EPICS Base R3.14.12: Known Problems

Any patch files linked below should be applied at the root of the EPICS base installation directory. Download them, then use the GNU Patch program as follows:

% cd /path/to/base-3.14.12.x
% patch -p0 < /path/to/file.patch

R3.14.12.8 Problems

The following known problem exists in version 3.14.12.8 of EPICS Base:

• 2018-09-17: This release does not build using GNU Make 3.81 because the undefine command was not added until the 3.82 release. Apply this patch to fix the problem if you can't upgrade your version of GNU Make. This bug does not appear in any other Base releases.

The 3.14 branch of Base is now no longer supported, so no further fixes will be documented here or patch files added to this page. Please upgrade to the 3.15 branch.

R3.14.12.7 Problems

The following significant problems have been reported in version 3.14.12.7 of EPICS Base and may also be present in earlier versions:

• 2018-02-02: Embedded IOCs using generalTime with an event system could be affected by a problem reported by Till Strauman, which is fixed by this patch. The fix will also be needed for R3.14.12.6 and later release series.
• 2018-03-09: If a database file creates an info tag on a record that has has an alias, the info tag may actually be created on the alias rather than the original record depending on the order of the two in the file. This problem was discovered and reported by Dirk Zimoch, and is only being fixed on the 3.15 and later branches. The workaround for databases that need both aliases and info tags on the same record is to create the alias/aliases after all the info tags, or to always use the double argument alias(record, new-name) syntax outside of the record definition.

R3.14.12.6 Problems

The following significant problems have been reported in version 3.14.12.6 of EPICS Base and may also be present in earlier versions:

• 2017-07-24: RTEMS IOCs may crash or display random messages if their callback ring buffer fills up, because the callbackRequest() routine passes a stack-allocated message buffer to be printed but the buffer can be reused before it gets displayed. This patch replaces the stack-allocated message with an array of static strings.
• 2017-04-24: IOCs with very large numbers of CA links may crash on shutdown due to a bug in the dbCa code. This patch solves the issue.
• 2017-04-17: macLib can lose errors when more than one macro is used in a string to be expanded. This patch fixes the problem and adds a self-test to prove it.
• 2017-02-27: An access fault in dbConvert on 64-bit systems is fixed with this patch. Thanks to Carsten Winkler, Freddie Akeroyd and Benjamin Franksen for finding and resolving this bug, which only occurs on systems where int and long are different sizes.

R3.14.12.5 Problems

The following significant problems have been reported in version 3.14.12.5 of EPICS Base and may also be present in earlier versions:

• 2017-06-13: An IOC crash resulting from an access security CA channel disconnectting can be fixed by applying this patch; this bug was fixed in R3.14.12.6 but has been reported twice now in older versions of Base.
• 2016-09-29: A race-condition reported in the pcaspy module turned out to be a bug in gdd, which is fixed by this patch. Thanks to Xiaoqiang Wang for finding and providing a fix for this bug, which does not affect the IOC at all so the patch only needs to be applied for installations that build CAS server tools such as the CA Gateway or pcaspy.
• 2016-08-11: This patch fixes a race condition in db_get_field_and_count() that was found back in May.
• 2016-03-30: Initializing a string record with a string longer than 40 characters can cause the IOC to crash at iocInit (see Launchpad bug 1563191). This patch prevents both bugs.
• 2016-03-30: The C++ version of the epicsThread API can corrupt memory in some circumstances when the thread exits (see Launchpad bug 1558206). This patch fixes the problem.
• 2015-11-09: Another patch for Microsoft VS 2015 users, this patch is required to prevent a compiler error when building for these targets.
• 2015-08-20: Users of the Microsoft VS 2015 compiler will need to apply this patch to Base to prevent a compiler error. Apparently only the Community Edition of this compiler works for 64-bit builds, the Express Edition exceeds a compiler limit when compiling Base.
• 2015-06-30: If EPICS Base is built with readline support, any IOC that calls epicsExit() from a thread other than the main thread is likely to leave the user's terminal in a weird state, requiring the user to run something like 'stty sane' to clean it up. This patch fixes the readline support code to clean up automatically by registering an epicsAtExit() routine.

