Re: mbswidth "failure" on Solaris

[Bruno Haible]

Hi,

> >     15 | e: {∇⃗×𝐸⃗ = -∂𝐵⃗/∂t}
> > -      |    ^~~~~~~~~~~~~~
> > +      |    ^~~~~~~~~~~~~~~~~

Indeed, mbswidth seems to have returned 3 more columns.

> The error (three more columns than expected) seems to indicate something
> related to the combining arrow.

No. The issue comes from the math symbols. The following test programs shows
it:

#include <config.h>
#include <stdio.h>
#include <locale.h>
#include <wchar.h>
#include "mbswidth.h"
int main ()
{
  setlocale (LC_ALL, "en_US.UTF-8");
  printf ("%d\n", (int) mbswidth ("{∇⃗×𝐸⃗ = -∂𝐵⃗/∂t}",0)); // 14 vs 17
  printf ("%d\n", wcwidth (0x2207)); // 1 vs. 2
  printf ("%d\n", wcwidth (0x20D7)); // 0
  printf ("%d\n", wcwidth (0x00D7)); // 1
  printf ("%d\n", wcwidth (0x1D438)); // 1
  printf ("%d\n", wcwidth (0x2202)); // 1 vs. 2
  printf ("%d\n", wcwidth (0x1D435)); // 1
}

The following patch should fix it.

The patch changes the behaviour of wcwidth(0x2202) for UTF-8 locales.
It would be possible to limit the change to the non-East-Asian UTF-8
locales (by using the function uc_locale_language() and testing
whether its result is not one of "zh", "ja", "ko"), but glibc does not
do this (it uses the same width across all UTF-8 locales), therefore
I'm not doing it here either.


2019-05-05  Bruno Haible  <address@hidden>

        wcwidth: Ensure width 1, not 2, for ambiguous characters.
        Reported by Kiyoshi KANAZAWA <address@hidden>
        via Akim Demaille <address@hidden>.
        * m4/wcwidth.m4 (gl_FUNC_WCWIDTH): Check the width of U+2202. Use an
        en_US.UTF-8 locale, since that is more likely to be present than an
        fr_FR.UTF-8 locale.
        * tests/test-wcwidth.c (main): Check the width of U+2202.
        * doc/posix-functions/wcwidth.texi: Mention the issue.

diff --git a/m4/wcwidth.m4 b/m4/wcwidth.m4
index 3952fd2..e9b5bf4 100644
--- a/m4/wcwidth.m4
+++ b/m4/wcwidth.m4
@@ -1,4 +1,4 @@
-# wcwidth.m4 serial 28
+# wcwidth.m4 serial 29
 dnl Copyright (C) 2006-2019 Free Software Foundation, Inc.
 dnl This file is free software; the Free Software Foundation
 dnl gives unlimited permission to copy and/or distribute it,
@@ -54,6 +54,8 @@ AC_DEFUN([gl_FUNC_WCWIDTH],
     dnl On OSF/1 5.1, wcwidth(0x200B) (ZERO WIDTH SPACE) returns 1.
     dnl On OpenBSD 5.8, wcwidth(0xFF1A) (FULLWIDTH COLON) returns 0.
     dnl This leads to bugs in 'ls' (coreutils).
+    dnl On Solaris 11.4, wcwidth(0x2202) (PARTIAL DIFFERENTIAL) returns 2,
+    dnl even in Western locales.
     AC_CACHE_CHECK([whether wcwidth works reasonably in UTF-8 locales],
       [gl_cv_func_wcwidth_works],
       [
@@ -80,7 +82,7 @@ int wcwidth (int);
 int main ()
 {
   int result = 0;
-  if (setlocale (LC_ALL, "fr_FR.UTF-8") != NULL)
+  if (setlocale (LC_ALL, "en_US.UTF-8") != NULL)
     {
       if (wcwidth (0x0301) > 0)
         result |= 1;
@@ -90,6 +92,8 @@ int main ()
         result |= 4;
       if (wcwidth (0xFF1A) == 0)
         result |= 8;
+      if (wcwidth (0x2202) > 1)
+        result |= 16;
     }
   return result;
 }]])],
diff --git a/tests/test-wcwidth.c b/tests/test-wcwidth.c
index eb7bdd2..8e9cea3 100644
--- a/tests/test-wcwidth.c
+++ b/tests/test-wcwidth.c
@@ -72,6 +72,22 @@ main ()
       ASSERT (wcwidth (0x200B) == 0);
       ASSERT (wcwidth (0xFEFF) <= 0);
 
+      /* Test width of some math symbols.
+         U+2202 is marked as having ambiguous width (A) in EastAsianWidth.txt
+         (see <https://www.unicode.org/Public/12.0.0/ucd/EastAsianWidth.txt>).
+         The Unicode Standard Annex 11
+         <https://www.unicode.org/reports/tr11/tr11-36.html>
+         says
+           "Ambiguous characters behave like wide or narrow characters
+            depending on the context (language tag, script identification,
+            associated font, source of data, or explicit markup; all can
+            provide the context). If the context cannot be established
+            reliably, they should be treated as narrow characters by default."
+         For wcwidth(), the only available context information is the locale.
+         "fr_FR.UTF-8" is a Western locale, not an East Asian locale, therefore
+         U+2202 should be treated like a narrow character.  */
+      ASSERT (wcwidth (0x2202) == 1);
+
       /* Test width of some CJK characters.  */
       ASSERT (wcwidth (0x3000) == 2);
       ASSERT (wcwidth (0xB250) == 2);
diff --git a/doc/posix-functions/wcwidth.texi b/doc/posix-functions/wcwidth.texi
index 741be8e..ecdf758 100644
--- a/doc/posix-functions/wcwidth.texi
+++ b/doc/posix-functions/wcwidth.texi
@@ -18,6 +18,10 @@ glibc 2.8.
 This function handles combining characters in UTF-8 locales incorrectly on some
 platforms:
 Mac OS X 10.3, OpenBSD 5.8.
address@hidden
+This function returns 2 for characters with ambiguous east asian width, even in
+Western locales, on some platforms:
+Solaris 11.4.
 @end itemize
 
 Portability problems not fixed by Gnulib: