Experimental Physics and
Industrial Control System

Dirk Zimoch <[email protected]> · Mon, 21 Jul 2014 10:15:48 +0200

On 16.07.2014 20:40, Mark Rivers wrote:

The problem is
     T  * volatile * buffer;
in epicsRingPointer.h.
I am not sure if "volatile" really makes any sense here.I strongly suggest NOT removing the volatile keyword.

In October 2008 we struggled with a very nasty bug in epicsRingPointer.h.  The symptoms were:

- Only failed on multi-processor systems
- Only failed with recent versions of gcc with optimization enabled
- Only failed 1 in 100,000 times

Eric Norum fixed the problem.  The changes included the following (comparison of epicsRingPointer.h from base 3.14.8.2 and 3.14.12.4)

-private: // Data
-    int nextPush;
-    int nextPop;
+private: /* Data */
+    volatile int nextPush;
+    volatile int nextPop;
      int size;
-    T **buffer;
+    T  * volatile * buffer;
  };

So he added that volatile keyword, and it would need a lot of testing to prove it is not needed.

Mark

In that case the volatile needs to be removed for the delete, because it

seems that the compiler thinks that there is no delete operator for

arrays of volatiles and implicitly converting volatile away is

considered impossible.

I just see that this is not a problem with vxWorks 6.9 but with the

Windows host build.

Eric, have you tested your change with Windows?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Dirk Zimoch
Sent: Wednesday, July 16, 2014 7:00 AM
To: [email protected]
Subject: Re: EPICS on VxWorks6.9

Newer C++ compilers consider more and more things in EPICS errors. And
maybe they are right.

The problem is
      T  * volatile * buffer;
in epicsRingPointer.h.

I am not sure if "volatile" really makes any sense here.

The other possibility is to cast away the "void" in the destructor.

delete [ ] (T*) buffer;

Or whatever the nicer C++ syntax is (probably using const_cast<T*>()?)

Dirk

On 16.07.2014 08:14, Amit Chauhan wrote:

Hi,

We are trying to put EPICS on VxWorks.

Various information related to our setup are:

           EPICS base version : 3.14.12.4

           VxWorks Version : 6.9

Workbench version : 3.3

           Host Architecture : win32 on x86

           Target Architecture: ppc-32

           Target Board :  MVME-5500 with PowerPC 7447 processor

On running the 'make' command, the EPICS build stops after a while with
an error.

*_Following is the output of the make command:_*

make -C ./configure install

make[1]: Entering directory `C:/EPICS/base-3-14-12-4/configure'

perl ../src/tools/makeMakefile.pl O.win32-x86 ../..

perl ../src/tools/makeMakefile.pl O.vxWorks-ppc32 ../..

perl -MExtUtils::Command -e mkpath O.Common

make -C O.win32-x86 -f ../Makefile TOP=../.. T_A=win32-x86 install

make[2]: Entering directory `C:/EPICS/base-3-14-12-4/configure/O.win32-x86'

make[2]: Nothing to be done for `install'.

make[2]: Leaving directory `C:/EPICS/base-3-14-12-4/configure/O.win32-x86'

make -C O.vxWorks-ppc32 -f ../Makefile TOP=../.. T_A=vxWorks-ppc32 install

make[2]: Entering directory
`C:/EPICS/base-3-14-12-4/configure/O.vxWorks-ppc32'

make[2]: Nothing to be done for `install'.

make[2]: Leaving directory
`C:/EPICS/base-3-14-12-4/configure/O.vxWorks-ppc32'

make[1]: Leaving directory `C:/EPICS/base-3-14-12-4/configure'

make -C ./src install

make[1]: Entering directory `C:/EPICS/base-3-14-12-4/src'

make -C ./tools install

make[2]: Entering directory `C:/EPICS/base-3-14-12-4/src/tools'

perl ../../src/tools/makeMakefile.pl O.win32-x86 ../../..

perl ../../src/tools/makeMakefile.pl O.vxWorks-ppc32 ../../..

perl -MExtUtils::Command -e mkpath O.Common

make -C O.win32-x86 -f ../Makefile TOP=../../.. T_A=win32-x86 install

make[3]: Entering directory `C:/EPICS/base-3-14-12-4/src/tools/O.win32-x86'