• 2015-04-29: The latest version of Cygwin (using gcc 4.9.2) has moved the declaration of the select() system call to a different header file. This patch adjusts the Base headers so Base builds properly with the latest Cygwin.
• 2015-04-09: The dbContextReadNotifyCacheAllocator clears its pool of free buffers on an array size change, but doesn't consider buffers currently in use, which will later be free'd. Such buffers were being returned to the pool and later reused in allocations for which they were too short. This patch tracks the size of buffers which are in use and only returns buffers which are the correct length to the pool. Smaller buffers are free'd immediately.

R3.14.12.4 Problems

The following significant problems have been reported in version 3.14.12.4 of EPICS Base and may also be present in earlier versions:

• 2015-03-20: The ao record type does not check converted raw values for overflow of a 32-bit integer before writing them to the RVAL field. On some architectures the result may be unexpected (caused by Undefined Behaviour). This patch corrects that by limiting the raw value to the 32-bit signed integer range. The ai and ao ROFF fields are now unsigned values, to allow a raw offset of 0x80000000.
• 2015-03-05: Sites that do certain kinds of cross-builds may have problems with the Cap5.so shared library being built and installed where it shouldn't have been. This patch should fix that.
• 2015-02-10: The CA Repeater may sometimes print bogus error messages, at least on Windows. This patch clears up those messages.
• 2015-02-05: The caget/caput/camonitor programs do not convert the value zero correctly in binary mode (-lb or -0b depending on channel data type). This patch fixes this problem.
• 2015-01-07: Some linux distributions apply strict compiler flags which cause the Cap5.c compilation to fail. Apply this patch to clean up the source file and remove those warnings.
• 2014-12-23: This patch fixes an error message from iocInit about forward-links over CA, thanks to Dirk Zimoch.
• 2014-12-02: If your build fails to find the xsubpp program (part of the Perl distribution) try applying this patch which is needed for recent versions of Perl such as the one included with Scientific Linux 7, RHEL 7 and CentOS 7.0.
• 2014-10-29: Windows builds need this patch to correctly create dependency files for C and C++ compilations. This will also be needed for other targets whose C/C++ compilers cannot generate dependency files so the build uses mkmf.pl to create them.
• 2014-10-08: IOC applications running under Cygwin need a file equivalent to the dllPath.bat that supports the Bash syntax. After applying this patch to Base the build system supports adding the target relPaths.sh to the iocBoot/ioc/Makefile which fulfils the above requirement. The relPaths.sh file must be sourced, not executed, to be able to perform the necessary modications to the environment.
• 2014-09-17: The previous patch from 2014-04-07 would cause a complete rebuild whenever you run make, and also had problems in gdd. The original patch file has been modified to correct these problems.
• 2014-04-07: The build rules in Base were not explicitly requiring that all include files have been installed before the compilation of C or C++ source files starts, thus on highly parallel machines some applications may not build properly with 'make -j'. This patch adds the necessary requirement to the rules.
• 2014-02-25: Xiaoqiang Wang proposed some fixes that allow static builds on the MinGW targets to succeed. This patch makes those changes and fixes another issue in blockingSockTest.
• 2014-02-17: This patch contains build configuration fixes for Apple iOS 7.0 and XCode 5.0 from Tom Pelaia.
• 2014-01-29: Matt Pearson pointed out a bug in epicsTime::strftime() which rounds its nanoseconds output without incrementing the whole seconds. It is not actually feasible to fix this and round properly in all cases, but this patch prevents the nanosecond value from getting larger than all 9s, which should be sufficient for most purposes.
• 2013-12-04: This version of Base added support for Apple's Xcode 5.0 which can build software for iOS 7.0 and the ARMv8 64-bit CPU on the newest iPhone 5S device. Unfortunately the Xcode upgrade broke the build of the ios-x86 iPhone simulator target, although the ios-arm target successfully built code which runs fine on the real hardware. See above (2014-02-17) for the fix.

