Re: Fix strtol in Turkish locales (bug 19242)

[Joseph Myers <joseph at codesourcery dot com>]

On Thu, 19 Nov 2015, Florian Weimer wrote:

> On 11/13/2015 11:06 PM, Joseph Myers wrote:
> 
> > This patch makes the code check for the relevant ranges of letters
> > rather than using toupper / isalpha (via macros) at all; locales
> > remain used for whitespace testing (explicitly correct according to
> > ISO C).  Note that the way the code worked, the only non-ASCII letter
> > that would previously have been accepted would have been the Turkish
> > 'Ä' (dotless 'I'), because the uppercase version of that in Turkish
> > locales is 'I'.  This patch means that will no longer be accepted,
> > which seems appropriate.
> 
> Have you tested the performance impact of this change?  The locale-based

No.

> approach uses a lookup table, and it's not clear if the fairly involved
> condition in the macro makes a difference.  (This code is probably quite
> performance-sensitive for some applications.)

An alternative version below just uses _nl_C_locobj_ptr in the existing 
calls to __iswalpha_l / __towupper_l / __isalpha_l / __toupper_l, so 
should be more obviously safe in performance terms, albeit less obvious 
about what the code is actually doing.

> Should sscanf have a test as well?

I think the tests in this patch cover all the copies of the code that 
actually get built into glibc (four copies for narrow strings and four for 
wide).  There are a lot of other interfaces that use this code more 
indirectly, including scanf functions, but I don't see a need to cover all 
of them in each test for a strtol issue.  (Although there ought to be at 
least one test somewhere of strtol_l etc. - none of those *_l functions 
appear to be tested - but that's part of the "lots of functions are 
untested" issue for which I posted scripts some time ago to identify 
untested functions, not an issue with any particular patch affecting those 
functions.)


Fix strtol in Turkish locales (bug 19242).

The implementations of strtol and related functions use
locale-specific conversions to upper case before determining whether a
character is a valid letter in the argument.  This means that in
Turkish locales such as tr_TR.UTF-8 and tr_TR.ISO-8859-9, "i" is
interpreted as not being a valid number, when if the base passed to
strtol is 19 or more it should be interpreted as the number 18.

ISO C explicitly says "The letters from a (or A) through z (or Z) are
ascribed the values 10 through 35", so clearly intends the standard
ASCII letters (otherwise you wouldn't generally have exactly 26
letters to ascribe such values) (whereas white-space must be
identified according to the locale).  In particular, 'i' and 'I' must
be understood to be in that sequence.

This patch makes the code do the case conversions and classification
in the C locale; the user's locale remains used for whitespace testing
(explicitly correct according to ISO C).  Note that the way the code
worked, the only non-ASCII letter that would previously have been
accepted would have been the Turkish 'Ä' (dotless 'i'), because the
uppercase version of that in Turkish locales is 'I'.  This patch means
that will no longer be accepted, which seems appropriate.

Tested for x86_64 and x86.

2015-11-19  Joseph Myers  <joseph@codesourcery.com>

	[BZ #19242]
	* stdlib/strtol_l.c (ISALPHA): Use _nl_C_locobj_ptr for locale.
	(TOUPPER): Likewise.
	* stdlib/tst-strtol-locale-main.c: New file.
	* stdlib/tst-strtol-locale.c: Likewise.
	* stdlib/Makefile (tests): Add tst-strtol-locale.
	[$(run-built-tests) = yes] (LOCALES): Add tr_TR.ISO-8859-9.
	[$(run-built-tests) = yes] ($(objpfx)tst-strtol-locale.out):
	Depend on $(gen-locales).
	* wcsmbs/tst-wcstol-locale.c: New file.
	* wcsmbs/Makefile (tests): Add tst-wcstol-locale.
	[$(run-built-tests) = yes] (LOCALES): Add tr_TR.UTF-8 and
	tr_TR.ISO-8859-9.
	[$(run-built-tests) = yes] ($(objpfx)tst-wcstol-locale.out):
	Depend on $(gen-locales).

