Experimental Physics and
Industrial Control System

Dirk Zimoch <[email protected]> · Fri, 24 Nov 2017 08:33:23 +0100

Hi Andrew,

So far I needed only that one file. VxWorks 6.2 and 6.3 use gcc 3.3.2

and 3.4.4. They may not have __sync_lock_test_and_set().

Actually I do not use vxWorks 6.2 or 6.3 but 5.5. :-)
If switching to vxWorks 6 I will go for 6.7 or newer.

But since we have those versions for another project (not using EPICS),

I thought it may be good to check for maximum compatibility.

As I get other errors now, vxWorks 6.3- compatibility may not be

possible. I need to try more fixes.

I will set up a Jenkis server in the next few days. However I fear due

to PSI networking regulations it will not be accessible from outside

PSI. Is it OK to send the messages to core-talk?

Dirk

On 23.11.2017 19:23, Johnson, Andrew N. wrote:

Hi Dirk,

The complete Boost version used in Base-7.0.1-pre1 can be found in the
any version of the V4 dBCommonCPP application; the full source is huge
though so I only pulled the files into 7.0.1 that I knew were needed. If
you can keep track of and send me a list of all the additional Boost
files that you need I will add them before the -rc1 release.

If you want EPICS to support VxWorks versions before 6.8 you *do* need
to actively support it by doing builds , running tests and reporting
problems (preferably as early in the development process as possible).
APS doesn’t use or have either of these versions installed so we can’t
do that for you.

- Andrew

--
Sent from my iPad

On Nov 23, 2017, at 9:34 AM, Dirk Zimoch <[email protected]
<mailto:[email protected]>> wrote:

On 22.11.2017 13:28, Dirk Zimoch wrote:

The vxWorks 6.2 and 6.3 builds failed here:
In file included from
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/detail/spinlock_pool.hpp:25,

                from
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/shared_ptr.hpp:38,

                from
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/shared_ptr.hpp:17,
                from
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/tr1/memory.hpp:56,
                from ../../src/misc/pv/sharedPtr.h:133,
                from ../pv/pvType.h:41,
                from ../pv/pvIntrospect.h:17,
                from ../pv/pvData.h:24,
                from ../../src/json/print.cpp:11:
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/detail/spinlock.hpp:47:52:
boost/smart_ptr/detail/spinlock_pt.hpp: No such file or directory

When comparing with vxWorks 6.6 and 6.7, I see the following includes:
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/detail/spinlock_pool.hpp
\
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/detail/spinlock.hpp
\
/usr/local/epics/base-7.0.1/include/os/vxWorks/boost/smart_ptr/detail/spinlock_sync.hpp
\

That means in spinlock.hpp vxWorks 6.6 and 6.7 take the branch
#elif defined( BOOST_SP_HAS_SYNC )
#  include <boost/smart_ptr/detail/spinlock_sync.hpp>

But vxWorks 6.2 and 6.3 want to take the branch
#elif defined(BOOST_HAS_PTHREADS)
#  include <boost/smart_ptr/detail/spinlock_pt.hpp>

I checked that vxWorks 6 indeed defines _POSIX_THREADS as checked in
posix_features.hpp:
     // POSIX defines _POSIX_THREADS > 0 for pthread support,
     // however some platforms define _POSIX_THREADS without
     // a value, hence the (_POSIX_THREADS+0 >= 0) check.
     // Strictly speaking this may catch platforms with a
     // non-functioning stub <pthreads.h>, but such occurrences should
     // occur very rarely if at all.
#     if defined(_POSIX_THREADS) && (_POSIX_THREADS+0 >= 0) &&
!defined(BOOST_HAS_WINTHREADS) && !defined(BOOST_HAS_MPTASKS)
#        define BOOST_HAS_PTHREADS
#     endif

But the file boost/smart_ptr/detail/spinlock_pt.hpp is missing. I
copied it from another boost installation to see what happens.
VxWorks 6.2 has pthread_mutex_lock(). It implements it with semTake()
but does some magic before to follow posix rules. Maybe it is better
to use semTake directly in a vxWorks specific implementation?
Also I don't think that vxWorks 6.2 supports multi-core processors.
Thus intLock() may be faster and still safe?

Now I get another error with vxWorks 6.2:

In file included from ../../src/misc/pvUnitTest.cpp:12:
../../src/misc/pv/pvUnitTest.h: In function `detail::testPassx
  testFieldEqual(const boost::shared_ptr<const
epics::pvData::PVStructure>&,
  const char*, typename PVD::value_type)':
../../src/misc/pv/pvUnitTest.h:156: error: syntax error before `>' token
../../src/misc/pv/pvUnitTest.h:158: error: syntax error before `>' token
../../src/misc/pv/pvUnitTest.h:160: error: syntax error before `>' token
../../src/misc/pv/pvUnitTest.h: In function `detail::testPassx
  testFieldEqual(const boost::shared_ptr<const
epics::pvData::PVStructure>&,
  const char*, typename PVD::const_svector)':
../../src/misc/pv/pvUnitTest.h:177: error: syntax error before `>' token

And with vxWorks 6.3:

/afs/psi.ch/project/vxworks/WindRiverGPP3.3/gnu/3.4.4-vxworks-6.3/x86-linux2/bin/../lib/gcc/powerpc-wrs-vxworks/3.4.4/../../../../include/c++/3.4.4/xstring:
<http://psi.ch/project/vxworks/WindRiverGPP3.3/gnu/3.4.4-vxworks-6.3/x86-linux2/bin/../lib/gcc/powerpc-wrs-vxworks/3.4.4/../../../../include/c++/3.4.4/xstring:>
In function `void testLogicException()':
/afs/psi.ch/project/vxworks/WindRiverGPP3.3/gnu/3.4.4-vxworks-6.3/x86-linux2/bin/../lib/gcc/powerpc-wrs-vxworks/3.4.4/../../../../include/c++/3.4.4/xstring:78:
<http://psi.ch/project/vxworks/WindRiverGPP3.3/gnu/3.4.4-vxworks-6.3/x86-linux2/bin/../lib/gcc/powerpc-wrs-vxworks/3.4.4/../../../../include/c++/3.4.4/xstring:78:>
internal compiler error: in expand_expr_real, at expr.c:8417
Please submit a full bug report,
with preprocessed source if appropriate.
Contact [email protected] <mailto:[email protected]> for
instructions.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System