Thanks for reporting the problem.
On 04/14/2017 02:27 PM, Gavin Smith wrote:
> /opt/solarisstudio12.3/bin/c99 -Xc -D_XPG6 -DHAVE_CONFIG_H -I. -I../..
> -D_REENTRANT -I/opt/csw/include -c -o regex.o regex.c
> "regex_internal.h", line 105: warning: macro redefined: gettext
That's odd. Where was the gettext macro originally defined?
> "regex_internal.h", line 734: warning: token-less macro argument
This warns about the use of a C99-ism in code that is protected by "#if
199901L <= __STDC_VERSION__", so it must be a false alarm, and you're
right tno not worry about it.
> Assembler: "/tmp/yabeAAAiUaidC", line 19010 : Value out of range
> "/tmp/yabeAAAiUaidC", line 19031 : Value out of range
Ouch. This appears to be a compiler bug, as a compiler should never
generate invalid assembly-language code. I cannot reproduce the problem
on Solaris 10 sparc (32-bit) with Oracle Developer Studio 12.5. You are
on Solaris 10 x86 (32-bit) and are using 12.3. Do you observe the same
compiler bug with 12.5 or 12.6 beta? If so, I suggest reporting it as a
bug to Oracle.
What happens if you compile with -xO2?
Although I cannot reproduce the problem in my environment, I think I see
why it's happening. I installed the attached patches into Gnulib; do
they fix your problem?
>From 3c20784ac87d0f0748a86f3bc6f06e18694658da Mon Sep 17 00:00:00 2001
From: Paul Eggert <egg...@penguin.cs.ucla.edu>
Date: Fri, 14 Apr 2017 17:38:58 -0700
Subject: [PATCH 1/2] intprops: port to Oracle Studio 12.3 x86
Problem reported by Gavin Smith in:
http://lists.gnu.org/archive/html/bug-gnulib/2017-04/msg00049.html
* lib/intprops.h (_GL_INT_OP_WRAPV_VIA_UNSIGNED):
Convert unsigned to signed via the usual rather than the standard way,
to avoid a compiler bug in Oracle Studio 12.3 x86.
---
ChangeLog | 9 +++++++++
lib/intprops.h | 41 +++++++++++++++++++++++++++++------------
2 files changed, 38 insertions(+), 12 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index ae1a8bd..97d1e59 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2017-04-14 Paul Eggert <egg...@penguin.cs.ucla.edu>
+
+ intprops: port to Oracle Studio 12.3 x86
+ Problem reported by Gavin Smith in:
+ http://lists.gnu.org/archive/html/bug-gnulib/2017-04/msg00049.html
+ * lib/intprops.h (_GL_INT_OP_WRAPV_VIA_UNSIGNED):
+ Convert unsigned to signed via the usual rather than the standard way,
+ to avoid a compiler bug in Oracle Studio 12.3 x86.
+
2017-04-08 Paul Eggert <egg...@cs.ucla.edu>
getopt: prefer - to _ in new file names
diff --git a/lib/intprops.h b/lib/intprops.h
index eb06b69..220b509 100644
--- a/lib/intprops.h
+++ b/lib/intprops.h
@@ -442,17 +442,34 @@ verify (TYPE_WIDTH (unsigned int) == UINT_WIDTH);
((overflow (a, b) \
|| (EXPR_SIGNED ((a) op (b)) && ((a) op (b)) < (tmin)) \
|| (tmax) < ((a) op (b))) \
- ? (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t, tmin, tmax), 1) \
- : (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t, tmin, tmax), 0))
-
-/* Return A <op> B, where the operation is given by OP. Use the
- unsigned type UT for calculation to avoid overflow problems.
- Convert the result to type T without overflow by subtracting TMIN
- from large values before converting, and adding it afterwards.
- Compilers can optimize all the operations except OP. */
-#define _GL_INT_OP_WRAPV_VIA_UNSIGNED(a, b, op, ut, t, tmin, tmax) \
- (((ut) (a) op (ut) (b)) <= (tmax) \
- ? (t) ((ut) (a) op (ut) (b)) \
- : ((t) (((ut) (a) op (ut) (b)) - (tmin)) + (tmin)))
+ ? (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t), 1) \
+ : (*(r) = _GL_INT_OP_WRAPV_VIA_UNSIGNED (a, b, op, ut, t), 0))
+
+/* Return A <op> B, where the operation is given by OP. Return the
+ low-order bits of the mathematically-correct answer. Use the
+ unsigned type UT for calculation to avoid undefined behavior on
+ signed integer overflow. Assume that conversion to the result type
+ T yields the low-order bits in the usual way. UT and T are the
+ same width, T is two's complement, and there is no padding or trap
+ representations.
+
+ According to the C standard, converting UT to T yields an
+ implementation-defined result or signal for values outside T's range.
+ So, the standard way to convert UT to T is to subtract TMIN from
+ greater-than-TMAX values before converting them to T, and to add
+ TMIN afterwards, where TMIN and TMAX are T's extrema.