R3.14.12.3 Problems

The following significant problems have been reported in version 3.14.12.3 of EPICS Base and may also be present in earlier versions:

• 2013-12-04: If an IOC loads a record instance of a record type that was not defined in its DBD files, the dbLoadRecords command may segfault before displaying the error message that explains the problem. This patch prevents that crash.
• 2013-09-13: Reading back a .TSEL field that is pointed to another record's .TIME field was exposing internal details (that the link target field gets changed internally to point the .VAL field) which don't need to be made visible. This patch hides the change, and should allow a .TSEL field to be saved and restored properly using autosave.
• 2013-07-10: Building base against VxWorks 6.9 fails for several reasons. Apply this patch to resolve a name conflict, this patch to configure the build rules properly for vxWorks 6.9, and this patch for a change in the gmtime_r() and localtime_r() APIs. Other changes may also be needed for the IOC to run properly.
• 2013-06-07: Hostnames that begin with numbers can cause problems on some systems. This patch modifies the libCom routine aToIPAddr() to fix this, and adds a new test program to check its functionality.
• 2013-05-06: The Windows implementation of the epicsThread API leaks a thread handle for each epicsThread that is closed, which could cause a resource shortage problem on long-running IOCs, CA client programs or other Windows programs using the epicsThread API. This patch from Giles Knap fixes the problem.
• 2013-04-24: The newer compiler in VxWorks 6.8 is clever enough to not link the object file from libCom/osi/os/vxWorks/atReboot.app into generated executable images, although this object is actually wanted. This patch changes how the sysAtReboot initialization takes place and ensures that the necessary steps will be taken properly at IOC startup to register the shutdown function.
• 2013-01-23: If you need to use a version of Perl which is not found at /usr/bin/perl some of the scripts that come with Base will run the wrong perl. This patch changes all of the Perl preamble lines to #!/usr/bin/env perl which should be correct on all modern Unix-like systems. This line is not used on Windows systems.
• 2013-01-09: A fix for bug 597054 is available for sites that want their period scan tasks to not drift on workstaion OSs. Apply this patch which prints warning messages when repeated over-run errors occur (patch modified on 2013-10-23).
• 2012-12-18: A race condition can occur when reloading the access security file from a subroutine record that is processed using caput rather than a longer-running CA client such as MEDM or caput -c. This patch fixes the problem.

R3.14.12.2 Problems

The following significant problems have been reported in version 3.14.12.2 of EPICS Base and may also be present in earlier versions:

• 2011-12-12: Sites using some older versions of the SunPRO C compiler may need to apply this patch to get it to build this version of Base.
• 2011-12-22: Databases built from a substitutions file that depend on another database built inside O.Common will fail without this patch.
• 2012-01-04: If you have problems on MacOS-X compiling Base in the src/cap5 directory related to it trying to compile for the PowerPC architecture, you're actually hitting a problem with Apple's installation of Perl; they didn't update the Perl configuration when they removed the PowerPC compiler support, which makes it impossible to compile Perl modules from C code without adding a workaround. If you need to use Perl scripts that use the Channel Access library you can edit the src/cap5/Makefile and change the Cap5_CFLAGS definition to be:

  Cap5_CFLAGS = -g -pipe -fno-common -DPERL_DARWIN -fno-strict-aliasing
      

• 2012-01-11: CA clients using long string support may do a get of a DBF_CTRL_CHAR with a 1-element array, but the IOC was rejecting 1-element gets for long strings since the value should always be a single zero. In this case though the clients are only interested in the attribute data so the get should be allowed. This patch permits it, and also ensures that sub-strings fetched from DBF_STRING fields are properly zero-terminated.
• 2012-01-26: A bug has just come to light that was introduced into the dbpf IOC command during refactoring between Base R3.14.9 and R3.14.10. On most architectures it is invisible, but the Windows C runtime library appears to have a problem with sscanf() format strings that end with the conversion specification sequence " %n", i.e. that have a space before a request to return the number of characters consumed. With the space present the %n request gets ignored, which prevents IOCs running on Windows from converting an integer value encoded in a DBF_STRING into a field of type DBF_ENUM. This patch fixes the bug in dbpf, and removes the spaces from the conversion spec.
• 2012-03-19: The fix-907761 patch posted on 2012-01-11 above has been replaced with an updated version.
• 2012-04-30: If you use rules in your Makefiles that look like the example below, you should apply this patch to prevent problems when you use the -s flag to gnumake, or make sure that you don't have any commands on the same line as $(ECHO).