diff --git a/stdlib/Makefile b/stdlib/Makefile
index 9e0c249..e8b5b8c 100644
--- a/stdlib/Makefile
+++ b/stdlib/Makefile
@@ -74,7 +74,8 @@ tests		:= tst-strtol tst-strtod testmb testrand testsort testdiv   \
 		   tst-makecontext3 bug-getcontext bug-fmtmsg1		    \
 		   tst-secure-getenv tst-strtod-overflow tst-strtod-round   \
 		   tst-tininess tst-strtod-underflow tst-tls-atexit	    \
-		   tst-setcontext3 tst-tls-atexit-nodelete
+		   tst-setcontext3 tst-tls-atexit-nodelete		    \
+		   tst-strtol-locale
 tests-static	:= tst-secure-getenv
 
 modules-names	= tst-tls-atexit-lib
@@ -122,7 +123,8 @@ endif
 include ../Rules
 
 ifeq ($(run-built-tests),yes)
-LOCALES := cs_CZ.UTF-8 de_DE.UTF-8 en_US.ISO-8859-1 tr_TR.UTF-8
+LOCALES := cs_CZ.UTF-8 de_DE.UTF-8 en_US.ISO-8859-1 tr_TR.UTF-8 \
+	   tr_TR.ISO-8859-9
 include ../gen-locales.mk
 
 $(objpfx)bug-strtod2.out: $(gen-locales)
@@ -131,6 +133,7 @@ $(objpfx)tst-strtod.out: $(gen-locales)
 $(objpfx)tst-strtod3.out: $(gen-locales)
 $(objpfx)tst-strtod4.out: $(gen-locales)
 $(objpfx)tst-strtod5.out: $(gen-locales)
+$(objpfx)tst-strtol-locale.out: $(gen-locales)
 endif
 
 # Testdir has to be named stdlib and needs to be writable
diff --git a/stdlib/strtol_l.c b/stdlib/strtol_l.c
index 8f6163d..392b31a 100644
--- a/stdlib/strtol_l.c
+++ b/stdlib/strtol_l.c
@@ -137,8 +137,8 @@
 # define UCHAR_TYPE wint_t
 # define STRING_TYPE wchar_t
 # define ISSPACE(Ch) __iswspace_l ((Ch), loc)
-# define ISALPHA(Ch) __iswalpha_l ((Ch), loc)
-# define TOUPPER(Ch) __towupper_l ((Ch), loc)
+# define ISALPHA(Ch) __iswalpha_l ((Ch), _nl_C_locobj_ptr)
+# define TOUPPER(Ch) __towupper_l ((Ch), _nl_C_locobj_ptr)
 #else
 # if defined _LIBC \
    || defined STDC_HEADERS || (!defined isascii && !defined HAVE_ISASCII)
@@ -150,8 +150,8 @@
 # define UCHAR_TYPE unsigned char
 # define STRING_TYPE char
 # define ISSPACE(Ch) __isspace_l ((Ch), loc)
-# define ISALPHA(Ch) __isalpha_l ((Ch), loc)
-# define TOUPPER(Ch) __toupper_l ((Ch), loc)
+# define ISALPHA(Ch) __isalpha_l ((Ch), _nl_C_locobj_ptr)
+# define TOUPPER(Ch) __toupper_l ((Ch), _nl_C_locobj_ptr)
 #endif
 
