These release notes describe changes that have been made since the previous release of this series of EPICS Base. Note that changes which were merged up from commits to the 3.15 branch are described further down this file under the 3.15 release to which they were originally committed. Thus it is important to read more than just the first section to understand everything that has changed in each release.
The PVA submodules each have their own individual sets of release notes which should also be read to understand what has changed since earlier releases:
This version of EPICS has not been released yet.
Add support for CA tools timeout from environment variable $EPICS_CLI_TIMEOUT
which sets the default the default timeout for caget et al.
The -w argument continues to take precedence.
On Windows targets, if a thread not created by epicsThreadCreate*() directly
or indirectly calls an epicsThread*() function, a specific tracking struct
is allocated. Prior to this release the allocation would not be free()d,
resulting in a memory leak.
A similar issue on POSIX targets was previously fixed.
The default compiler for FreeBSD targets changes from GCC to clang.
Add a new IOC shell command dbCreateAlias allow record aliases to be added.
Intended for use before iocInit. eg. to add an alias "bar" for a record "foo".
dbLoadRecords("some.db") # includes: record(ai, "foo") { ...
dbCreateAlias("foo", "bar")
iocInit()
In some cases, RSRV may queue a subscription update, but not flush it. This partially addresses this issue.
Previously, if a subRecord has an invalid INP* link, it was silently failing
(and not running the proc function). Now the the status code returned by the
subroutine is returned from dbProcess().
Fall back to the previous behavior when searching for readline.h with older compilers.
Look for /opt/local/include/readline on OSX.
The SCAN and callback threads are now stopped during normal IOC shutdown.
The environment variable $EPICS_FREELIST_BYPASS may be set to YES to cause the freeListLib functions to always call directly to malloc()/free(). May be useful when troubleshooting some kinds of memory allocation bugs which would otherwise be "hidden". eg. use-after-free data races. This may also improve the results of dynamic analysis tools which are not aware of this internal free list.
The compress record now supports the use of partially-filled buffers when using
any of the N-to-one algorithms. This is achieved by setting the new field PBUF
to YES.
The "ts" filter can now retrieve the record's timestamp in several numeric
and string formats, some of which support full nanosecond precision.
Hal$ caget -a test:channel
test:channel 2021-03-11 18:23:48.265386 42
Hal$ caget -f9 'test:channel.{"ts": {"num": "dbl"}}'
test:channel.{"ts": {"num": "dbl"}} 984331428.265386105
Hal$ caget 'test:channel.{"ts": {"str": "iso"}}'
test:channel.{"ts": {"str": "iso"}} 2021-03-11T18:23:48.265386+0100
Hal$ caget -f1 'test:channel.{"ts": {"num": "ts"}}'
test:channel.{"ts": {"num": "ts"}} 2 984331428.0 265386163.0
More information is included in the filters documentation, which can be found in
the html/filters.html document that is generated during the build
The longout record can now be configured using its new OOPT and OOCH fields to (not) write to its output link depending on the contents of VAL, in a similar manner to the calcout record. More information can be found on the reference page for the longout record type that accompanies this release.
This functionality was suggested in lp# 1398215 and may be added to other output record types if the community finds it useful, please send feedback about the feature to tech-talk.
When built with optional GNU libreadline support, the interactive IOC shell
will perform tab completion for command names as well as for some arguments
of the built-in commands. For example, the record name argument of dbpr,
and the path name argument of cd.
Externally defined commands have a limited ability to opt into completion by
using the new iocshArgStringRecord and iocshArgStringPath argument types.
Both function identically to iocshArgString but indicate how to suggest
completion strings.
Builds on macOS (darwin-x86 or darwin-aarch64 targets) normally use Apple's libedit library in readline compatibility mode, which doesn't support the tab completion API that GNU readline provides. You can use Homebrew or some other third-party package manager to install the GNU readline package, then edit the configure/os/CONFIG_SITE.darwinCommon.darwinCommon file to have EPICS use the real thing to get tab completion in the IOC shell. The default settings in that file currently look for and use a Homebrew-installed readline if present.
The floating point modulo function FMOD(NUM,DEN) has been added to the CALC
expression engine and is available to all software using that (calc and calcout
record types, access security library and some extensions).
Thanks to several attendees at the 2022 EPICS Codeathon the number of header files with Doxygen annotations in the EPICS Core has again increased.
The top-level make targets uninstall, archuninstall and similar no
longer trigger the clean target which empties build directories, this
was a bug introduced in 7.0.5.
The make distclean target now properly deletes the generated file(s)
modules/RELEASE.<host>.local which are essential to build the external
submodules under the modules directory, and should not crash if the
build is configured with INSTALL_LOCATION pointing to an empty external directory (i.e. if you run make distclean twice in succession). When
INSTALL_LOCATION is set in the files configure/CONFIG_SITE or
configure/CONFIG_SITE.local the modules/RELEASE.<host>.local file
will now be regenerated in case the install path has been modified.
Note that passing INSTALL_LOCATION=<path> on the make command-line will
only work if you have run make distclean immediately beforehand, as the
modules/RELEASE.<host>.local file must be recreated using the new path.
The capr.pl script can now display records from older Base versions to
which fields have since been added, and shows long strings and array data
up to 10 elements, use the new -n option to increase that number.
The script is fully event-driven and prints all the field data received by
the end of the CA wait time (-w option which defaults to 2 seconds).
The interest level can now be specified using the -l option before the
PV name, and the new -D flag outputs debugging information.
VxWorks 6.9 can do its own OS clock time synchronization, if it has been
configured by setting SNTPC_PRIMARY_IPV4_ADDR. Since EPICS 3.15.3 the
IOC time support code has checked for the existence of the VxWorks time
synchronization task and avoided starting the EPICS one if the OS task
exists and the OS clock gives a "recent" time (i.e. after when EPICS was
compiled), unless the environment variable EPICS_TS_FORCE_NTPTIME is
also set. However a logic error in that code required the environment
variable to be set in more cases than it should have.
This error has been fixed and the IOC should work normally if the VxWorks
task is configured and running. The TIMEZONE value for the year is also
now calculated at initialization in this configuration, previously it was
only done when the IOC synchronzation task was used. Setting the above
environment variable will now cause the IOC support code to shut down the
VxWorks synchronization thread (if running) before starting the EPICS one.
Running the iocsh command ClockTime_Report now shows whether the VxWorks
task is running as well as giving the state of the IOC synchronization task.
The ClockTime_Init command can also be used to stop or restart the IOC
time synchronization task while the IOC is running, depending on the 0 or
1 parameter passed to it. This last change also applies to RTEMS IOCs.
This change may cause channel filters which manipulate array updates to fail to compile.
To avoid potential speculation issues arising from overlapping code pointers
with data values, union dbfl_ref is modified to remove the dtor member.
dtor is moved out into the enclosing struct db_field_log.
So eg. using a db_field_log* p, the expression p->u.r.dtor must be
changed to (p)->dtor.
The order over operations when processing a waveformRecord is adjusted so that updates to NORD is posted with the correct timestamp.
When built with a compiler supporting __has_include<>, the presence
of the <readline/readline.h> will be used to automatically determine
a default value for COMMANDLINE_LIBRARY.
Mingw builds with readline support now link -ltermcap instead of -lcurses.
This should not effect sites which set explicitly set COMMANDLINE_LIBRARY
as the only definition in Base now has the form COMMANDLINE_LIBRARY ?= ....
The Perl CA bindings have been fixed to handle zero-length long string data properly.
The aao record types gains the same DOL functionality found
in other output record types (ao, longout, etc.)
During iocInit(), the environment variable RSRV_SERVER_PORT is set
with the TCP port number selected.
DBD files generated by the dbdExpand.pl script are now sorted within each
item type by the primary name of the item. The result should resolve any
issues with reproducable builds. No option is provided to prevent the sorting,
previously the order was essentially random and varied each time.
Records are now output by this program in order, sorted by name. The new flag
-s can be given to output the records in the same order they were read in,
instead of sorting them.
Note that there are currently no build rules provided with Base which make use of this program.
SIMM=RAW support has been added for the relevant output record types (ao, bo, mbbo, mbboDirect). RAW simulation mode will have those records do the appropriate conversion and write RVAL to the location pointed to by SIOL.
On some targets, if a thread not created by epicsThreadCreate*() directly
or indirectly calls an epicsThread*() function, a specific tracking struct
is allocated.
Prior to this release, on POSIX and WIN32 targets, this
struct would not be free()d, resulting in a memory leak.
This release fixed the leak on POSIX targets.
See the associated github issue 241 for WIN32 status.
On some targets, if a thread not created by epicsThreadCreate*() directly
or indirectly calls an epicsThread*() function, a specific tracking struct
is allocated.
Prior to this release, on POSIX and WIN32 targets, this
allocation would not be free()d, resulting in a memory leak.
This release fixed the leak on POSIX and WIN32 targets (excluding MSVC before vs2012, and the WINE runtime).
On some targets, if a thread not created by epicsThreadCreate*() directly
or indirectly calls an epicsThread*() function, a specific tracking struct
is allocated.
Prior to this release, on POSIX and WIN32 targets, this
struct would not be free()d, resulting in a memory leak.
This release fixed the leak on POSIX targets.
See the associated github issue 241 for WIN32 status.
This now works again, it was broken in 2019 (7.0.3.1) by an errant commit.
The DISP field can be set to a non-zero value to prevent records being changed from outside the IOC (this is ancient behavior), but has never been documented as being usable at design-time (DCT=Yes in the Record Reference tables). This has now been changed.
The epicsInt8 and thus DBF_CHAR types have always been unsigned on
architectures where char is unsigned, for example on many PowerPC CPU
architectures. This was counter-intuitive, and resulted in IOC behavior
differing between architectures when converting DBF_CHAR values into a
signed integer or floating point type.
WARNING: This fix may change behavior of existing databases on target
architectures with unsigned char (mainly PowerPC) when using input links to
read from CHAR arrays. Architectures with signed char (usually x86) should
be unaffected, although some compilers might generate new warnings.
Several types of hardware links (VME_IO, CAMAC_IO, etc) now accept
hexadecimal and octal numbers. (Hexadecimal numbers had already been valid
up to EPICS R3.15.) This change may introduce incompatibilities when using
numbers with leading 0 as they will now be parsed as octal.
Heinz Junkes provided a new implementation of the epicsEvent API suitable for
RTEMS Posix targets (RTEMS 5.1 and later). In review a few issues related to
overflow of timeout values surfaced in this and other embedded implementations,
and these were also been fixed in this Pull Request. The API documentation for
this and some other routines has also been updated.
The names of breakpoint tables were made unnecessarily strict when DBD file
processing was moved to Perl for the 3.15 release series. Table names may now
contain the special characters _ - : ; . [ ] < > in addition
to letters and digits.
Prevents Use of uninitialized value warnings from convertRelease.pl.
Many internal error messages now emit ANSI escape sequences to highlight the words "ERROR" and "WARNING" in an attempt to make occurrences more noticeable during IOC startup.
The macros ERL_ERROR and ERL_WARNING are defined for external usage,
and expand as string constants. eg.
#include <errlog.h>
#ifndef ERL_ERROR
# define ERL_ERROR "ERROR"
#endif
void fn() {
...
errlogPrintf(ERL_ERROR ": something bad happens :(\n");ANSI escapes are automatically removed from errlog output not destined
for a terminal. For example, for logClient, if stderr is redirected,
or if unsupported ($TERM not set, or Windows < 10).
The dbnd server side filter now passes through alarm and property
change events, even when not exceeding the deadband.
The bit fields B0 - B1F of this record are now always updated and have a
monitor posted when the VAL field is set and the record processed. It is now
possible to initialize the record's value by setting the bit fields inside a
database file as long as no other method was used to initialize it (suc as
setting VAL directly, using DOL, or by an initial readback from device
support). A new internal field OBIT was added to store information about
monitors posted on the bit fields.
Some scripts now make use of features that were introduced to this Perl version that was released in 2009.
GH:183 These were broken in a previous release, but now work again.