+ However, in practice there is no need to subtract and add TMIN.
+ E.g., GCC converts to signed integers in the usual way; see:
+ https://gcc.gnu.org/onlinedocs/gcc/Integers-implementation.html
+ All other known C compilers are similar to GCC in this respect.
+ Furthermore, Oracle Studio Studio 12.3 x86 has a bug when
+ implementing the standard way; see:
+ http://lists.gnu.org/archive/html/bug-gnulib/2017-04/msg00049.html
+
+ So, implement this operation in the usual way rather than in
+ the standard way. */
+
+#define _GL_INT_OP_WRAPV_VIA_UNSIGNED(a, b, op, ut, t) \
+ ((t) ((ut) (a) op (ut) (b)))
#endif /* _GL_INTPROPS_H */
--
2.9.3
>From 931d22eb17dca7efc9c355176fe3303d3ab33bb9 Mon Sep 17 00:00:00 2001
From: Paul Eggert <egg...@penguin.cs.ucla.edu>
Date: Fri, 14 Apr 2017 18:55:42 -0700
Subject: [PATCH 2/2] intprops: try to avoid tickling similar bugs
* lib/intprops.h (_GL_INT_OP_CALC): Document that UT no longer
needs to be the same width as T; it can be wider.
Change callers so that UT is at least as wide as unsigned int,
as I suspect that this is less likely to run into compiler bugs.
---
ChangeLog | 6 ++++++
lib/intprops.h | 19 +++++++++----------
2 files changed, 15 insertions(+), 10 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index 97d1e59..5ed1517 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,11 @@
2017-04-14 Paul Eggert <egg...@penguin.cs.ucla.edu>
+ intprops: try to avoid tickling similar bugs
+ * lib/intprops.h (_GL_INT_OP_CALC): Document that UT no longer
+ needs to be the same width as T; it can be wider.
+ Change callers so that UT is at least as wide as unsigned int,
+ as I suspect that this is less likely to run into compiler bugs.
+
intprops: port to Oracle Studio 12.3 x86
Problem reported by Gavin Smith in:
http://lists.gnu.org/archive/html/bug-gnulib/2017-04/msg00049.html
diff --git a/lib/intprops.h b/lib/intprops.h
index 220b509..ca8740d 100644
--- a/lib/intprops.h
+++ b/lib/intprops.h
@@ -389,10 +389,10 @@ verify (TYPE_WIDTH (unsigned int) == UINT_WIDTH);
(_Generic \
(*(r), \
signed char: \
- _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned char, \
+ _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
signed char, SCHAR_MIN, SCHAR_MAX), \
short int: \
- _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned short int, \
+ _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
short int, SHRT_MIN, SHRT_MAX), \
int: \
_GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
@@ -406,10 +406,10 @@ verify (TYPE_WIDTH (unsigned int) == UINT_WIDTH);
#else
# define _GL_INT_OP_WRAPV(a, b, r, op, builtin, overflow) \
(sizeof *(r) == sizeof (signed char) \
- ? _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned char, \
+ ? _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
signed char, SCHAR_MIN, SCHAR_MAX) \
: sizeof *(r) == sizeof (short int) \
- ? _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned short int, \
+ ? _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
short int, SHRT_MIN, SHRT_MAX) \
: sizeof *(r) == sizeof (int) \
? _GL_INT_OP_CALC (a, b, r, op, overflow, unsigned int, \
@@ -431,9 +431,8 @@ verify (TYPE_WIDTH (unsigned int) == UINT_WIDTH);
/* Store the low-order bits of A <op> B into *R, where the operation
is given by OP. Use the unsigned type UT for calculation to avoid
- overflow problems. *R's type is T, with extremal values TMIN and
- TMAX. T must be a signed integer type. Return 1 if the result
- overflows. */
+ overflow problems. *R's type is T, with extrema TMIN and TMAX.
+ T must be a signed integer type. Return 1 if the result overflows. */
#define _GL_INT_OP_CALC(a, b, r, op, overflow, ut, t, tmin, tmax) \
(sizeof ((a) op (b)) < sizeof (t) \
? _GL_INT_OP_CALC1 ((t) (a), (t) (b), r, op, overflow, ut, t, tmin, tmax) \
@@ -449,9 +448,9 @@ verify (TYPE_WIDTH (unsigned int) == UINT_WIDTH);
low-order bits of the mathematically-correct answer. Use the
unsigned type UT for calculation to avoid undefined behavior on
signed integer overflow. Assume that conversion to the result type
- T yields the low-order bits in the usual way. UT and T are the
- same width, T is two's complement, and there is no padding or trap
- representations.
+ T yields the low-order bits in the usual way. UT is at least as
+ wide as T and is no narrower than unsigned int, T is two's
+ complement, and there is no padding or trap representations.
According to the C standard, converting UT to T yields an
implementation-defined result or signal for values outside T's range.
--
2.9.3