I really like Shuah's idea for a configuration parameter to allow this.
Garrett, could you try the attached patch and let us know if it works
for you? If you call configure with a --enable-non32bit-int parameter
now, a warning message will be generated and the build will be allowed
on targets where ints are not 32 bits.
Best Regards,
Ric
On Wed, 2009-08-26 at 00:16 +0000, Garrett Cooper wrote:
> On Tue, Aug 25, 2009 at 10:42 AM, Shuah Khan<[email protected]> wrote:
> > On Tue, 2009-08-25 at 15:19 +0000, Anton Pak wrote:
> >> I tried many times OpenHPI on x86_64.
> >> It worked fine in the configurations x86 client - x86_64 daemon and x86_64
> >> client - x86 daemon.
> >> However, the functioning on non-x86 64-bit platform or on Itanium is still
> >> an open question for me.
> >>
> >> Anton Pak
> >
> > Instead of changing the error to a warning, itcould be controlled via a
> > config option, so this error could be ignored on certain architectures
> > using this config tune.
> >
> > -- Shuah
> >>
> >> On Tue, 25 Aug 2009 19:12:03 +0400, Shuah Khan <[email protected]> wrote:
> >>
> >> > On Tue, 2009-08-25 at 05:42 +0000, [email protected] wrote:
> >> >> Hello!
> >> >>
> >> >> We have the following lines in include/SaHpi.h:
> >> >>
> >> >> typedef unsigned char SaHpiUint8T;
> >> >> typedef unsigned short SaHpiUint16T;
> >> >> typedef unsigned int SaHpiUint32T;
> >> >> typedef signed char SaHpiInt8T;
> >> >> typedef signed short SaHpiInt16T;
> >> >> typedef signed int SaHpiInt32T;
> >> >>
> >> >> Also I suspect there can be marshalling issues, i.e. when
> >> >> daemon on platform with sizeof(int) == 4 but library is not and vice
> >> >> versa.
> >> >>
> >> >> Anton Pak
> >> >
> >> > The proposed fix might help compile OpenHpi on a 64-bit platform,
> >> > however I agree with the concerns from others that we will see run-time
> >> > issues. Running 32-bit binary and libraries on a 64-bit platform might
> >> > be an option.
> >> >
> >> > -- Shuah
> >> >
> >> >
> >> >>
> >> >>
> >> >> > Hi Garrett,
> >> >> >
> >> >> > I was wondering about your addition of the cross_compiling test. If
> >> >> > there is OpenHPI code that may behave badly on a system where an int
> >> >> > isn't 4 bytes, I'm thinking that a warning should be issued even for a
> >> >> > cross compile to such an architecture as well. Like:
> >> >> >
> >> >> > if test "x$OH_SIZEOF_INT" != "x4"; then
> >> >> > AC_MSG_WARN([
> >> >> > *** int is not 4 bytes, it is $OH_SIZEOF_INT bytes on this platform
> >> >> > *** undefined behavior may result from this.
> >> >> > ])
> >> >> > fi
> >> >> >
> >> >> > Will having a warning instead of an error cause problems for LTP when
> >> >> > cross compiling?
> >> >> >
> >> >> > Does anyone else in the OpenHPI community oppose changing this error
> >> >> to
> >> >> > a warning?
> >> >> >
> >> >> > Best Regards,
> >> >> > Ric White
> >> >> >
> >> >> >
> >> >> > On Wed, 2009-08-19 at 23:11 +0000, Garrett Cooper wrote:
> >> >> >> Hi,
> >> >> >> Found this bug while trying to cross-compile with the
> >> >> >> openhpi-test-suite in LTP, and made the ERROR into WARN. Please fix
> >> >> >> this item as per the attached patch.
> >> >> >> Thanks,
> >> >> >> -Garrett
> >> >> >>
> >> >> >> PS Please CC my address in all correspondence w.r.t. this email
> >> >> thread.
> >> >> >>
> >> >> >> Summary:
> >> >> >>
> >> >> >> 1. int != 4 shouldn't be checked for while cross-compiling.
> >> >> >> 2. Warn instead of erroring out, because we have a _lot_ of 64-bit
> >> >> >> platforms that we test on which will be negatively impacted by this
> >> >> >> check.