GH:194 This was broken in a previous release, but now works again.
- Many code comments have been spell-checked and corrected.
- Passing a
-DDEBUGcompiler flag no longer breaks the build. - Parallel builds of RTEMS-mvme2100 and RTEMS-mvme2700 targets now work.
- Illegal characters seen in JSON strings in a database file should now get a better error message.
- lp:1938459 GH:191 int64in only checks lower 32 bits for change
- lp:1941875 Buggy warning message "Record/Alias name '...' should not contain non-printable ...
- GH:187 waveformRecord missing PACT=true?
- GH:189 Fix a couple memory leaks and a segfault
- GH:200 and GH:201 Fix timers on MS Windows for non-EPICS threads
GH:192 Both GCC and CLANG compiler intrisics used for the epicsAtomic APIs have been revised; implementations using CLANG should now run faster as they now use the compiler's built-in atomic functions instead of taking a mutex.
GH:185
This was done to simplify the code and may have improved performance slightly for some uses. Support for the old NTP-specific struct l_fp has been dropped but all other routines and methods of the class epicsTime function as before.
Many of the built-in record types have had improvements to their documentation with additional fields added to the tables, rewrites of descriptions and links to other documents added or fixed.
These target architectures have been removed:
- darwin-ppc, darwin-ppcx86
- linux-386, linux-486, linux-586, linux-686, linux-athlon (cross-build)
- linux-cris, linux-cris_v10, linux-cris_v32 (cross-build)
- RTEMS-at91rm9200ek, RTEMS-gen68360, RTEMS-mcp750, RTEMS-mvme167, RTEMS-psim (cross-build)
The new major release of the RTEMS real-time OS contains many changes including the ability to support SMP systems. This release of EPICS can still be built with RTEMS 4.9.x or 4.10.x and should work just the same as earlier releases, although due to code having moved around we recommend thorough testing before this release is first used in production systems.
This release of EPICS comes with support for several new RTEMS targets running on RTEMS 5:
- RTEMS-beagleboneblack
- RTEMS-pc686
- RTEMS-qoriq_e500 (MVME2500)
- RTEMS-xilinx_zynq_a9_qemu
- RTEMS-xilinx_zynq_zedboard
The EPICS support for RTEMS 4 has always relied on RTEMS-specific
kernel APIs which cannot be used on an SMP system, so a new port was
created to use the Posix real-time APIs that are now recommended for
RTEMS 5. Note that a single installation of EPICS cannot build both
RTEMS 4 and RTEMS 5 targets, if you need to support targets running
on both versions you must use a separate installation, and be sure
to run make distclean if switching a single source tree from one
to the other (both header files and dependency files are different
between the two and must be cleaned out).
The configuration variable RTEMS_VERSION in the EPICS config file
configure/os/CONFIG_SITE.Common.RTEMS must be set to the full 3-
part version number for RTEMS 4 releases, e.g. 4.9.1, 4.10.2
but for RTEMS 5.1 and later it must only contain the major version
number e.g. 5.
Some RTEMS BSPs can be built and may work with the newer libbsd network stack which RTEMS is moving over to, but most of the MVME boards (and the uC5282) still require the legacy network stack.
The dependency on bspExt has been removed, EPICS now provides its own routine for VMEbus probing (or uses one built into the BSP).
Anyone using this release on RTEMS is advised to discuss problems building or running it on either the tech-talk or core-talk email lists so the core developers can help with and find out about any problems with the old or new port.
Known Issues:
- MVME2100 and MVME2700 need changes to the RTEMS 5 BSP to build.
- VMEBus support is not yet available for the MVME2500 BSP.
- There are some known issues with floating point on MVME2500, probably related to its newer e500 FPU.
- Changed network driver for beatnik to work with libbsd. Some issues with DHCP, but network stack usable. Can load env from NVRAM.
The optional argument to epicsEnvShow can now be a glob pattern.
The function epicsStrnGlobMatch(char* str, size_t len, char* pattern)
works exactly the same as epicsStrGlobMatch() but takes an additional
length arguments which limits the number of characters of str to match.
A process using libca which does not find an existing caRepeater process
will attempt to start one by running the caRepeater executable.
This is not always possible, usually when caRepeater is not in $PATH.
Now, instead of printing a warning, an internal caRepeater thread
will be started (as is done be RTEMS and vxWorks targets).
If this fallback occurs, the lifetime of the caRepeater thread may be shorter than the lifetime of a separate caRepeater process would have been.
It remains the recommended practice to explicitly start a caRepeater
instance. Examples of both systemd (caRepeater.service) and sysv
(S99caRepeater) scripts may be found under bin/.
When used with one argument, the var command can be used with a glob pattern
for printing matching variables.
The FINAL_LOCATION make variable has for some time been an undocumented
means of performing a staged build. This is a build which "installs" to
a temporary location, which will later be moved to a final location.
This has now been added to configure/CONFIG_SITE.
Usage analogous to the autotools recipe
./configure --prefix=/usr/lib/epics
make install DESTDIR=/tmp/buildwould be
make INSTALL_LOCATION=/tmp/build FINAL_LOCATION=/usr/lib/epicsFINAL_LOCATION is now correctly used in systemd and sysv init scripts
caRepeater.service, S99caRepeater, and S99logServer.
IOC shell will now ensure ${PWD} is set on startup,
and updated by the cd iocsh function.
Two new fields have been added to dbCommon so will be present in all
records: AMSG and UTAG.
AMSG can hold an arbitrary 40-character string, providing additional
information about the alarm condition indicated in STAT and SEVR. With no
alarm it will hold an empty string. The new recGblSetSevrMsg() function can
be used in place of recGblSetSevr() to signal an alarm while providing a
message.
For example, a device support's read_bi() routine for a hypothetical
multi-channel ethernet attached device might flag a communication error
between the IOC and controller, or an error involving a certain channel like
this:
static long read_bi(biRecord* prec) {
...
if (!priv->connected) {
recGblSetSevrMsg(prec, COMM_ALARM, INVALID_ALARM,
"No controller connected");
return S_dev_noDevice;
}
if (!priv->err) {
recGblSetSevrMsg(prec, READ_ALARM, INVALID_ALARM,
"Channel %u disconnexted", priv->chan);
return S_dev_noDevice;
}
return status;
}UTAG holds an epicsUInt64 value which is semantically part of the record's
timestamp (TIME). The value defaults to zero if not explicitly set. Device
support or an event time provider which supports this feature may write a tag
value directly to the dbCommon::utag field.
TSEL links will copy both TIME and UTAG between records if the link type
supports this (CA links do not).
A utag server side channel filter has been added which can be configured to
filter out monitor updates which don't pass the test (UTAG & M) == V where
M and V are client specified integers. For example running the command
camonitor BPM0:X.{utag:{M:1,V:1}} will only show updates for which
(UTAG & 1) == 1 i.e. the least significant bit of the UTAG field is set.
This feature is intended for use by intelligent devices which can provide contextual information along with a value/alarm/time. For example, a beam diagnostic device which is aware of whether a beam signal should be present (eg. from a global timing system).
Two new optional methods have been added to the Link Support Entry Table
(struct lset): lset::getAlarmMsg() and lset::getTimeStampTag(). See
comments in dbLink.h for details on implementing these.
Two new accessor functions have also been added which call these methods:
dbGetAlarmMsg() and dbGetTimeStampTag().
User code wishing to call these interfaces while maintaining compatibility with older versions of Base may add some of the following macro definitions, and ensure that the variables referenced by output pointers are initialized.
#ifndef HAS_ALARM_MESSAGE
# define recGblSetSevrMsg(REC, STAT, SEVR, ...) recGblSetSevr(REC, STAT, SEVR)
#endif
#ifndef dbGetAlarmMsg
# define dbGetAlarmMsg(LINK, STAT, SEVR, BUF, BUFLEN) dbGetAlarm(LINK, STAT, SEVR)
#endif
#ifndef dbGetTimeStampTag
# define dbGetTimeStampTag(LINK, STAMP, TAG) dbGetTimeStamp(LINK, STAMP)
#endifThe unit test programs that are run by the make runtests or make tapfiles
commands get executed by a .t wrapper script which is normally generated by
the EPICS makeTestfile.pl program. Those generated wrapper scripts now
impose a time-limit on the test program they execute, and will kill it if it
runs for longer than 500 seconds (8 minutes 20) without exiting. That
time-limit can be changed for any such test by modifying the Makefile which
creates and runs the .t wrapper script.
Setting the environment variable EPICS_UNITTEST_TIMEOUT to the desired
number of seconds while the Makefile is generating the test script changes the
timeout in that script. For example:
TESTSCRIPTS_HOST += hourLongTest.t
hourLongTest.t: export EPICS_UNITTEST_TIMEOUT=3600
When selecting such a timeout remember that different Continuous Integration systems such as GitHub Actions and Appveyor run on processors with different speeds, so allow enough head-room for slower systems to complete the test.
Test programs written directly in Perl as a .plt script should implement a
similar timeout for themselves. The "netget" test in Base does this in a way
that works on Windows as well as Unix-like hosts.
Krisztian Loki reported segfaults occurring when a Soft Channel aai record INP field was a DB link to an array field of a compress record. This was caused by the aai record's pass-0 device support initialization clashing with the semantics of the new link support API.
The aai record has been modified to allow the Soft Channel device support to request a pass-1 initialization callback. See the Device Support section of the Array Analogue Input Record Reference pages in this release for the API changes, which are fully backwards compatible for existing aai device support.
Kay Kasemir reported that
it is possible to change the Base record type's default DTYP if a device()
entry is seen before the recordtype() definition to which it refers. The
default DTYP is the first device loaded, which is normally the Soft Channel
support from Base. A warning was being displayed by dbdExpand when a device()
entry was see first, but that was easily missed.
The DBD file parser in dbdExpand.pl has now been modified to make this an error,
although the registerRecordDeviceDriver.pl script will still accept device()
entries without having their recordtype() loaded since this is necessary to
compile device supports as loadable modules.
On Posix systems, epicsMutex now support priority inheritance if available.
The IOC needs to run with SCHED_FIFO engaged to use these.
Support for Posix implementations before POSIX.1-2001 (_XOPEN_SOURCE < 500,
glibc version < 2.3.3) has been dropped.
The IOC shell's epicsMutexShowAll command prints "PI is enabled" if both
libc and kernel support is present.
Since 7.0.3.1 a Windows IOC could not run for more than 49.7 days; at that time the periodic scan threads would stop processing. This issue should now have been fixed and the Monotonic time functions on Windows should return values which count at nanosecond resolution. However we have not waited 49.7 days to test the final software, so there is a small chance that it's still broken.
This fixes lauchpad bug #1896295.
Thanks to Jeong Han Lee this release comes with build support for Apple's new
M1/M2 CPUs running macOS, using the target name darwin-aarch64.
It should also be possible to build universal binaries containing code for
both the Intel and arm64 processors under either target name: In the
appropriate configure/os/CONFIG_SITE.Common.darwin-* file add the other
architecture class name to the ARCH_CLASS variable (after a space).
The new epicsStrSimilarity() routine in epicsString.h uses a modified
Levenshtein distance to compare two strings, with a character case difference
being half the weight of a full substitution. The double return value falls in
the range 0.0 (identical) through 1.0 (no characters matching), or -1.0 for
error. This is used to provide a new "Did you mean ..." suggestion when a .db
file provides an invalid choice string for a DBF_MENU or DBF_DEVICE field.
Target architectures that support command-line programs that run the main()
routine can now be marked as such in their VALID_BUILDS definition. This
enables a new set of Makefile target variables PROD_CMD (similar to
PROD_HOST), LIBRARY_CMD (like LIBRARY_HOST, etc.), LOADABLE_LIBRARY_CMD,
OBJS_CMD, SCRIPTS_CMD, TARGETS_CMD, TESTLIBRARY_CMD, TESTSCRIPTS_CMD
and TESTPROD_CMD. The CA client tools and programs such as caRepeater are now built for all such targets (previously they were built for all targets except where the OS was VxWorks, RTEMS and iOS).