  $(ECHO) Creating foo; ../create-foo.sh > foo

• 2012-05-14: Windows users can't call the function casStatsFetch() because the epicsShareFunc marker is missing from src/rsrv/rsrv.h. This patch fixes that.
• 2012-05-16: RTEMS CExp users may need to apply this patch to be able to call nfsMount(). This change is not needed if you use iocsh to run your startup script.
• 2012-05-21: A buffer overflow bug in dbReportDeviceConfig() can be fixed by applying this patch.
• 2012-09-04: An off-by-one string overflow bug in caput -S can be fixed by applying this patch.
• 2012-10-10: The monitor deadband fields of several record types can get stuck if the ALST or MLST fields become NAN. This patch resolves that problem.
• 2012-12-05: On Windows systems, do not use WinZIP to extract the source tree from the Base tar.gz file if you are planning to build it for the cygwin-x86 host architecture. Doing so causes the build to fail when it reaches the src/bpt directory. This might not happen if your Cygwin installation is configured to use DOS line-endings, but we haven't tried that to see. Use the tar program provided with Cygwin to extract the source tree instead.
• 2012-12-13: Cygwin users must apply this patch to be able to build Base when SHARED_LIBRARIES=YES. Without it you can expect to see this error when the build first runs antelope.exe: RULES_BUILD:230: recipe for target `parse.c' failed

R3.14.12.1 Problems

The following significant problems were reported in version 3.14.12.1 of EPICS Base and subsequently fixed in version 3.14.12.2:

• 2011-05-23: Cygwin builds on drives other than C: can fail when gnumake first tries to link the antelope.exe program, with an error message like ld: cannot find -lCom. This patch fixes that problem, which was not present in R3.14.12.
• 2011-06-27: On MacOS X machines when running a 64-bit kernel the two versions of the EpicsHostArch script both return the wrong result. This patch modifies those scripts to give the correct output.
• 2011-07-07: Some combination of Windows 7 and MinGW Make try to compile C files using the C++ compiler, which doesn't work. Apply this patch to delete the old build rule that causes this problem.
• 2011-07-21: VxWorks systems that do not use a PowerPC CPU may see the error message initHandlerAddrList() unsolicitedHandlerEPICS not in sym table. Apply this patch to resolve this problem.
• 2011-07-26: Some versions of GCC for specific processors don't like certain legal C++ constructs which are rarely used and result in an udpiiu.cpp:1391: error: generic thunk code fails for method ... message. Apply this patch to resolve this problem, which is bug 816678.
• 2011-10-20: Multi-threaded CA clients (such as recent versions of the sequener) that run inside the IOC process may miss monitors from local channels without this patch, which fixes bug 878372 and bug 878387.
• 2011-11-01: The caget and camonitor programs have not properly handled array PVs since R3.14.12, especially when using the -#n option with multiple PVs at the same time. The camonitor code could also crash if an array PV disconnects and later reconnects with a larger size array. This patch resolves these problems and fixes bug 794749.
• 2011-11-08: Building 3.14.12.1 using gcc version 4.6 or later requires this patch which is a combination of the two patches attached to bug 736273 which was fixed back in August.
• 2011-11-09: There is a bug in the makeBaseApp.pl script when using the cygwin-x86 architecture; it outputs windows-style paths in the generated configure/RELEASE file, i.e. C:/EPICS/base instead of /cygdrive/c/EPICS/base. This patch fixes the problem.

R3.14.12 Problems

The following significant problems were reported in version 3.14.12 of EPICS Base and subsequently fixed in version 3.14.12.1:

• 2010-12-14: When building IOCs with a Win32 target architecture, the generated xxx_registerRecordDeviceDriver.cpp file can cause a number of warnings to be reported because we don't properly escape the back-slash path-separator characters in the bldTop string.
  2010-12-16: Stephen Norum reported the above and some related issues in the convertRelease.pl and fullPathName.pl scripts. This patch can be applied to fix these problems.
• 2011-01-07: If errlogPrintf() is called during shutdown, it can cause a recursive assert failure. This patch from Michael Davidsaver fixes the issue. This is bug 693534.
• 2011-01-13: The native element count for array fields is truncated by the IOC server to 16 bits, so (unless you use the dynamic array sizing features that were new in this release) it is not possible to transfer more than 65534 elements of an array from an IOC without applying this patch and rebuilding Base and the IOC. Note that the CA client library is not affected, just the rsrv server code. This is bug 701673.
• 2011-02-16: Problems with an IOC running on Windows when communicating with local CA clients might be solved by applying this patch, which might also solve some issues with non-IOC CA client programs. This is bug 717252.
• 2011-03-01: Systems that generate a lot of errlog output can trigger a bug in msgbufGetNode() that overwrites older messages in the output buffer. This patch fixes that problem and modifies the errlog test program to match.