Thread: regression test errors: netbsd 1.3.2/i386
I just compiled up the current snapshot for a test under netbsd 1.3.2/i386. Everything works except the followng: int8 .. failed <-- seems like large numbers show as negative geometry .. failed <-- seems like roundoff differences union .. failed <-- all differences have extra rows Perhaps these are all "well-known" problems, but if any details on failures would help, let me know. If the geometry errors really are roundoffs, I'll be glad to make a netbsd "expected" file before the final release. Cheers, Brook
Brook Milligan wrote:
>
> I just compiled up the current snapshot for a test under netbsd
> 1.3.2/i386. Everything works except the followng:
>
> int8 .. failed <-- seems like large numbers show as negative
Your system is apparently not yet supported for 8 byte integers. If you
want netbsd to be supported, you need to figure out if you have support
for those in your compiler (gcc seems to have it across several
processors) and you need i/o support for them in printf/scanf.
configure tries to detect these automatically so look at what configure
tries to do and see why it missed for your system.
> union .. failed <-- all differences have extra rows
? Do other people see a problem with unions?
- Tom
> I just compiled up the current snapshot for a test under netbsd
> 1.3.2/i386. Everything works except the followng:
>
> int8 .. failed <-- seems like large numbers show as negative
Your system is apparently not yet supported for 8 byte integers. If you
want netbsd to be supported, you need to figure out if you have support
for those in your compiler (gcc seems to have it across several
processors) and you need i/o support for them in printf/scanf.
configure tries to detect these automatically so look at what configure
tries to do and see why it missed for your system.
Configure does check (and passes), but doesn't seem to record anything
about the int8 tests. As a result, config.h (and hence int8.h) knows
nothing about the results of the configure test. The patches below
get configure to record the tests in config.h. With these patches (to
last night's snapshot) the int8 regression test passes on netbsd
1.3.2/i386. I'm actually surprised they work without these patches
for anyone, really.
Cheers,
Brook
===========================================================================
--- configure.in.orig Thu Sep 10 01:00:35 1998
+++ configure.in Thu Sep 10 15:21:36 1998
@@ -572,6 +572,7 @@
[AC_DEFINE(HAVE_LONG_INT_64) AC_MSG_RESULT(yes)],
AC_MSG_RESULT(no),
AC_MSG_RESULT(assuming not on target machine))
+AC_SUBST(HAVE_LONG_INT_64)
AC_MSG_CHECKING(whether 'long long int' is 64 bits using %lld)
AC_TRY_RUN([#include <stdio.h>
@@ -609,6 +610,7 @@
[AC_DEFINE(HAVE_LONG_LONG_INT_64_lld) AC_MSG_RESULT(yes)],
AC_MSG_RESULT(no),
AC_MSG_RESULT(assuming not on target machine))
+AC_SUBST(HAVE_LONG_LONG_INT_64_lld)
AC_MSG_CHECKING(whether 'long long int' is 64 bits using %qd)
AC_TRY_RUN([#include <stdio.h>
@@ -646,6 +648,7 @@
[AC_DEFINE(HAVE_LONG_LONG_INT_64_qd) AC_MSG_RESULT(yes)],
AC_MSG_RESULT(no),
AC_MSG_RESULT(assuming not on target machine))
+AC_SUBST(HAVE_LONG_LONG_INT_64_qd)
dnl Checks for library functions.
AC_PROG_GCC_TRADITIONAL
===========================================================================
--- include/config.h.in.orig Wed Aug 26 01:01:16 1998
+++ include/config.h.in Thu Sep 10 15:22:45 1998
@@ -222,8 +222,11 @@
/* Set to 1 if type "long int" works and is 64 bits */
#undef HAVE_LONG_INT_64
-/* Set to 1 if type "long long int" works and is 64 bits */
-#undef HAVE_LONG_LONG_INT_64
+/* Set to 1 if type "long long int" works with %lld and is 64 bits */
+#undef HAVE_LONG_LONG_INT_64_lld
+
+/* Set to 1 if type "long long int" works with %qd and is 64 bits */
+#undef HAVE_LONG_LONG_INT_64_qd
/*
* Code below this point should not require changes
> I just compiled up the current snapshot for a test under netbsd
> 1.3.2/i386. Everything works except the followng:
>
> union .. failed <-- all differences have extra rows
? Do other people see a problem with unions?
Still having problems with unions from last night's snapshot. The
diffs are below in case they can help pinpoint the problem. Does no
one see anything like these additional zeros?
Cheers,
Brook
===========================================================================
--- expected/union.out Fri May 29 07:22:42 1998
+++ results/union.out Thu Sep 10 16:51:42 1998
@@ -122,7 +122,10 @@
-34.84
-1.2345678901234e-200
0
-(5 rows)
+0
+0
+0
+(8 rows)
QUERY: SELECT f1 AS ten FROM FLOAT8_TBL
UNION ALL
@@ -131,15 +134,19 @@
---------------------
0
-34.84
+0
+0
-1004.3
-1.2345678901234e+200
-1.2345678901234e-200
0
-34.84
+0
+0
-1004.3
-1.2345678901234e+200
-1.2345678901234e-200
-(10 rows)
+(14 rows)
QUERY: SELECT f1 AS nine FROM FLOAT8_TBL
UNION
@@ -153,9 +160,11 @@
-34.84
-1.2345678901234e-200
0
+0
+0
123456
2147483647
-(9 rows)
+(11 rows)
QUERY: SELECT f1 AS ten FROM FLOAT8_TBL
UNION ALL
@@ -164,6 +173,8 @@
---------------------
0
-34.84
+0
+0
-1004.3
-1.2345678901234e+200
-1.2345678901234e-200
@@ -172,7 +183,7 @@
-123456
2147483647
-2147483647
-(10 rows)
+(12 rows)
QUERY: SELECT f1 AS five FROM FLOAT8_TBL
WHERE f1 BETWEEN -1e6 AND 1e6
@@ -185,8 +196,9 @@
-34.84
-1.2345678901234e-200
0
+ 0
123456
-(5 rows)
+(6 rows)
QUERY: SELECT f1 AS five FROM VARCHAR_TBL
UNION
I made some more int8 fixes today, so make sure you have tested the current version before rooting around. > > I just compiled up the current snapshot for a test under netbsd > > 1.3.2/i386. Everything works except the followng: > > > > int8 .. failed <-- seems like large numbers show as negative > > Your system is apparently not yet supported for 8 byte integers. If you > want netbsd to be supported, you need to figure out if you have support > for those in your compiler (gcc seems to have it across several > processors) and you need i/o support for them in printf/scanf. > > configure tries to detect these automatically so look at what configure > tries to do and see why it missed for your system. > > Configure does check (and passes), but doesn't seem to record anything > about the int8 tests. As a result, config.h (and hence int8.h) knows > nothing about the results of the configure test. The patches below > get configure to record the tests in config.h. With these patches (to > last night's snapshot) the int8 regression test passes on netbsd > 1.3.2/i386. I'm actually surprised they work without these patches > for anyone, really. Yes, I forgot to rename some stuff in config.h. I did that today. > > Cheers, > Brook > > =========================================================================== > --- configure.in.orig Thu Sep 10 01:00:35 1998 > +++ configure.in Thu Sep 10 15:21:36 1998 > @@ -572,6 +572,7 @@ > [AC_DEFINE(HAVE_LONG_INT_64) AC_MSG_RESULT(yes)], > AC_MSG_RESULT(no), > AC_MSG_RESULT(assuming not on target machine)) > +AC_SUBST(HAVE_LONG_INT_64) > > AC_MSG_CHECKING(whether 'long long int' is 64 bits using %lld) > AC_TRY_RUN([#include <stdio.h> > @@ -609,6 +610,7 @@ > [AC_DEFINE(HAVE_LONG_LONG_INT_64_lld) AC_MSG_RESULT(yes)], > AC_MSG_RESULT(no), > AC_MSG_RESULT(assuming not on target machine)) > +AC_SUBST(HAVE_LONG_LONG_INT_64_lld) > > AC_MSG_CHECKING(whether 'long long int' is 64 bits using %qd) > AC_TRY_RUN([#include <stdio.h> > @@ -646,6 +648,7 @@ > [AC_DEFINE(HAVE_LONG_LONG_INT_64_qd) AC_MSG_RESULT(yes)], > AC_MSG_RESULT(no), > AC_MSG_RESULT(assuming not on target machine)) > +AC_SUBST(HAVE_LONG_LONG_INT_64_qd) > > dnl Checks for library functions. > AC_PROG_GCC_TRADITIONAL > =========================================================================== > --- include/config.h.in.orig Wed Aug 26 01:01:16 1998 > +++ include/config.h.in Thu Sep 10 15:22:45 1998 > @@ -222,8 +222,11 @@ > /* Set to 1 if type "long int" works and is 64 bits */ > #undef HAVE_LONG_INT_64 > > -/* Set to 1 if type "long long int" works and is 64 bits */ > -#undef HAVE_LONG_LONG_INT_64 > +/* Set to 1 if type "long long int" works with %lld and is 64 bits */ > +#undef HAVE_LONG_LONG_INT_64_lld > + > +/* Set to 1 if type "long long int" works with %qd and is 64 bits */ > +#undef HAVE_LONG_LONG_INT_64_qd > > /* > * Code below this point should not require changes > > -- Bruce Momjian | 830 Blythe Avenue maillist@candle.pha.pa.us | Drexel Hill, Pennsylvania 19026 + If your life is a hard drive, | (610) 353-9879(w) + Christ can be your backup. | (610) 853-3000(h)