If you have created your own site-specific target architectures you may need to
update the VALID_BUILDS variable if it gets set in your locally added
configure/os/CONFIG.Common.<arch> files. This is usually only needed for
cross-compiled targets though since CONFIG.Common.UnixCommon sets it.
The other VALID_BUILDS types are "Host" for target architectures that can
compile and run their own programs (PROD_HOST etc.), and "Ioc" for targets
that can run IOCs (PROD_IOC etc.).
The YAJL parser and generator routines in libcom and in the IOC's dbStatic parser now support the JSON5 standard. This adds various features to JSON without altering the API for the code other than adding a new option to the YAJL parser which can be used to disable JSON5 support if desired. The new features include:
- The ability to handle numeric values
Infinity,-InfinityandNaN. - String values and map keys may be enclosed in single quotes
', inside which the double-quote character"doesn't have to be escaped with a back-slash\, although a single-quote character'(or apostrophy) must be escaped inside a single-quoted string. - Numbers may start with a plus sign,
+. - Integers may be expressed in hexadecimal with a leading
0xor0X. - Floating-point numbers may start or end with their decimal point
.(after the sign or before the exponent respectively if present). - Map keys that match the regex
[A-Za-z_][A-Za-z_0-9]*don't have to be enclosed in quotes at all. The dbStatic parser adds.+-to the characters allowed but will add quotes around such keys before passing them to YAJL. - Arrays and maps allow a comma before the closing bracket/brace character.
- The YAJL parser will elide a backslash followed by a newline characters from a string value. The dbStatic parser doesn't allow that however.
Code that must also compile against the older API can use the new C macro
HAS_JSON5 to detect the new version. This macro is defined on including
either the yajl_parse.h or yajl_gen.h headers, which also provide the
new configuration options to turn on JSON5 support.
All APIs in the IOC that previously accepted JSON will now accept JSON5. This includes JSON field modifiers (channel filters), JSON link addresses, constant input link array values and database info-tag values. JSON values that get parsed by the dbLoadRecords() routine are still more liberal than the other uses as the ability to use unquoted strings that was called "relaxed JSON" is still supported, whereas the JSON5 standard and the YAJL parser only allow unquoted strings to be used for keys in a JSON map.
This also fixes lauchpad bug #1714455.
- The libCom routines
epicsStrnRawFromEscaped()anddbTranslateEscape()declared in epicsString.h no longer accept octal escaped characters such as\123or\41. - The routine
epicsStrnEscapedFromRaw()now generates hex excaped characters for unprintable characters such as\x1f. - Hex escape character sequences
\xXXmust now contain exactly 2 hex digits. - An escape sequence
\0now generates a zero byte in the raw string, but the other digits1-9should not appear after a back-slash.
These changes are to more closely follow the JSON5 standard, which doesn't
support octal character escapes or the \a (Bel, \x07) escape sequence.
Input database links can now use channel filters, it is not necessary to make them CA links for the filters to work.
The Soft Channel device support for ai records now returns failure when fetching the INP link fails.
Several modifications have been made to properly support zero-length array values inside the IOC and over Channel Access. Some of these changes may affect external code that interfaces with the IOC, either directly or over the CA client API so we recommend thorough testing of any external code that handles array fields when upgrading to this release.
Since these changes affect the Channel Access client-side API they will
require rebuilding any CA Gateways against this version or Base to
properly handle zero-length arrays. The caget, caput and camonitor
client programs are known to work with empty arrays as long as they were
built with this or a later version of EPICS.
When called with COUNT=0 this macro no longer returns the number of bytes required for a scalar (1 element) but for an empty array (0 elements). Make sure code that uses this doesn't call it with COUNT=0 when it really means COUNT=1.
Note that the db_access.h header file is included by cadef.h so the change can impact Channel Access client programs that use this macro.
The ca_array_put() and ca_array_put_callback() routines now accept an
element count of zero, and will write a zero-length array to the PV if
possible. No error will be raised if the target is a scalar field though,
and the field's value will not be changed.
The ca_array_get_callback() and ca_create_subscription() routines
still accept a count of zero to mean fetch as many elements as the PV
currently holds.
Client programs should be prepared for the count fields of any
struct event_handler_args or struct exception_handler_args passed to
their callback routines to be zero.
The soft device support for the array records aai, waveform, and subArray as well as the aSub record type now correctly report reading 0 elements when getting an empty array from an input link.
The dbpf command now accepts array values, including empty arrays, when provided as a JSON string. This must be enclosed in quotes so the iocsh argument parser sees the JSON as a single argument:
epics> dbpf wf10:i32 '[1, 2, 3, 4, 5]'
DBF_LONG[5]: 1 = 0x1 2 = 0x2 3 = 0x3 4 = 0x4 5 = 0x5
Record links that get a scalar value from an array that is currently
empty will cause the record that has the link field to be set to an
INVALID/LINK alarm status.
The record code must call dbGetLink() with pnRequest=NULL for it to
be recognized as a request for a scalar value though.
This changes the semantics of passing pnRequest=NULL to dbGetLink(),
which now behaves differently than passing it a pointer to a long integer
containing the value 1, which was previously equivalent.
The latter can successfully fetch a zero-element array without triggering
a LINK alarm.
Record links that put a zero-element array into a scalar field will now set
the target record to INVALID/LINK alarm without changing the field's value.
Previously the field was set to 0 in this case (with no alarm).
The target field must be marked as special(SPC_DBADDR) to be recognized
as an array field, and its record support must define a put_array_info()
routine.
The record processing code for records with output links has been modified to update the timestamp via recGblGetTimeStamp() before processing the output links. This ensures that other records which get processed via an output link can use TSEL links to fetch the timestamp corresponding to the data processed by the output link.
This change could result in a slightly earlier timestamp for records whose output link is handled by a device driver, but only if the device driver does not handle its own timestamping via TSE -2 and instead uses TSE 0 or TSE -1 to get current time or best time, and the time spent in the device driver is greater than your timestamp provider resolution. For these situations it is recommended to set TSE to -2 and set the timestamp in the driver code.
A new iocsh command registerAllRecordDeviceDrivers is provided and also
defined as a function in iocshRegisterCommon.h. This uses dynamic symbol
lookup with epicsFindSymbol() to perform the same function as a generated
*_registerRecordDeviceDriver() function. This allows for an alternative
approach to dynamic loading of support modules without code generation.
This feature is not intended for use by IOCs constructed using the standard EPICS application build process and booted from a startup script in an iocBoot subdirectory, although it might work in some of those cases — the generated registerRecordDeviceDriver.cpp file is normally required to link everything referred to in the DBD file into the IOC's executable. It also won't work with some static build configurations, or if the symbol table has been stripped from the executable.
It is now possible to use a JSON Const link with a string value to initialize
an aai or waveform record that has FTVL set to CHAR through the INP link.
The string length is not limited to 40 characters. This should also work for
aSub record inputs similarly configured as long strings.
record(waveform, "wf") {
field(NELM, 100)
field(FTVL, CHAR)
field(INP, {const:"This is a waveform and more than 40 characters"})
}
GNUmake added the directive undefine in version 3.82 to allow variables to
be undefined. Support for this has been added to the EPICS Release file parser,
so undefine can now be used in configure/RELEASE files to unset variables.
Build configuration files for a new cross-build architecture linux-aarch64
have been added, and the targets linux-arm_el and linux-arm_eb removed.
The 64-bit ARM architecture target doesn't have build files for self-hosting
yet but they should be relatively easy to add, contributions welcome!
The following bugs/issues have fixes included in this release:
- lp: 1884339, Inaccessible CA servers on Windows
- github: 83 osdTimeGetCurrent doesn't work for subprocess on macOS
- Recent Cygwin build problem with a missing
TCP_NODELAYdeclaration.
Builds of the Perl CA bindings weren't working properly when the Perl installation was from Conda. This release also fixed the capr.pl script to handle the INT64 data types, and to be able to properly handle missing fields, as happens if the IOC is running an older EPICS version for example.
The implementation of the epicsMessageQueue used on RTEMS has switched from
the native RTEMS-specific one to the EPICS generic version, avoiding a bug
in the RTEMS Kernel message queue code.
Historically, there have been very few restrictions on which characters may be present in record and alias names. Base 3.14.12.3 added a warning for names containing space, single or double quote, period/dot, or dollar sign.
Bad character ' ' in record name "bad practice"
7.0.4.1 Turns this warning into an error, and adds a new warning if a record name begins with a minus, plus, left square bracket, or left curly bracket.
The following launchpad bugs have fixes included in this release:
- lp: 1812084, Build failure on RTEMS 4.10.2
- lp: 1829919, IOC segfaults when calling dbLoadRecords after iocInit
- lp: 1838792, epicsCalc bit-wise operators on aarch64
- lp: 1853148, mingw compiler problem with printf/scanf formats
- lp: 1852653, USE_TYPED_DSET incompatible with C++
- lp: 1862328, Race condition on IOC start leaves rsrv unresponsive
- lp: 1866651, thread joinable race
- lp: 1868486, epicsMessageQueue lost messages
- lp: 1868680, Access Security file reload (asInit) fails
Internally, the Com and ca libraries now express dllimport/export (Windows)
and symbol visibility (GCC) using library-specific macros (eg. LIBCOM_API)
instead of the macros epicsShareFunc, epicsShareClass, epicsShareDef etc.
that are defined in the shareLib.h header.
This change may affect some user code which uses the epicsShare* macros
without having explicitly included the shareLib.h header themselves.
Such code should be changed to include shareLib.h directly.
A new helper script makeAPIheader.pl and build rules to generate a
library-specific *API.h header file has been added. Run makeAPIheader.pl -h
for information on how to use this in your own applications, but note that the
resulting sources will not be able to be compiled using earlier versions of
EPICS Base.
At the iocShell prompt help <cmd> now prints a descriptive usage message
for many internal IOCsh commands in addition to the command parameters.
Try help * to see all commands, or a glob pattern such as help db* to see
a subset.
External code may provide usage messages when registering commands using a
new const char *usage member of the iocshFuncDef structure.
The iocsh.h header also now defines a macro IOCSHFUNCDEF_HAS_USAGE which
can be used to detect Base versions that support this feature at compile-time.
configure/RELEASE files are parsed by both GNUmake and the convertRelease.pl
script. While GNUmake is quite relaxed about what characters may be used in a
RELEASE variable name, the convertRelease.pl script parser has only recognized
variable names that match the Perl regular expression \w+, i.e. upper and
lower-case letters, digits and underscore characters.
The script has been modified so now RELEASE variable names must start with a
letter or underscore, and be followed by any number of letters, digits,
underscore or hyphen characters, matching the regular expression
[A-Za-z_][A-Za-z_0-9-]*. The hyphen character - was not previously allowed
and if used would have prevented a build from finding include files and
libraries in any module using that in its RELEASE variable name.
This change does disallow names that start with a digit which used to be
allowed, but hopefully nobody has been relying on that ability. The regular
expression used for names can be found in the file src/tools/EPICS/Release.pm
and can be adjusted locally if necessary.
On *NIX targets caRepeater will now partially daemonize by redirecting stdin/out/err to /dev/null. This prevents caRepeater from inheriting the stdin/out of a process, like caget, which has spawned it in the background. This has been known to cause problems in some cases when caget is itself being run from a shell script.
caRepeater will now understand the -v argument to retain stdin/out/err
which may be necessary to see any error messages it may emit.
IOCs now emit a warning when a database file containing the state record is
loaded. This record has been deprecated for a while and will be removed
beginning with EPICS 7.1. Consider using the stringin record instead.
The record types in Base now define their device support entry table (DSET) structures in the record header file. While still optional, developers of external support modules are encouraged to start converting their code to use the record's new definitions instead of the traditional approach of copying the structure definitions into each source file that needs them. By following the instructions below it is still possible for the converted code to build and work with older Base releases.
This would also be a good time to modify the device support to use the type-safe device support entry tables that were introduced in Base-3.16.2 -- see this entry below for the description of that change, which is also optional for now.
Look at the aiRecord for example. Near the top of the generated aiRecord.h
header file is a new section that declares the aidset:
/* Declare Device Support Entry Table */
struct aiRecord;
typedef struct aidset {
dset common;
long (*read_ai)(struct aiRecord *prec);
long (*special_linconv)(struct aiRecord *prec, int after);
} aidset;
#define HAS_aidsetNotice that the common members (number, report(), init(), init_record()
and get_ioint_info() don't appear directly but are included by embedding the
dset common member instead. This avoids the need to have separate definitions
of those members in each record dset, but does require those members to be
wrapped inside another set of braces {} when initializing the data structure
for the individual device supports. It also requires changes to code that
references those common members, but that code usually only appears inside the
record type implementation and very rarely in device supports.
An aiRecord device support that will only be built against this or later versions of EPICS can now declare its dset like this:
aidset devAiSoft = {
{ 6, NULL, NULL, init_record, NULL },
read_ai, NULL
};
epicsExportAddress(dset, devAiSoft);However most device support that is not built into EPICS itself will need to
remain compatible with older EPICS versions, which is why the ai record's header
file also declares the preprocessor macro HAS_aidset. This makes it easy to
define the aidset in the device support code when it's needed, and not when
it's provided in the header:
#ifndef HAS_aidset
typedef struct aidset {
dset common;
long (*read_ai)(aiRecord *prec);
long (*special_linconv)(aiRecord *prec, int after);
} aidset;
#endif
aidset devAiSoft = {
{ 6, NULL, NULL, init_record, NULL },
read_ai, NULL
};
epicsExportAddress(dset, devAiSoft);The above typedef struct declaration was copied directly from the new
aiRecord.h file and wrapped in the #ifndef HAS_aidset conditional.
This same pattern should be followed for all record types except for the lsi,
lso and printf record types, which have published their device support entry
table structures since they were first added to Base but didn't previously embed
the dset common member. Device support for these record types therefore can't
use the dset name since the new definitions are different from the originals and
will cause a compile error, so this pattern should be used instead:
#ifndef HAS_lsidset
struct {
dset common;
long (*read_string)(lsiRecord *prec);
}
#else
lsidset
#endif
devLsiEtherIP = {
{5, NULL, lsi_init, lsi_init_record, get_ioint_info},
lsi_read
};IMPORTANT NOTE: Some record types in this release will not be compatible with device support binaries compiled against earlier versions of those record types, because importing the record documentation from the EPICS Wiki as described below also modified the order of some of the fields in the record definitions. As long as all support modules and IOCs are rebuilt from source after updating them to use this release of EPICS Base, these changes should not have any affect.
On supported targets (Linux, Mac, Windows) logClient will attempt to avoid dropping undelivered log messages when the connection to the log server is closed/reset.
Many internal timers and delay calculations use a monotonic clock epicsTimeGetMonotonic() instead of the realtime epicsTimeGetCurrent(). This is intended to make IOCs less susceptible to jumps in system time.
A new statement is added to enable IOC shell commands to signal error conditions, and for scripts to respond. This first is through the new function
int iocshSetError(int err);A script may be prefixed with eg. "on error break" to stop at the failed command.
on error continue | break | wait [value] | haltA suggested form for IOC shell commands is:
static void doSomethingCallFunc(const iocshArgBuf *args)
{
iocshSetError(doSomething(...)); /* return 0 == success */
}Allows built trees to be copied or moved without invalidating RPATH entires.
The LINKER_USE_RPATH Makefile variable (see configure/CONFIG_SITE) may be
set to YES, NO, and a new third option ORIGIN. This is limited to
targets using the ELF executable format (eg. Linux).
When LINKER_USE_RPATH=ORIGIN, the variable LINKER_ORIGIN_ROOT is set to
one of the parents of the build directory. Any libraries being linked
to which are found under this root will have a relative RPATH entry.
Other libraries continue to result in absolute RPATH entries.
An effect of this might change a support library from being linked with
-Wl,-rpath /build/epics-base/lib/linux-x86
to being linked with
-Wl,-rpath \$ORIGIN/../../../epics-base/lib/linux-x86
if the support module directory is /build/mymodule
and LINKER_ORIGIN_ROOT=/build.
The API functions epicsGetExecDir() and epicsGetExecName() are also
added to osiFileName.h to provide runtime access to the directory or
filename of the executable with which the process was started.
Previously, setting STATIC_BUILD=NO implied LINKER_USE_RPATH=NO.
This is no longer the case. Setting LINKER_USE_RPATH=YES will
always emit RPATH entries. This was found to be helpful when linking
against some 3rd party libraries which are only available as shared objects.
Host names given in a HAG entry of an IOC's Access Security Configuration
File (ACF) have to date been compared against the hostname provided by the CA
client at connection time, which may or may not be the actual name of that
client. This allows rogue clients to pretend to be a different host, and the
IOC would believe them.
An option is now available to cause an IOC to ask its operating system to look
up the IP address of any hostnames listed in its ACF (which will normally be
done using the DNS or the /etc/hosts file). The IOC will then compare the
resulting IP address against the client's actual IP address when checking
access permissions at connection time. This name resolution is performed at
ACF file load time, which has a few consequences:
-
If the DNS is slow when the names are resolved this will delay the process of loading the ACF file.
-
If a host name cannot be resolved the IOC will proceed, but this host name will never be matched.
-
Any changes in the hostname to IP address mapping will not be picked up by the IOC unless and until the ACF file gets reloaded.
Optionally, IP addresses may be added instead of, or in addition to, host names in the ACF file.
This feature can be enabled before iocInit with
var("asCheckClientIP",1)
or with the VxWorks target shell use
asCheckClientIP = 1A new routine epicsThreadCreateOpt() is an alternative to
epicsThreadCreate() which takes some arguments via a structure (struct epicsThreadOpts) to allow for future extensions.
typedef struct epicsThreadOpts {
unsigned int priority;
unsigned int stackSize;
unsigned int joinable;
} epicsThreadOpts;
#define EPICS_THREAD_OPTS_INIT { \
epicsThreadPriorityLow, epicsThreadStackMedium, 0}
epicsThreadId epicsThreadCreateOpt(const char * name,
EPICSTHREADFUNC funptr, void * parm, const epicsThreadOpts *opts);The final opts parameter may be NULL to use the default values of thread
priority (low) and stack size (medium). Callers wishing to provide alternative
settings for these thread options or to create a joinable thread (see below)
should create and pass in an epicsThreadOpts structure as shown below.
Always initialize one of these structures using the EPICS_THREAD_OPTS_INIT
macro to ensure that any additional fields that get added in the future are
set to their default values.
void startitup(void) {
epicsThreadOpts opts = EPICS_THREAD_OPTS_INIT;
epicsThreadId tid;
opts.priority = epicsThreadPriorityMedium;
tid = epicsThreadCreateOpt("my thread", &threadMain, NULL, &opts);
}C or C++ Code that also needs to build on earlier versions of Base can use
#ifdef EPICS_THREAD_OPTS_INIT to determine whether the
epicsThreadCreateOpt() API is available on this Base version.
The stackSize member of the epicsThreadOpts structure and the equivalent
parameters to the epicsThreadCreate() and epicsThreadMustCreate() routines
can now be passed either one of the epicsThreadStackSizeClass enum values or
a value returned from the epicsThreadGetStackSize() routine.
If the new joinable flag of an epicsThreadOpts structure is non-zero (the
default value is zero), the new API routine epicsThreadMustJoin() must be
called with the thread's epicsThreadId when/after the thread exits, to free
up thread resources. This function will block until the thread's main function
has returned, allowing the parent to wait for its child thread. The child's
epicsThreadId will no longer be valid and should not be used after the
epicsThreadMustJoin() routine returns.
A thread that was originally created with its joinable flag set may itself
call epicsThreadMustJoin(), passing in its own epicsThreadId. This marks the
thread as no longer being joinable, so it will then free the thread resources
itself when its main function returns. The epicsThreadId of a thread that is
not joinable gets invalidated as soon as its main function returns.
Previously IOC executables that made calls to devLib routines would fail to link when built for some non-VME based RTEMS targets, which would have to be explicitly filtered out by sites that build Base for those targets. This fix makes that no longer necessary, all RTEMS targets should now link although the IOC won't be able to be used with the VME I/O on those systems (that we don't have VMEbus I/O support for in libCom).
Add a fast path to epicsTimeGetCurrent() and related calls in the common case where only the default OS current time provider is registered. This path does not take the global mutex guarding the time providers list, potentially reducing lock contention.
The size of the queue used by dbEvent to push monitor updates has been
slightly increased based on DBR_TIME_DOUBLE to better fill an ethernet frame.
This may result in slightly fewer, but larger frames being sent.
Report NOBT as "precision" through the dbAccess API. This is not accessible through CA, but is planned to be used through QSRV.
-
The GNUmake build targets
cvscleananddepcleanare now available from any directory; previously they were only available from application top directories. -
The approach that EPICS Base uses for building submodules inside the parent module looks useful for support modules too. The rules for building submodules have been modified and extracted into a new
RULES_MODULESfile, so a support module will be able to use them too without having to copy them into its ownmodules/Makefile. There are some specific requirements that support modules and their submodules must follow, which are described as comments in the newbase/configure/RULES_MODULESfile itself.
In the past, a build of EPICS using sources checked out from the repository branch between official releases would have shown the version number of the previous release, followed by a -DEV suffix, for example 7.0.2.1-DEV.
The policy that controls when the number gets updated has been changed, and now immediately after a release has been tagged the version number will be updated to the next patch release version, plus the -DEV suffix as before. Thus following 7.0.2.2 the version number will show as 7.0.2.3-DEV. This does not require the next official release to be numbered 7.0.2.3 though, it could become 7.0.3 or even 7.1.0 if the changes incorporated into it are more substantial than bug fixes.
Turns out this is ~10x slower to query than CLOCK_MONOTONIC.
The linker flag -flat_namespace has been restored for creating shared
libraries, although not for loadable libraries (bundles). This was required
for building using the latest versions of Apple XCode.
A regression was introduced in 7.0.2 which caused record chains with loops to
be incorrectly broken. Processing should be skipped when a DB_LINK with
Process Passive (PP) closes a loop to a synchronous record.
Instead in 7.0.2 the targeted record would be processed if processing began
with a remote action (or some other caller of dbPutField()). This would
result in the loop running a second time. The loop would be broken on the
second iteration.
Support for some obsolete dbStaticLib Database Configuration Tool (DCT) APIs was removed some time ago, but vestiges of them still remained. The following routines and macros and have now finally been removed:
int dbGetFieldType(DBENTRY *pdbentry)int dbGetLinkType(DBENTRY *pdbentry)DCT_STRINGDCT_INTEGERDCT_REALDCT_MENUDCT_MENUFORMDCT_INLINKDCT_OUTLINKDCT_FWDLINKDCT_NOACCESSDCT_LINK_CONSTANTDCT_LINK_FORMDCT_LINK_PV
The dbhcr command used to work before iocInit as well as afterwards. It
displays all records that have hardware addresses (VME_IO, CAMAC_IO,
GPIB_IO, INST_IO etc.) but stopped working if run before iocInit due to the
rewrite of the link address parser code in dbStaticLib. This release fixes that
issue, although in some cases the output may be slightly different than it used
to be.
The list of tracked bugs fixed in this release can be found on the Launchpad Milestone page for EPICS Base 7.0.2.
The four separate Git branches core/master, libcom/master, ca/master and
database/master have been recombined into one branch called 7.0. Keeping
these as 4 separate branches in the same repository made it impossible to
create merge requests that contained changes in more than one of these
modules. The layout of the source files has not changed at all however, so the
source code for libcom, ca and the database are still found separately under
the module subdirectory.
A failure when fetching the simulation mode through SIML will not put the
record into INVALID alarm state anymore. Instead, as long as the record's
current alarm severity (SEVR)is NO_ALARM, its alarm status (STAT) will be
set to LINK_ALARM without increasing the severity. This allows clients to get
some notification of a failing or bad SIML link without otherwise affecting
record processing.