 #define INTERNAL(X) INTERNAL1(X)
diff --git a/stdlib/tst-strtol-locale-main.c b/stdlib/tst-strtol-locale-main.c
new file mode 100644
index 0000000..8293e07
--- /dev/null
+++ b/stdlib/tst-strtol-locale-main.c
@@ -0,0 +1,87 @@
+/* Test strtol functions work with all ASCII letters in Turkish
+   locales (bug 19242).
+   Copyright (C) 2015 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <locale.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <wchar.h>
+
+#define STR_(X) #X
+#define STR(X) STR_(X)
+#define FNPFXS STR (FNPFX)
+#define CONCAT_(X, Y) X ## Y
+#define CONCAT(X, Y) CONCAT_ (X, Y)
+#define FNX(FN) CONCAT (FNPFX, FN)
+
+#define TEST(LOC, STR, EXP_VAL, FN, TYPE, FMT)				\
+  do									\
+    {									\
+      CHAR *ep;								\
+      TYPE val = FNX (FN) (STR, &ep, 36);				\
+      printf ("%s: " FNPFXS #FN " (" SFMT ") == " FMT "\n", LOC, STR, val); \
+      if (val == (TYPE) (EXP_VAL) && *ep == 0)				\
+	printf ("PASS: %s: " FNPFXS #FN " (" SFMT ")\n", LOC, STR);	\
+      else								\
+	{								\
+	  printf ("FAIL: %s: " FNPFXS #FN " (" SFMT ")\n", LOC, STR);	\
+	  result = 1;							\
+	}								\
+    }									\
+  while (0)
+
+static int
+test_one_locale (const char *loc)
+{
+  if (setlocale (LC_ALL, loc) == NULL)
+    {
+      printf ("setlocale (LC_ALL, \"%s\") failed\n", loc);
+      return 1;
+    }
+  int result = 0;
+  for (int i = 10; i < 36; i++)
+    {
+      CHAR s[2];
+      s[0] = L_('A') + i - 10;
+      s[1] = 0;
+      TEST (loc, s, i, l, long int, "%ld");
+      TEST (loc, s, i, ul, unsigned long int, "%lu");
+      TEST (loc, s, i, ll, long long int, "%lld");
+      TEST (loc, s, i, ull, unsigned long long int, "%llu");
+      s[0] = L_('a') + i - 10;
+      s[1] = 0;
+      TEST (loc, s, i, l, long int, "%ld");
+      TEST (loc, s, i, ul, unsigned long int, "%lu");
+      TEST (loc, s, i, ll, long long int, "%lld");
+      TEST (loc, s, i, ull, unsigned long long int, "%llu");
+    }
+  return result;
+}
+
+static int
+do_test (void)
+{
+  int result = 0;
+  result |= test_one_locale ("C");
+  result |= test_one_locale ("tr_TR.UTF-8");
+  result |= test_one_locale ("tr_TR.ISO-8859-9");
+  return result;
+}
+
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"
diff --git a/stdlib/tst-strtol-locale.c b/stdlib/tst-strtol-locale.c
new file mode 100644
index 0000000..739d0b0
--- /dev/null
+++ b/stdlib/tst-strtol-locale.c
@@ -0,0 +1,25 @@
+/* Test strtol functions work with all ASCII letters in Turkish
+   locales (bug 19242).  Narrow string version.
+   Copyright (C) 2015 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#define CHAR char
+#define SFMT "\"%s\""
+#define FNPFX strto
+#define L_(C) C
+
+#include <tst-strtol-locale-main.c>
diff --git a/wcsmbs/Makefile b/wcsmbs/Makefile
index f333fa8..c1bb937 100644
--- a/wcsmbs/Makefile
+++ b/wcsmbs/Makefile
@@ -47,13 +47,14 @@ strop-tests :=  wcscmp wcsncmp wmemcmp wcslen wcschr wcsrchr wcscpy wcsnlen \
 		wcscspn wmemchr wmemset
 tests := tst-wcstof wcsmbs-tst1 tst-wcsnlen tst-btowc tst-mbrtowc \
 	 tst-wcrtomb tst-wcpncpy tst-mbsrtowcs tst-wchar-h tst-mbrtowc2 \
-	 tst-c16c32-1 wcsatcliff $(addprefix test-,$(strop-tests))
+	 tst-c16c32-1 wcsatcliff tst-wcstol-locale \
+	 $(addprefix test-,$(strop-tests))
 
 include ../Rules
 
 ifeq ($(run-built-tests),yes)
 LOCALES := de_DE.ISO-8859-1 de_DE.UTF-8 en_US.ANSI_X3.4-1968 hr_HR.ISO-8859-2 \
-	   ja_JP.EUC-JP zh_TW.EUC-TW
+	   ja_JP.EUC-JP zh_TW.EUC-TW tr_TR.UTF-8 tr_TR.ISO-8859-9
 include ../gen-locales.mk
 
 $(objpfx)tst-btowc.out: $(gen-locales)
@@ -62,6 +63,7 @@ $(objpfx)tst-mbrtowc.out: $(gen-locales)
 $(objpfx)tst-mbrtowc2.out: $(gen-locales)
 $(objpfx)tst-wcrtomb.out: $(gen-locales)
 $(objpfx)wcsmbs-tst1.out: $(gen-locales)
+$(objpfx)tst-wcstol-locale.out: $(gen-locales)
 endif
 
 CFLAGS-wcwidth.c = -I../wctype
diff --git a/wcsmbs/tst-wcstol-locale.c b/wcsmbs/tst-wcstol-locale.c
new file mode 100644
index 0000000..ea27e8e
--- /dev/null
+++ b/wcsmbs/tst-wcstol-locale.c
@@ -0,0 +1,25 @@
+/* Test strtol functions work with all ASCII letters in Turkish
+   locales (bug 19242).  Wide string version.
+   Copyright (C) 2015 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#define CHAR wchar_t
+#define SFMT "L\"%ls\""
+#define FNPFX wcsto
+#define L_(C) L ## C
+
+#include "../stdlib/tst-strtol-locale-main.c"

-- 
Joseph S. Myers
joseph@codesourcery.com