> >> >> >>
> >> >> >> Signed-off-by: Garrett Cooper <[email protected]>
> >> >> >>
> >> >> >> ---
> >> >> >>
> >> >> /nfs.mac/ltp-upgrade/contrib/ltp/src/testcases/open_hpi_testsuite//configure.ac.orig
> >> >> 2009-08-19
> >> >> >> 16:04:08.000000000 -0700
> >> >> >> +++
> >> >> >>
> >> >> /nfs.mac/ltp-upgrade/contrib/ltp/src/testcases/open_hpi_testsuite//configure.ac
> >> >> 2009-08-19
> >> >> >> 16:10:47.000000000 -0700
> >> >> >> @@ -44,14 +44,13 @@
> >> >> >> AC_PROG_LN_S
> >> >> >> AC_PROG_MAKE_SET
> >> >> >>
> >> >> >> -dnl die on int != 32bits. This is too instrumental to our code
> >> >> right
> >> >> >> now.
> >> >> >> +dnl Warn when int != 32bits. This is too instrumental to our code
> >> >> >> right now.
> >> >> >> dnl AC_MSG_CHECKING(int is 4 bytes)
> >> >> >> OH_SET_SIZES
> >> >> >> -if test "x$OH_SIZEOF_INT" != "x4"; then
> >> >> >> - AC_MSG_ERROR([
> >> >> >> +if test x$cross_compiling != xyes && test "x$OH_SIZEOF_INT" != "x4";
> >> >> >> then
> >> >> >> + AC_MSG_WARN([
> >> >> >> *** int is not 4 bytes, it is $OH_SIZEOF_INT bytes on this platform
> >> >> >> -*** undefined behavior will result
> >> >> >> -*** please contact the openhpi development team to fix this
> >> >> >> +*** undefined behavior may result from this.
> >> >> >> ])
> >> >> >> fi
>
> Shuah,
> Whichever way you find necessary is fine as long as it doesn't
> break builds and there are sufficient warnings at both buildtime and
> at runtime that say `Hey -- this code might not work because of
> integer sizing issues!' so you guys can cover yourself from FAQ-like
> questions as much as possible :).
> Thanks,
> -Garrett
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now. http://p.sf.net/sfu/bobj-july
> _______________________________________________
> Openhpi-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/openhpi-devel
--- configure.ac 2009-08-26 18:34:15.000000000 -0600
+++ configure.ac.non32 2009-08-26 18:33:06.000000000 -0600
@@ -44,15 +44,32 @@
AC_PROG_LN_S
AC_PROG_MAKE_SET
+enabled_non32bit="no"
+AC_ARG_ENABLE([non32bit-int],
+ [ --enable-non32bit-int Allow compile on non 32bit int architectures [[default=no]]],
+ [if test "x$enableval" = "xyes"; then
+ enabled_non32bit="yes"
+ fi])
+
dnl die on int != 32bits. This is too instrumental to our code right now.
dnl AC_MSG_CHECKING(int is 4 bytes)
OH_SET_SIZES
if test "x$OH_SIZEOF_INT" != "x4"; then
- AC_MSG_ERROR([
-*** int is not 4 bytes, it is $OH_SIZEOF_INT bytes on this platform
-*** undefined behavior will result
-*** please contact the openhpi development team to fix this
-])
+ if test "x$enabled_non32bit" = "xyes"; then
+ AC_MSG_WARN([
+*** Size of int is not 4 bytes, it is $OH_SIZEOF_INT bytes on this platform.
+*** The --enable_non32bit_int arg has been used to force compilation regardless.
+*** Execution on this platform is not supported and could produce undefined
+*** behavior.
+ ])
+ else
+ AC_MSG_ERROR([
+*** Size of int is not 4 bytes, it is $OH_SIZEOF_INT bytes on this platform.
+*** Execution on this platform is not supported and could produce undefined
+*** behavior.
+*** Please contact the openhpi development team to fix this
+ ])
+ fi
fi
AM_CONDITIONAL(CROSS_COMPILATION, test x$cross_compiling = xyes)
@@ -164,7 +181,6 @@
AM_CONDITIONAL(OPENHPID_STATIC, test x$openhpid_static = xyes)
-
AC_ARG_ENABLE([clients],
[ --enable-clients build HPI client programs [[default=yes]]],
[if test "x$enableval" = "xyes"; then
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
Openhpi-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openhpi-devel