This routine was removed in Base-3.16.1 but has been reimplemented in this release by special request. Note that the error message strings that it returns when verification fails have changed, but are still designed for display to the user.
Records that support simulation mode have two new fields, SSCN (Simulation
Scan Mode) and SDLY (Simulation Delay). SSCN is a menu field that provides
an alternate value for the SCAN field to be used while the record is in
simulation mode. This is especially useful for I/O scanned records, for which
simulation mode was not working at all. Setting SDLY to a positive value
makes the record process asynchronously in simulation mode, with the second
stage processing happening after the specified time (in seconds).
This change permits IOCs to be built that omit the CA server (RSRV) by
removing its registrar entry which is now provided in the new rsrv.dbd file.
Other server layers can be built into the IOC (alongside RSRV or in place of
it) by registering them in a similar manner. The dbServer API is documented
with Doxygen comments in the header file.
Specific IOC server layers can be disabled at runtime by adding their name to
the environment variable EPICS_IOC_IGNORE_SERVERS (separated by spaces if more
than one should be ignored).
EPICS 7.0.1 contains the IOC Database, RSRV server and the Channel Access client code from EPICS Base 3.16.1 along with all the original record types and soft device support, but GDD and the Portable Channel Access Server have been unbundled and are now available separately. In their place we have brought in the more recently written EPICS V4 C++ libraries (collectively referred to as the PVA modules). The directory tree for EPICS is somewhat larger as a result, and the original structure of the Base directories has been split into 4 separate Git repositories. External modules should build against this new structure with little or no changes needed, except that some allowance may be needed for the merging of the V4 modules.
There should be rather more description and documantation of these changes than is currently available, but as developers we generally much prefer to write code than documentation. Send questions to the tech-talk mailing list and we'll be happy to try and answer them!
The list of tracked bugs fixed in this release can be found on the Launchpad Milestone page for EPICS Base 3.16.2.
Two new iocsh commands and some associated underlying APIs have been added to
show the state of the queues that feed the three callback tasks and the
scanOnce task, including a high-water mark which can optionally be reset. The
new iocsh commands are callbackQueueShow and scanOnceQueueShow; both take
an optional integer argument which must be non-zero to reset the high-water
mark.
Event numbers greater than or equal to NUM_TIME_EVENTS are now allowed if
supported by the registered event time provider, which must provide its own
advancing timestamp validation for such events.
Time events numbered 0 through (NUM_TIME_EVENTS-1) are still validated by code
in epicsGeneralTime.c that checks for advancing timestamps and enforces that
restriction.
Type-safe versions of the device and driver support structures dset and
drvet have been added to the devSup.h and drvSup.h headers respectively. The
original structure definitions have not been changed so existing support
modules will still build normally, but older modules can be modified and new
code written to be compatible with both.
The old structure definitions will be replaced by the new ones if the macros
USE_TYPED_DSET and/or USE_TYPED_DRVET are defined when the appropriate
header is included. The best place to define these is in the Makefile, as with
the USE_TYPED_RSET macro that was introduced in Base-3.16.1 and described
below. See the comments in devSup.h for a brief usage example, or look at
this commit
to the ipac module to see a module conversion.
A helper function DBLINK* dbGetDevLink(dbCommon *prec) has also been added
to devSup.h which fetches a pointer to the INP or OUT field of the record.
This release includes the ability to run the EPICS unit tests built for a
special version of the RTEMS-pc386 target architecture on systems that have an
appropriate QEMU emulator installed (qemu-system-i386). It is also now
possible to create sub-architectures of RTEMS targets, whereas previously the
EPICS target architecture name had to be RTEMS-$(RTEMS_BSP).
The new target RTEMS-pc386-qemu builds binaries that can be run in the
qemu-system-i386 PC System emulator. This target is a derivative of the
original RTEMS-pc386 target but with additional software to build an in-
memory file-system, and some minor modifications to allow the unit tests to
work properly under QEMU. When this target is enabled, building any of the
make targets that cause the built-in self-tests to be run (such as make runtests) will also run the tests for RTEMS using QEMU.
To allow the new 3-component RTEMS target name, the EPICS build system for
RTEMS was modified to allow a configure/os/CONFIG.Common.<arch> file to set
the RTEMS_BSP variable to inform the build what RTEMS BSP to use. Previously
this was inferred from the value of the T_A make variable, but that prevents
having multiple EPICS targets that build against the same BSP. All the
included RTEMS target configuration files have been updated; build
configuration files for out-of-tree RTEMS targets will continue to work as the
original rules are used to set RTEMS_BSP if it hasn't been set when needed.
This release adds three new link types: "state", "debug" and "trace". The
"state" link type gets and puts boolean values from/to the dbState library
that was added in the 3.15.1 release. The "debug" link type sets the
jlink::debug flag in its child link, while the "trace" link type also causes
the arguments and return values for all calls to the child link's jlif and
lset routines to be printed on stdout. The debug flag can no longer be set
using an info tag. The addition of the "trace" link type has allowed over 200
lines of conditional diagnostic printf() calls to be removed from the other
link types.
The "calc" link type can now be used for output links as well as input links. This allows modification of the output value and even combining it with values from other input links. See the separate JSON Link types document for details.
A new start_child() method was added to the end of the jlif interface table.
The lset methods have now been properly documented in the dbLink.h header
file using Doxygen annotations, although we do not run Doxygen on the source
tree yet to generate API documentation.
Link types that utilize child links must now indicate whether the child will
be used for input, output or forward linking by the return value from its
parse_start_map() method. The jlif_key_result enum now contains 3 values
jlif_key_child_inlink, jlif_key_child_outlink and jlif_key_child_fwdlink
instead of the single jlif_key_child_link that was previously used for this.
Some additional build rules have been added to help debug configuration
problems with the build system. Run make show-makefiles to get a sorted list
of all the files that the build system includes when building in the current
directory.
A new pattern rule for PRINT.% can be used to show the value of any GNUmake
variable for the current build directory (make sure you are in the right
directory though, many variables are only set when inside the O.<arch> build
directory). For example make PRINT.T_A will display the build target
architecture name from inside a O.<arch> directory but the variable will be
empty from an application top or src directory. make PRINT.EPICS_BASE will
show the path to Base from any EPICS application directory though.
The IOC contains a mechanism involving the PUTF and RPRO fields of each record to ensure that if a record is busy when it receives a put to one of its fields, the record will be processed again to ensure that the new field value has been correctly acted on. Until now that mechanism only worked if the put was to the asynchronous record itself, so puts that were chained from some other record via a DB link did not cause reprocessing.
In this release the mechanism has been extended to propagate the PUTF state
across DB links until all downstream records have been reprocessed. Some
additional information about the record state can be shown by setting the TPRO
field of an upstream record, and even more trace data is displayed if the
debugging variable dbAccessDebugPUTF is set in addition to TPRO.
A new iocsh command dbli lists the info fields defined in the database, and
can take a glob pattern to limit output to specific info names. The newly
added dbStaticLib function dbNextMatchingInfo() iterates through the info
fields defined in the current record, and is used to implement the new
command.
The "DataBase Print Record" command dbpr now generates slightly better
output, with more field types having their own display methods. This release
also includes additional protection against buffer overflows while printing
long links in dbpr, and corrects the output of long strings from the dbgf
command.
The VAL fields and related fields of these records are now DBF_LONG. (Not
DBF_ULONG in order to prevent Channel Access from promoting them to
DBF_DOUBLE.) Additional bit fields B10...B1F have been added.
Device support that accesses VAL or the bit fields directly (most don't) and
aims for compatibility with old and new versions of these records should use
at least 32 bit integer types to avoid bit loss. The number of bit fields can
be calculated using 8 * sizeof(prec->val) which is correct in both versions.
The epicsReadline refactoring work described below unfortunately disabled the VxWorks implementation of the osdReadline.c API that uses ledlib for command editing and history. This functionality has now been restored, see Launchpad bug #1741578.
Constant links can now hold 64-bit integer values, either as scalars or arrays. Only base 10 is supported by the JSON parser though, the JSON standard doesn't allow for hexadecimal numbers.
The third-party YAJL library that has been included in libCom for several
years has been upgraded to version 2.1.0 and several bugs fixed. This has an
updated API, requiring any code that uses it to parse its own JSON files to be
modified to match. The changes are mainly that it uses size_t instead
unsigned int for string lengths, but it also uses long long instead of
long for JSON integer values, which was the main motivation for the upgrade.
The self-tests that YAJL comes with have been imported and are now run as an
EPICS Unit Test program, and the JSON syntax accepted by the parser was
extended to permit trailing commas in both arrays and maps. The difference
between the old and new YAJL APIs can be detected at compile time by looking
for the macro EPICS_YAJL_VERSION which is defined in the yajl_common.h
header file along with a brief description of the API changes.
A new optional parameter can be given when specifying a calc JSON link. The
time parameter is a string containing a single letter A..L that selects
one of the input links to be used for the timestamp of calculation if
requested. The timestamp will be fetched atomically with the value from the
chosen input link (providing that input link type supports the readLocked()
method).
A soft channel output record with the OUT link unset uses the CONSTANT link
type. The new link type code was causing some soft channel device supports to
return an error status from the write method of that link type, which would
cause a ca_put() operation to such a record to generate an exception. This has
been silenced by giving the constant link types a dummy putValue method. A new
test program has been added to prevent regressions of this behaviour.
In the 3.16.1 release a crash can occur in the IOC's RSRV server when a large array is made even larger; the previous array buffer was not being released correctly. See Launchpad bug #1706703.
The IOC now supports the 64-bit integer field types DBF_INT64 and
DBF_UINT64, and there are new record types int64in and int64out derived
from the longin and longout types respectively that use the DBF_INT64
data type for their VAL and related fields. The usual range of Soft Channel
device support are included for these new record types.
All internal IOC APIs such as dbAccess can handle the new field types and
their associated request values DBR_INT64 and DBR_UINT64, which are
implemented using the epicsInt64 and epicsUInt64 typedef's from the
epicsTypes.h header.
The waveform record type has been updated to support these new field types.
All waveform device support layers must be updated to recognize the new type
enumeration values, which had to be inserted before the FLOAT value in the
enum dbfType and in menuFtype. C or C++ code can detect at compile-time
whether this version of base provides 64-bit support by checking for the
presence of the DBR_INT64 macro as follows (Note that DBF_INT64 is an
enum tag and not a preprocessor macro):
#ifdef DBR_INT64
/* Code where Base has INT64 support */
#else
/* Code for older versions */
#endif
If the code uses the old db_access.h types (probably because it's calling
Channel Access APIs) then it will have to test against the EPICS version
number instead, like this:
#include <epicsVersion.h>
#ifndef VERSION_INT
# define VERSION_INT(V,R,M,P) ( ((V)<<24) | ((R)<<16) | ((M)<<8) | (P))
#endif
#ifndef EPICS_VERSION_INT
# define EPICS_VERSION_INT VERSION_INT(EPICS_VERSION, EPICS_REVISION, EPICS_MODIFICATION, EPICS_PATCH_LEVEL)
#endif
#if EPICS_VERSION_INT >= VERSION_INT(3,16,1,0)
/* Code where Base has INT64 support */
#else
/* Code for older versions */
#endif
Channel Access does not (and probably never will) directly support 64-bit
integer types, so the new field types are presented to the CA server as
DBF_DOUBLE values. This means that field values larger than 2^52
(0x10_0000_0000_0000 = 4503599627370496) cannot be transported over Channel
Access without their least significant bits being truncated. The EPICS V4
pvAccess network protocol can transport 64-bit data types however, and a
future release of the pvaSrv module will connect this ability to the fields of
the IOC.
Additional 64-bit support will be provided in later release. For instance the JSON parser for the new Link Support feature only handles integers up to 32 bits wide, so constant array initializer values cannot hold larger values in this release.
