[evlog-dev] [Bug] evlog cannot work properly if compiled with gcc 3.2

Hi,

I find the bitfield test and default test failed when I ran them under my
Redhat 8.0 which uses gcc 3.2. I did a search in google and found below
words in
http://www.redhat.com/docs/manuals/linux/RHL-8.0-Manual/release-notes/x86/.

Note that because of significant ABI fixes, the C++ compiler included in GCC
3.2 produces code that is not binary compatible with previous versions of
GCC, including versions 3.1.x and 2.96. Additionally, there are ABI fixes
for the C compiler related to long long bitfields and where
__attribute__((aligned (xxx))) type definitions used as base type bitfields
work differently than using __attribute__((aligned (xxx))) directly on the
bitfield.

I've made a patch according to this and tested on i386 machines with both
gcc 2.96 and 3.2. However I am not sure it is complete or not. So would
anyone like to double check the code or test it on other platforms (i.e.
ppc, s390)?

Another thing gcc changed is the __LINE__. It uses the last line number of a
macro instead of the first now. Don't know why, but now the line number is
the same as those of ppc and s390 :)


diff -Nuar evlog/user/lib/template/template.c
evlog-new/user/lib/template/template.c
--- evlog/user/lib/template/template.c	2002-10-29 23:57:51.000000000 +0800
+++ evlog-new/user/lib/template/template.c	2003-02-04
01:08:11.000000000 +0800
@@ -2542,6 +2542,7 @@
 	att->ta_flags |= EVL_ATTR_EXISTS;
 }
 
+#define GCC_VERSION __GNUC__ * 100 + __GNUC_MINOR__
 /*
  * att is a (possibly unnamed) bit-field in template t.  bfStart points to
  * the address of the bit-field storage unit currently being accumulated,
@@ -2590,8 +2591,14 @@
 		 */
 		adjustNextPerBitFields(notes, *bfStart, *bfOffset);
 		if (bfBits > intBits) {
+#if GCC_VERSION < 302 /* gcc < 3.2 */
 			alignNextPtr(notes, t->tm_data, __alignof__(long
long));
+#endif
 		} else {
+#if GCC_VERSION >= 302 /* gcc >= 3.2 */
+			if ( (baseType(att) != TY_LONGLONG) &&
+			     (baseType(att) != TY_ULONGLONG) )
+#endif
 			alignNextPtr(notes, t->tm_data,
 				computeAttAlignment(att));
 		}
@@ -2600,6 +2607,7 @@
 	}
 
 	bytesRemaining = notes->pn_end - *bfStart;
+	bytesRemaining = notes->pn_end - *bfStart;
 	stgUnitBytes = bitsToBytes(*bfOffset + bfBits);
 	if (stgUnitBytes > bytesRemaining) {
 		goto outOfData;
diff -Nuar evlog/user/lib/test/defaultTest.oldgcc.out
evlog-new/user/lib/test/defaultTest.oldgcc.out
--- evlog/user/lib/test/defaultTest.oldgcc.out	1970-01-01
08:00:00.000000000 +0800
+++ evlog-new/user/lib/test/defaultTest.oldgcc.out	2003-02-04 00:53:03.
000000000 +0800
@@ -0,0 +1,20 @@
+event type 1001:
+defaultTest.c:54: This event record has a message string, but no further
data.
+
+event type 1002:
+defaultTest.c:60: This event record has a message string plus 3 ints.
+00000000 0B 00 00 00 15 00 00 00  CE 07 00 00
+event type 1003:
+Hey!  Something happened at line 67 of defaultTest.c:
+Basically the same data as for event type 1002,
+but formatted with a non-default template.
+11/21/1998
+event type 1004:
+defaultTest.c:75: This event record has a message string, plus another
string.
+00000000 54 68 69 73 20 69 73 20  74 68 65 20 6F 74 68 65
+00000010 72 20 73 74 72 69 6E 67  2E 00
+event type 1003:
+Hey!  Something happened at line 82 of defaultTest.c:
+This is another instance of event type 1003,
+with a different message and different ints.
+9/30/2000
diff -Nuar evlog/user/lib/test/defaultTest.out
evlog-new/user/lib/test/defaultTest.out
--- evlog/user/lib/test/defaultTest.out	2002-07-25 04:15:40.000000000 +0800
+++ evlog-new/user/lib/test/defaultTest.out	2003-02-04
00:54:28.000000000 +0800
@@ -1,20 +1,20 @@
 event type 1001:
-defaultTest.c:54: This event record has a message string, but no further
data.
+defaultTest.c:56: This event record has a message string, but no further
data.
 
 event type 1002:
-defaultTest.c:60: This event record has a message string plus 3 ints.
+defaultTest.c:63: This event record has a message string plus 3 ints.
 00000000 0B 00 00 00 15 00 00 00  CE 07 00 00
 event type 1003:
-Hey!  Something happened at line 67 of defaultTest.c:
+Hey!  Something happened at line 71 of defaultTest.c:
 Basically the same data as for event type 1002,
 but formatted with a non-default template.
 11/21/1998
 event type 1004:
-defaultTest.c:75: This event record has a message string, plus another
string.
+defaultTest.c:78: This event record has a message string, plus another
string.
 00000000 54 68 69 73 20 69 73 20  74 68 65 20 6F 74 68 65
 00000010 72 20 73 74 72 69 6E 67  2E 00
 event type 1003:
-Hey!  Something happened at line 82 of defaultTest.c:
+Hey!  Something happened at line 86 of defaultTest.c:
 This is another instance of event type 1003,
 with a different message and different ints.
 9/30/2000
diff -Nuar evlog/user/lib/test/defaultTest.sh
evlog-new/user/lib/test/defaultTest.sh
--- evlog/user/lib/test/defaultTest.sh	2002-07-25 04:06:02.000000000 +0800
+++ evlog-new/user/lib/test/defaultTest.sh	2003-02-04
01:39:52.000000000 +0800
@@ -9,7 +9,13 @@
 
 if [ "$ARCH" = "i386" ]
 then
-	TEST_TEMPLATE=defaultTest.out
+	MAJOR=`gcc -dumpversion |cut -d'.' -f1`
+	MINOR=`gcc -dumpversion |cut -d'.' -f2`
+	if [ $MAJOR -eq 3 -a $MINOR -ge 2 -o $MAJOR -gt 3 ]; then
+		TEST_TEMPLATE=defaultTest.out
+	else
+		TEST_TEMPLATE=defaultTest.oldgcc.out
+	fi
 else
 	TEST_TEMPLATE=defaultTest.$ARCH.out
 fi
diff -Nuar evlog/user/Makefile evlog-new/user/Makefile
--- evlog/user/Makefile	2002-12-16 10:35:09.000000000 +0800
+++ evlog-new/user/Makefile	2003-02-04 01:28:45.000000000 +0800
@@ -195,6 +195,7 @@
 	install -m 755 ./lib/test/defaultTest $(STATEDIR)/test
 	install -m 755 ./lib/test/defaultTest.sh $(STATEDIR)/test
 	install -m 644 ./lib/test/defaultTest.out $(STATEDIR)/test
+	install -m 644 ./lib/test/defaultTest.oldgcc.out $(STATEDIR)/test
 	install -m 644 ./lib/test/defaultTest.ia64.out $(STATEDIR)/test
 	install -m 644 ./lib/test/defaultTest.ppc.out $(STATEDIR)/test
 	install -m 644 ./lib/test/defaultTest.s390.out $(STATEDIR)/test

Thanks,
-Zhu Yi


-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com