERET CAPITAL LETTER LONG I + ?\\u3263)) ;; CIRCLED HANGUL RIEUL \"")) (Assert-elc-is-escape-quoted) (delete-region (point-min) (point-max)) (insert ;; Just a single four character escape. - #r" (defvar testing-mule-compilation-handling - (string ?\u371E)) ;; kDefinition beautiful; pretty, used") + (read + "#r\" (defvar testing-mule-compilation-handling + (string ?\\u371E)) ;; kDefinition beautiful; pretty, used\"")) (Assert-elc-is-escape-quoted) (delete-region (point-min) (point-max)) (insert ;; Just a single eight character escape. - #r" (defvar testing-mule-compilation-handling - (string ?\U0002A65B)) ;; kDefinition (Cant.) decayed teeth;") + (read + "#r\" (defvar testing-mule-compilation-handling + (string ?\\U0002A65B)) ;; kDefinition (Cant.) decayed teeth;\"")) (Assert-elc-is-escape-quoted) (delete-region (point-min) (point-max)) (insert - ;; A single latin-1 hex digit escape + ;; A single latin-1 hex digit escape No run-time #'read call, + ;; non-Mule can handle this too. #r" (defvar testing-mule-compilation-handling - (string ?\xab)) ;; LEFT-POINTING DOUBLE ANGLE QUOTATION MARK") + (string ?\xab)) ;; LEFT-POINTING DOUBLE ANGLE QUOTATION MARK") (Assert-elc-has-no-specified-encoding) (delete-region (point-min) (point-max)) (insert - ;; A single latin-1 character + ;; A single latin-1 character. No run-time #'read call. #ru" (defvar testing-mule-compilation-handling - (string ?\u00AB)) ;; LEFT-POINTING DOUBLE ANGLE QUOTATION MARK") + (string ?\u00AB)) ;; LEFT-POINTING DOUBLE ANGLE QUOTATION MARK\")") (Assert-elc-has-no-specified-encoding) (delete-region (point-min) (point-max)) (insert - ;; Just ASCII. + ;; Just ASCII. No run-time #'read call #r" (defvar testing-mule-compilation-handling (string ?A)) ;; LATIN CAPITAL LETTER A") (Assert-elc-has-no-specified-encoding) + (delete-region (point-min) (point-max)) + + ;; This bug exists because the coding-cookie insertion code looks at + ;; the input buffer, not the output buffer. + ;; + ;; It looks at the input buffer because byte-compile-dynamic and + ;; byte-compile-dynamic-docstrings currently need to be + ;; unconditionally turned off for Mule files, since dynamic + ;; compilation of function bodies and docstrings fails if you can't + ;; call (point) and trivially get the byte offset in the file. + ;; + ;; And to unconditionally turn those two features off, you need to + ;; know before byte-compilation whether the byte-compilation output + ;; file contains non-Latin-1 characters, or perhaps to check after + ;; compilation and redo; but we don't do the latter. + ;; + ;; To fix this bug, we need to add Mule support to + ;; byte-compile-dynamic and byte-compile-dynamic-docstrings. Or drop + ;; support for those features entirely. + (insert + "(defvar testing-mule-compilation-handling (eval-when-compile + (decode-char 'ucs #x371e))) ;; kDefinition beautiful; pretty, used\"") + (Known-Bug-Expect-Failure + (Assert-elc-is-escape-quoted)) (delete-region (point-min) (point-max)))) ) Index: lisp/mule/vietnamese.el =================================================================== RCS file: /pack/xemacscvs/XEmacs/xemacs/lisp/mule/vietnamese.el,v retrieving revision 1.9 diff -u -r1.9 vietnamese.el --- lisp/mule/vietnamese.el 2007/07/22 22:03:50 1.9 +++ lisp/mule/vietnamese.el 2007/08/21 12:34:15 ＠＠ -205,7 +205,7 ＠＠ (set-language-info-alist "Vietnamese" '((charset vietnamese-viscii-lower vietnamese-viscii-upper) - (coding-system viscii vscii viqr) + (coding-system viscii) (coding-priority viscii) (locale "vietnamese" "vi") ;; Not available in packages. -- On the quay of the little Black Sea port, where the rescued pair came once more into contact with civilization, Dobrinton was bitten by a dog which was assumed to be mad, though it may only have been indiscriminating. (Saki) _______________________________________________ XEmacs-Patches mailing list XEmacs-Patches(a)xemacs.org http://calypso.tux.org/cgi-bin/mailman/listinfo/xemacs-patches