A new environment parameter EPICS_CA_MCAST_TTL is used to set the Time To Live
(TTL) value of any IP multi-cast CA search or beacon packets sent.
A new environment parameter EPICS_CA_AUTO_ARRAY_BYTES is now used by libca and
RSRV (CA clients and the IOC CA server). The default is equivalent to setting
EPICS_CA_AUTO_ARRAY_BYTES=YES which removes the need to set
EPICS_CA_MAX_ARRAY_BYTES and always attempts to allocate sufficiently large
network buffers to transfer large arrays properly over the network. In this case
the value of the EPICS_CA_MAX_ARRAY_BYTES parameter is ignored.
Explicitly setting EPICS_CA_AUTO_ARRAY_BYTES=NO will continue to honor the
buffer setting in EPICS_CA_AUTO_ARRAY_BYTES as in previous releases.
The default setting for EPICS_CA_AUTO_ARRAY_BYTES can be changed by adding the
line
EPICS_CA_AUTO_ARRAY_BYTES=NOto the configure/CONFIG_SITE_ENV file before building Base. Sites that wish to
override this only for specific IOC architectures can create new files for each
architecture named configure/os/CONFIG_SITE_ENV.<target-arch> with the above
setting in before building Base. The configuration can also be explicitly
changed by setting the environment variable in the IOC's startup script,
anywhere above the iocInit line.
The PCAS server (used by the PV Gateway and other CA servers) now always behaves
as if EPICS_CA_AUTO_ARRAY_BYTES is set to YES (it ignores the configuration
parameter and environment variable).
Drop support for CA clients advertising protocol versions less than 4.
This effects clients from Base older than 3.12.0-beta1. Newer clients will continue to be able to connect to older servers. Older clients will be ignored by newer servers.
This allows removal of UDP echo and similar protocol features which are not compatible with secure protocol design practice.
The subArray record can now be used as a lookup-table from a constant array specified in its INP field. For example:
record(subArray, "powers-of-2") {
field(FTVL, "LONG")
field(MALM, 12)
field(INP, [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048])
field(INDX, 0)
field(NELM, 1)
}
The INDX field selects which power of 2 to set the VAL field to. In previous releases the INP field would have to have been pointed to a separate waveform record that was initialized with the array values somehow at initialization time.
Most Soft Channel input device support routines have supported fetching the timestamp through the INP link along with the input data. However before now there was no guarantee that the timestamp provided by a CA link came from the same update as the data, since the two were read from the CA input buffer at separate times without maintaining a lock on that buffer in between. This shortcoming could be fixed as a result of the new link support code, which allows code using a link to pass a subroutine to the link type which will be run with the link locked. The subroutine may make multiple requests for metadata from the link, but must not block.
A major new feature introduced with this release of EPICS Base is an Extensible Link Type mechanism, also known as Link Support or JSON Link Types. This addition permits new kinds of link I/O to be added to an IOC in a similar manner to the other extension points already supported (e.g. record, device and driver support).
A new link type must implement two related APIs, one for parsing the JSON
string which provides the link address and the other which implements the link
operations that get called at run-time to perform I/O. The link type is built
into the IOC by providing a new link entry in a DBD file.
This release contains two new JSON link types, const and calc:
-
The
constlink type is almost equivalent to the old CONSTANT link type with the updates described below to accept arrays and strings, except that there is no need to wrap a scalar string constant inside array brackets since a constant string will never be confused with a PV name. -
The
calclink type allows CALC expressions to be used to combine values from other JSON links to produce its value. Until additional JSON link types are created though, thecalclink type has little practical utility as it can currently only fetch inputs from othercalclinks or fromconstlinks.
field(INP, {calc:{expr:"A+B+1",
args:[5, # A
{const:6}] # B
}
}
)
The new link types are documented in a separate document that gets generated at build time and installed as html/links.html.
The API to allow device support to use JSON addresses is currently
incomplete; developers are advised not to try creating device support that
specifies a JSON_LINK address type.
For link fields in external record types and soft device support to be able to use the new link types properly, various changes are required to utilize the new Link Support API as defined in the dbLink.h header file and outlined below. The existing built-in Database and Channel Access link types have been altered to implement the link APIs, so will work properly after these conversions:
- Make all calls to
recGblInitConstantLink()unconditional on the link type, i.e. change this code:
if (prec->siml.type == CONSTANT) {
recGblInitConstantLink(&prec->siml, DBF_USHORT, &prec->simm);
}into this:
recGblInitConstantLink(&prec->siml, DBF_USHORT, &prec->simm);Note that recGblInitConstantLink() still returns TRUE if the field was
successfully initialized from the link (implying the link is constant).
This change will work properly with all Base releases currently in use.
- Code that needs to identify a constant link should be modified to use
the new routine
dbLinkIsConstant()instead, which returns TRUE for constant or undefined links, FALSE for links whosedbGetLink()routine may return different values on different calls. For example this:
if (prec->dol.type != CONSTANT)should become this:
if (!dbLinkIsConstant(&prec->dol))When the converted software is also required to build against older versions of Base, this macro definition may be useful:
#define dbLinkIsConstant(lnk) ((lnk)->type == CONSTANT)- Any code that calls dbCa routines directly, or that explicitly checks if a link has been resolved as a CA link using code such as
if (prec->inp.type == CA_LINK)will still compile and run, but will only work properly with the old CA link type. To operate with the new extensible link types such code must be modified to use the new generic routines defined in dbLink.h and should never attempt to examine or modify data inside the link. After conversion the above line would probably become:
if (dbLinkIsVolatile(&prec->inp))A volatile link is one like a Channel Access link which may disconnect and reconnect without notice at runtime. Database links and constant links are not volatile; unless their link address is changed they will always remain in the same state they started in. For compatibility when building against older versions of Base, this macro definition may be useful:
#define dbLinkIsVolatile(lnk) ((lnk)->type == CA_LINK)- The current connection state of a volatile link can be found using the
routine
dbIsLinkConnected()which will only return TRUE for a volatile link that is currently connected. Code using the older dbCa API returning this information used to look like this:
stat = dbCaIsLinkConnected(plink);which should become:
stat = dbIsLinkConnected(plink);Similar changes should be made for calls to the other dbCa routines.
- A full example can be found by looking at the changes to the calcout record type, which has been modified in this release to use the new dbLink generic API.
Previously a constant link (i.e. a link that did not point to another PV, either locally or over Channel Access) was only able to provide a single numeric value to a record initialization; any string given in a link field that was not recognized as a number was treated as a PV name. In this release, constant links can be expressed using JSON array syntax and may provide array initialization of values containing integers, doubles or strings. An array containing a single string value can also be used to initialize scalar strings, so the stringin, stringout, lsi (long string input), lso (long string output), printf, waveform, subArray and aai (analog array input) record types and/or their soft device supports have been modified to support this.
Some examples of constant array and string initialized records are:
record(stringin, "const:string") {
field(INP, ["Not-a-PV-name"])
}
record(waveform, "const:longs") {
field(FTVL, LONG)
field(NELM, 10)
field(INP, [1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
}
record(aai, "const:doubles") {
field(FTVL, DOUBLE)
field(NELM, 10)
field(INP, [0, 1, 1.6e-19, 2.718, 3.141593])
}
record(aSub, "select") {
field(FTA, STRING)
field(NOA, 4)
field(INPA, ["Zero", "One", "Two", "Three"])
field(FTB, SHORT)
field(NOB, 1)
field(FTVA, STRING)
field(NOVA, 1)
field(SNAM, "select_asub")
}
Reminder: Link initialization with constant values normally only occurs at record initialization time. The calcout and printf record types are the only exceptions in the Base record types to this rule, so it is generally not useful to change a const link value after iocInit.
A database file can now provide a "relaxed JSON" value for a database field value or an info tag. Only a few field types can currently accept such values, but the capability is now available for use in other places in the future. When writing to a JSON-capable field at run-time however, only strictly compliant JSON may be used (the dbStaticLib parser rewrites relaxed JSON values into strict JSON before passing them to the datase for interpretation, where the strict rules must be followed).
"Relaxed JSON" was developed to maximize compatibility with the previous database parser rules and reduce the number of double-quotes that would be needed for strict JSON syntax. The parser does accept strict JSON too though, which should be used when machine-generating database files. The differences are:
-
Strings containing only the characters
a-z A-Z 0-9 _ - + .do not have to be enclosed in double-quote characters. -
The above rule applies to map keys as well as to regular string values.
-
The JSON keywords
null,trueandfalse(all lower-case) will be recognized as keywords, so they must be quoted to use any of these single words as a string. -
Comments may be used, introduced as usual by the
#character and extending to the end of the line.
A JSON field or info value is only enclosed in quotes when the value being provided is a single string, and even here the quotes can be omitted in some cases as described above. The following shows both correct and incorrect excerpts from a database file:
record(ai, math:pi) {
field(INP, {const: 3.14159265358979}) # Correct
field(SIOL, "{const: 3.142857}") # Wrong
info(autosave, { # White-space and comments are allowed
fields:[DESC, SIMM],
pass0:[VAL]
}) # Correct
}
Note that the record, field and info-tag names do not accept JSON values, so
they follows the older bareword rules for quoting where the colon : and
several additional characters are legal in a bareword string. Only the value
(after the comma) is parsed as JSON. The autosave module has not been modified
to accept JSON syntax, the above is only an example of how JSON might be used.
The way comments are parsed by the iocsh interpreter has changed. The
interpreter can be selectively disabled from echoing comments coming from a
script by starting those lines with #- rather than just #.
The table of record support functions (rset methods for short) no longer has
entries of type RECSUPFUN (which says: any number and type of arguments).
Instead, rset methods are now typed by default. The RECSUPFUN typedef has
been deprecated and casts to it as well as using the untyped struct rset
will create compilation warnings.
Existing code (e.g. external record supports) will generate such warnings when compiled against this version of Base, but it will work without changes.
For a conversion period, the new typed rset definitions are activated by
defining USE_TYPED_RSET, preferably by setting USR_CPPFLAGS += -DUSE_TYPED_RSET inside a Makefile. After activating the new typed rset in
this way and making the following changes, the result should still compile and
work properly against older versions of Base.
The first parameter of init_record and process has been changed to struct dbCommon *. Record types that use void* here should be changed to use
struct dbCommon*, and cast the argument to their own xxxRecord *.
When compiled against this release, compiler warnings about incompatible types for the method pointers should be taken seriously. When compiled against older versions of base, such warnings are unavoidable.
Record types written in C++ need to take more drastic measures because of the stricter type checking in C++. To remain compatible with older versions of base you will need to use something like:
#include "epicsVersion.h"
#ifdef VERSION_INT
# if EPICS_VERSION_INT < VERSION_INT(3,16,0,2)
# define RECSUPFUN_CAST (RECSUPFUN)
# else
# define RECSUPFUN_CAST
# endif
#else
# define RECSUPFUN_CAST (RECSUPFUN)
#endif
and then replace (RECSUPFUN) with RECSUPFUN_CAST when initializing the
rset. Further changes might also be needed, e.g. to adapt const-ness of
method parameters.
The build rules associated with the CapFast-related tools sch2edif and
e2db and the database optimization tool dbst have been removed, along with
the DB_OPT build configuration variable.
The compressRecord has a new field BALG which can select between FIFO
(append) and LIFO (prepend) ordering for insertion of new elements. FIFO
ordering is the default, matching the behviour of previous versions.
Valgrind is a software debugging suite provided by many Linux distributions. The header valgrind/valgrind.h is now included in, and installed by, Base. When included by a C or C++ source file this header defines some macros which expand to provide hints to the Valgrind runtime. These have no effect on normal operation of the software, but when run using the valgrind tool they can help to find memory leaks and buffer overflows. Suitable hints have been added to several free-lists within libCom, including freeListLib, allowing valgrind to provide more accurate information about the source of potential leaks.
valgrind.h automatically disables itself when the build target is not
supported by the valgrind tool. It can also explicitly be disabled by defining
the macro NVALGRIND. See src/libCom/Makefile for a commented-out example.
As a matter of policy valgrind.h will never be included by any header file installed by Base, so its use will remain purely an implementation detail hidden from application software. Support modules which choose to use valgrind.h are advised to do likewise.
The IOC record locking code has been re-written with an expanded API; global locks are no longer required by the IOC database implementation.
The new API functions center around dbScanLockMany(), which behaves like
dbScanLock() applied to an arbitrary group of records. dbLockerAlloc() is
used to prepare a list or record pointers, then dbScanLockMany() is called.
When it returns, all of the records listed may be accessed (in any order) until
dbScanUnlockMany() is called.
The Application Developer's Guide has been updated to describe the API and implementation is more detail.
Previously a global mutex lockSetModifyLock was locked and unlocked during
dbScanLock(), acting as a sequencing point for otherwise unrelated calls. The
new dbLock.c implementation does not include any global mutex in dbScanLock()
or dbScanLockMany(). Locking and unlocking of unrelated lock sets is now
completely concurrent.
A Perl script and Makefile rules have been added to allow modules to generate a C header file with a macro defined with an automatically updated identifier. This is a VCS revision ID (Darcs, Git, Mercurial, Subversion, and Bazaar are supported) or the date/time of the build if no VCS system is in use.
The makeBaseApp example template has been updated with a new device support which makes this identifier visible via a lsi (long string input) record.
The epicsTime routines that used to return epicsTimeERROR now return a
specific S_time_ status value, allowing the caller to discover the reason for
any failure. The identifier epicsTimeERROR is no longer defined, so any
references to it in source code will no longer compile. The identifier
epicsTimeOK still exists and has the value 0 as before, so most code that uses
these APIs can be changed in a way that is backwards-compatible with the
previous return status.
Time providers that have to return a status value and still need to be built with earlier versions of Base can define the necessary status symbols like this:
#include "epicsTime.h"
#ifndef M_time
/* S_time_... status values were not provided before Base 3.16 */
#define S_time_unsynchronized epicsTimeERROR
#define S_time_...whatever... epicsTimeERROR
#endif
The epicsReadline code has been reorganized to allow the commandline history
editor to be disabled at runtime. The EPICS_COMMANDLINE_LIBRARY build setting
still selects the preferred editor, but the new IOCSH_HISTEDIT_DISABLE
environment variable can be set at runtime to disable history editing and make
the IOC or other program use the basic editor instead. This is useful when
starting and controlling an IOC from another program through its stdin and
stdout streams since history editors often insert invisible escape codes into
the stdout stream, making it hard to parse.
Added a new macro callbackGetPriority(prio, callback) to the callback.h
header and removed the need for dbScan.c to reach into the internals of its
CALLBACK objects.
Thanks to Jeong Han Lee this release comes with build support for Apple's new
M1/M2 CPUs running macOS, using the target name darwin-aarch64.
On MS Windows, epicsThread names are made available to the OS and debugger
using SetThreadDescription() if available as well as using the older
exception mechanism.
The waitable timer changes in 3.15.9 broke calls to epicsThreadSleep() and
similar routines that used timers (including ca_pend_event()) when made from
threads that were not started using the epicsThread APIs.
This problem
has now been fixed.
The epicsEventWaitWithTimeout() and epicsThreadSleep() functions have
been changed to use waitable timers. On Windows 10 version 1803 or higher
they will use high resolution timers for more consistent timing.
See this Google Groups thread for a comparison of the performance of different timers.
The build target inc now works again after a very long hiatus. It now
generates and installs just the dbd, header and html files, without compiling
any C/C++ code. This can be used to speed up CI jobs that only generate
documentation.
- The error status returned by a record support's
special()method is now propagated out of thedbPut()routine again (broken since 3.15.0). - gh: #80, VS-2015 and later have working strtod()
- lp: #1776141, Catch buffer overflow from long link strings
- lp: #1899697, Records in wrong PHAS order
The names of the generated junit xml test output files have been changed
from <testname>.xml to <testname>-results.xml, to allow better
distinction from other xml files. (I.e., for easy wildcard matching.)
Issues reported by various static code checkers.
The following launchpad bugs have fixes included in this release:
- lp: 1812084, Build failure on RTEMS 4.10.2
- lp: 1829770, event record device support broken with constant INP
- lp: 1829919, IOC segfaults when calling dbLoadRecords after iocInit
- lp: 1838792, epicsCalc bit-wise operators on aarch64
- lp: 1841608, logClient falsely sends error logs on all connections
- lp: 1853168, undefined
reference to
clock_gettime() - lp: 1862328, Race condition on IOC start leaves rsrv unresponsive
- lp: 1868486, epicsMessageQueue lost messages
This release contains changes that make it possible to integrate another test running and reporting system (such as Google's gtest) into the EPICS build system. The built-in test-runner and reporting system will continue to be used by the test programs inside Base however.
These GNUmake tapfiles and test-results build targets now collect a list of
the directories that experienced test failures and display those at the end of
running and/or reporting all of the tests. The GNUmake process will also only
exit with an error status after running and/or reporting all of the test
results; previously the -k flag to make was needed and even that didn't always
work.
Continuous Integration systems are recommended to run make tapfiles (or if
they can read junittest output instead of TAP make junitfiles) followed by
make -s test-results to display the results of the tests. If multiple CPUs are
available the -j flag can be used to run tests in parallel, giving the maximum
jobs that should be allowed so make -j4 tapfiles for a system with 4 CPUs say.
Running many more jobs than you have CPUs is likely to be slower and is not
recommended.
The code that implements bit operations for Calc expressions has been reworked
to better handle some CPU architectures and compilers. As part of this work a
new operator has been added: >>> performs a logical right-shift, inserting
zero bits into the most significant bits (the operator >> is an arithmetic
right-shift which copies the sign bit as it shifts the value rightwards).
The IOC's error logging system has been updated significantly to fix a number of issues including:
- Only send errlog messages to iocLogClient listeners
- Try to minimize lost messages while the log server is down:
- Detect disconnects sooner
- Don't discard the buffer on disconnect
- Flush the buffer immediately after a server reconnects
VxWorks IOCs (and potentially RTEMS IOCs running GeSys) have had problems with
garbled error messages from dbStaticLib routines for some time — messages
printed before iocInit were being queued through the errlog thread instead of
being output immediately. This has been fixed by initializing the main thread
with its OkToBlock flag set instead of cleared. IOCs running on other
operating systems that use iocsh to execute the startup script previously had
that set anyway in iocsh so were not affected, but this change might cause other
programs that don't use iocsh to change their behavior slightly if they use
errlogPrintf(), epicsPrintf() or errPrintf().
The camonitor program didn't properly cope if subscribed to a channel whose data type changed when its IOC was rebooted without restarting the camonitor program. This has now been fixed.
The remaining record types have had their reference pages moved from the Wiki, and some new reference pages have been written to cover the analog array and long string input and output record types plus the printf record type, none of which were previously documented. The wiki reference pages covering the fields common to all, input, and output record types have also been added, thanks to Rolf Keitel. The POD conversion scripts have also been improved and they now properly support linking to subsections in a different document, although the POD changes to add the cross-links that appeared in the original wiki pages still needs to be done in most cases.
The clock_gettime() routine is no longer used under MinGW since newer versions
don't provide it any more.
If multiple IOCs were started at the same time, by systemd say, they could race to obtain the Channel Access TCP port number 5064. This issue has been fixed.
Most Linux architectures should now configure themselves automatically to use
the GNU Readline library if its main header file can be found in the expected
place, and not try to use Readline if the header file isn't present. For older
Linux architectures where libncurses or libcurses must also be linked with, the
manual configuration of the COMMANDLINE_LIBRARY variable in the appropriate
configure/os/CONFIG_SITE.Common.<arch> file will still be necessary.
The EPICS_TIMEZONE environment parameter provided time-zone information for
the IOC's locale in the old ANSI format expected by VxWorks for its TIMEZONE
environment variable, and can also used by RTEMS to set its TZ environment
variable. However the TIMEZONE value has to be updated every year since it
contains the exact dates of the daylight-savings time changes. The Posix TZ
format that RTEMS uses contains rules that for calculating those dates, thus its
value would only need updating if the rules (or the locale) are changed.
This release contains changes that replace the EPICS_TIMEZONE environment
parameter with one called EPICS_TZ and a routine for VxWorks that calculates
the TIMEZONE environment variable from the current TZ value. This routine
will be run once at start-up, when the EPICS clock has synchronized to its NTP
server. The calculations it contains were worked out and donated to EPICS by
Larry Hoff in 2009; it is unforunate that it has taken 10 years for them to be
integrated into Base.
The default value for the EPICS_TZ environment parameter is set in the Base
configure/CONFIG_SITE_ENV file, which contains example settings for most EPICS
sites that use VxWorks, and a link to a page describing the Posix TZ format for
any locations that I missed.
If a VxWorks IOC runs continuously without being rebooted from December 31st to
the start of daylight savings time the following year, its TIMEZONE value will
be wrong as it was calculated for the previous year. This only affects times
that are converted to a string on the IOC however and is easily fixed; just run
the command tz2timezone() on the VxWorks shell and the calculation will be
redone for the current year. IOCs that get rebooted at least once before the
start of summer time will not need this to be done.
A new server-side filter has been added to the IOC for reducing the number and frequency of monitor updates from a channel by a client-specified factor. The filter's behaviour is quite simplistic, it passes the first monitor event it sees to the client and then drops the next N-1 events before passing another event. For example to sample a 60Hz channel at 1Hz, a 10Hz channel every 6 seconds, or a 1Hz channel once every minute:
Hal$ camonitor 'test:channel.{"dec":{"n":60}}'
...
More information is included in the filters documentation, which can be found
in the html/filters.html document that is generated during the build.
The remaining record types that had 3.14 reference documentation in the EPICS Wiki have had that documentation converted and imported into their DBD files. The preferred form for future updates to the record type descriptions is now an emailed patch file, a Pull Request through GitHub, or a Merge Request through Launchpad. Note that in some cases the behavior of a record type in a 7.0.x release may differ from that of the same record type in a 3.15 release, although this would be unusual, so it may be important to indicate the branch that your changes apply to.
NOTE: These documentation changes have modified the order of the fields in some record definitions. As a result this release is not compatible with record or device support binaries that were compiled against earlier releases.
The make target test-results should now work properly on Windows. Some Perl
installations used versions of prove.bat that would only display the results of
up to 3 tests or didn't return an error status in the event of tests failing. The
build system now calls its own perl script to summarize the results instead of
passing a list of TAP filenames to prove.
If a macro EPICS_NO_CALLBACK is defined, then callback.h will no longer
(re)define CALLBACK. The name CALLBACK is used by the WIN32 API, and
redefinition in callback.h cause errors if some windows headers are later
included.
Code which defines EPICS_NO_CALLBACK, but still wishes to use callbacks,
should use the alternate name epicsCallback introduced in 3.15.6, 3.16.2, and
7.0.2. It is also possible, though not encouraged, to use struct callbackPvt
which has been present since the callback API was introduced.
Bruno Martins reported a problem with the CA client library at shutdown in a process that uses multiple CA client contexts. The first context that triggers the CA client exit handler prevents any others from being able to clean up because it resets the ID of an internal epicsThreadPrivate variable which is shared by all clients. This action has been removed from the client library, which makes cleanup of clients like this possible.
Apple removed some Perl header files from macOS Mojave that were available in their SDK, requiring a change to the include paths used when compiling the CA bindings. The new version should build on new and older macOS versions, and these changes may also help other targets that have an incomplete installation of Perl (the build will continue after printing a warning that the Perl CA bindings could not be built).
This routine was a simple wrapper around the C89 function tmpnam()
which is now seen as unsafe and causes warning messages to be generated by
most modern compilers. The two internal uses of this function have been
modified to call epicsTempFile() instead. We were unable to find any
published code that used this function, so it was removed immediately instead
of being deprecated.
The Perl DBD file parser has been made slightly more liberal; the order in which DBD files must be parsed is now more flexible, so that a record type definition can now be parsed after a device support that referred to that record type. A warning message will be displayed when the device support is seen, but the subsequent loading of the record type will be accepted without triggering an error. See Launchpad bug 1801145.
The EPICS Wiki pages describing a number of standard record types has been converted into the Perl POD documentation format and added to the DBD files, so at build-time an HTML version of these documents is generated and installed into the htmls directory. Thanks to Tony Pietryla.
This displays the version numbers of EPICS Base and the CA protocol.
The new command epicsEnvUnset varname can be used to
unset an environment variable.
The libCom macro expansion library has been modified so that when the
SUPPRESS_WARNINGS flag is set it will no longer include any ,undefined
or ,recursive indicators in its output when undefined or recursive
macros are encountered. These indicators were harmless when the output was fed
into an IOC along with a definition for the macro, but when the msi
tool was used to generate other kinds of files they caused problems. If the
msi -V flag is used the markers will still be present in the output
whenever the appropriate condition is seen.
In addition to fixing its response to discovering parsing errors in its substitution input file (reported as Launchpad bug 1503661) so it now deletes the incomplete output file, the msi program has been cleaned up a little bit internally.
The waveform record has been posting monitors on its NORD field since Base 3.15.0.1; we finally got around to doing the equivalent in all the other built-in record types, which even required modifying device support in some cases. This fixes Launchpad bug 1730727.
Some documentation has been added to the dbdToHtml.pl script
explaining how Perl POD (Plain Old Documentation) markup can be added to
.dbd files to generate HTML documentation for the record types. To see
these instructions, run perl bin/<host>/dbdToHtml.pl -H
or perldoc bin/<host>/dbdToHtml.pl.
Changing from numeric to named soft events introduced an incompatibility
when a numeric event 1-255 is converted from a DOUBLE, e.g. from a calc record.
The post_event() API is not marked deprecated any more.
Also scanpel has been modified to accept a glob pattern for
event name filtering and to show events with no connected records as well.
Added a new OS-independent typedef for multicast socket options, and a test file to check their correct operation.
This feature is mostly meant for use by developers; configuration
settings that would normally appear in base/configure/CONFIG_SITE can now
be put in a locally created base/configure/CONFIG_SITE.local file instead
of having go modify or replace the original. A new .gitignore pattern
tells git to ignore all configure/*.local files.
The Application Developers' Guide says this is allowed and disables the limit on the log-file, but it hasn't actually worked for some time (if ever). Note that the iocLogServer will be removed from newer Base release sometime soon as its functionality can be implemented by other dedicated log servers such as logstash or syslog-ng.
Fixes lp:1786858 and part of lp:1786966.
The files in the startup directory have not been maintained in recent years and have grown crufty (technical term). This release includes the following updates to these files:
- The Perl
EpicsHostArch.plscript has been rewritten, and support for a few previously missing host architectures has been added to it. - The
EpicsHostArch.plscript has also been moved into the standardsrc/toolsdirectory, from where it will be installed intolib/perl. In this new location it is no longer executable, so it must be run by theperlexecutable. - The build system has been adjusted to look for
EpicsHostArch.plin both places if theEPICS_HOST_ARCHenvironment variable has not been set at build-time. - Sites that used the original Perl script to set
EPICS_HOST_ARCHas part of their standard environment will need to adjust their scripts when they upgrade to this release. - The
EpicsHostArchshell script has been replaced with a wrapper routine that calls the PerlEpicsHostArch.plscript. Sites that rely on this script to setEPICS_HOST_ARCHshould consider switching to the Perl script instead. - The
Site.cshrcandSite.profilefiles have been renamed tounix.cshandunix.sh, respectively. - The existing
win32.batfile has been cleaned up and a newwindows.batfile added for 64-bit targets. The contents of these files should be seen as examples, don't uncomment or install parts for software that you don't explicitly know that you need.
The latest version of XCode will not compile calls to system() or
clock_settime() for iOS targets. There were several places in Base
where these were being compiled, although there were probably never called. The
code has now been modified to permit iOS builds to complete again.
A check has been added to recGblResetAlarms() that prevents records
from getting an alarm severity higher than INVALID_ALARM. It is still possible
for a field like HSV to get set to a value that is not a legal alarm severity,
but the core IOC code should never copy such a value into a record's SEVR or
ACKS fields. With this fix the record's alarm severity will be limited to
INVALID_ALARM.
The following launchpad bugs have fixes included:
- lp: 1786320, dbCa subscribes twice to ENUM
- lp: 541221,
assert (pca->pgetNative)failed in ../dbCa.c - lp: 1747091, epicsTimeGetEvent() / generalTime bug
- lp: 1743076, Segfault
in
ca_attach_context()during exits - lp: 1751380, Deadlock
in
ca_clear_subscription() - lp: 1597809, Setting NAME field in DB file may break IOC
- lp: 1770292,
get_alarm_double()inconsistent across record types - lp: 1771298, Conversion of NaN to integer relies on undefined behavior
Removed the settings for 2017; fixed the hour of the change for MET.
Initialize the first time-stamp from the first monitor, not the client-side current time in this configuration.
Windows builds using Visual Studio 2015 and later now use the -FS
compiler option to allow parallel builds to work properly.
We now give the -FC option to tell the compiler to print absolute
paths for source files in diagnostic messages.
The Posix implementation of epicsEventWaitWithTimeout() was limiting the
timeout delay to at most 60 minutes (3600.0 seconds). This has been changed to
10 years; significantly longer maximum delays cause problems on systems where
time_t is still a signed 32-bit integer so cannot represent absolute
time-stamps after 2038-01-19. Our assumption is that such 32-bit systems will
have been retired before the year 2028, but some additional tests have been
added to the epicsTimeTest program to detect and fail if this assumption is
violated.
This release adds several new make targets intended for use by developers and Continuous Integration systems which simplify the task of running the built-in self-test programs and viewing the results. Since these targets are intended for limited use they can have requirements for the build host which go beyond the standard minimum set needed to build and run Base.
The new make target test-results will run the self-tests if
necessary to generate a TAP file for each test, then summarizes the TAP output
files in each test directory in turn, displaying the details of any failures.
This step uses the program prove which comes with Perl, but also needs
cat to be provided in the default search path so will not work on most
Windows systems.
The new make target junitfiles will run the self-tests if necessary
and then convert the TAP output files into the more commonly-supported JUnit
XML format. The program that performs this conversion needs the Perl module
XML::Generator to have been installed.
The new make target clean-tests removes any test result files from
previous test runs. It cleans both TAP and JUnit XML files.
The attempt to fix DNS related delays for short lived CLI programs (eg. caget) in lp:1527636 introduced a bug which cased these short lived clients to crash on exit. This bug should now be fixed.
When a National Instruments network variables CA server is already running on a Windows system and an IOC or PCAS server is started, the IOC's attempt to bind a TCP socket to the CA server port number fails, but Windows returns a different error status value than the IOC is expecting in that circumstance (because the National Instruments code requests exclusive use of that port, unlike the EPICS code) so the IOC fails to start properly. The relevent EPICS bind() checks have now been updated so the IOC will request that a dynamic port number be allocated for this TCP socket instead when this happens.
Code has been added to the IOC startup to better protect it against bad
periodic scan rates, including against locales where . is not
accepted as a decimal separator character. If the scan period in a menuScan
choice string cannot be parsed, the associated periodic scan thread will no
longer be started by the IOC and a warning message will be displayed at iocInit
time. The scanppl command will also flag the faulty menuScan value.
Loading of database files has been optimized to avoid over-proportionally
long loading times for large databases. As a part of this, the alphabetical
ordering of records instances (within a record type) has been dropped. In the
unexpected case that applications were relying on the alphabetic order, setting
dbRecordsAbcSorted = 1 before loading the databases will retain the
old behavior.
The routine dbRenameRecord() has been removed, as it was intended
to be used by database configuration tools linked against a host side version
of the dbStatic library that is not being built anymore.
In addition to the more detailed change descriptions below, the following Launchpad bugs have also been fixed in this release:
- lp:1440186 Crash due
to a too small buffer being provided in
dbContextReadNotifyCache() - lp:1479316 Some data races found using Helgrind
- lp:1495833 biRecord prompt groups are nonsensical
- lp:1606848 WSAIoctl
SIO_GET_INTERFACE_LISTfailed in Windows
When using the Microsoft compilers a new build system variable is provided that
controls whether whole program optimization is used or not. For static builds
using Visual Studio 2010 this optimization must be disabled. This is controlled
in the files configure/os/CONFIG_SITE.Common.windows-x64-static and
configure/os/CONFIG_SITE.Common.win32-x86-static by setting the variable
OPT_WHOLE_PROGRAM=NO to override the default value YES that would otherwise
be used.
Note that enabling this optimization slows down the build process. It is not
possible to selectively disable this optimization, when building a particular
module say; Microsoft's linker will restart itself automatically with the
-LTCG flag set and display a warning if it is asked to link any object
files that were compiled with the -GL flag.
Dynamic array sizing support was added to the IOC server (RSRV) in the Base-3.14.12 release, but has not until now been supported in the Portable Channel Access Server (PCAS). Channel Access server applications using the PCAS may not need to be modified at all; if they already push monitors with different gdd array lengths, those variable sizes will be forwarded to any CA clients who have requested variable length updates. The example CAS server application has been modified to demonstrate this feature.
In implementing the above, the gdd method gdd::put(const gdd *) now
copies the full-sized array from the source gdd if the destination gdd is of
type array, has no allocated memory and a boundary size of 0.
The EPICS timestamp library (epicsTime) inside libCom's OSI layer has
been extended by routines that convert from struct tm to the EPICS
internal epicsTime type, assuming UTC - i.e. without going through
the timezone mechanism. This solves issues with converting from the structured
type to the EPICS timestamp at driver level from multiple threads at a high
repetition rate, where the timezone mechanism was blocking on file access.
The build configuration files that allow cross-building of the 32-bit
win32-x86-mingw cross-target have been adjusted to default to building shared
libraries (DLLs) as this is now supported by recent MinGW compilers. The 64-bit
windows-x64-mingw cross-target was already being built that way by default. The
configuration options to tell the minGW cross-compiler to link programs with
static versions of the compiler support libraries have now been moved into the
CONFIG_SITE.linux-x86.<target> files.
The iocInit code now performs a sanity check of the current time
returned by the generalTime subsystem and will print a warning if the wall-clock
time returned has not been initialized yet. This is just a warning message; when
a time provider does synchonize the IOC will subsequently pick up and use the
correct time. This check code also primes the registered event system provider
if there is one so the epicsTimeGetEventInt() routine will work on IOCs
that ask for event time within an interrupt service routine.
The osiClockTime provider's synchronization thread (which is only used on some embedded targets) will now poll the other time providers at 1Hz until the first time it manages to get a successful timestamp, after which it will poll for updates every 60 seconds as before.
The routine generalTimeGetExceptPriority() was designed for use by
backup (lower priority) time providers like the osiClockTime provider which do
not have their own absolute time reference and rely on other providers for an
absolute time source. This routine no longer implements the ratchet mechanism
that prevented the time it returned from going backwards. If the backup clock's
tick-timer runs fast the synchronization of the backup time provider would never
allow it to be corrected backwards when the ratchet was in place. The regular
epicsTimeGetCurrent() API still uses the ratchet mechanism, so this
change will not cause the IOC to see time going backwards.
The build configuration files for builds using the Microsoft compilers have been
updated, although there should be no noticable difference at most sites. One
extra compiler warning is now being suppressed for C++ code, C4344: behavior change: use of explicit template arguments results in ... which is gratuitous
and was appearing frequently in builds of the EPICS V4 modules.
Cross-builds of the windows-x64 target from a win32-x86 host have been
removed as they don't actually work within the context of a single make
run. Significant changes to the build configuration files would be necessary for
these kinds of cross-builds to work properly, which could be done if someone
needs them (email Andrew Johnson before working on this, and see
this stack-overflow answer for a starting point).
In preparation for moving to git in place of the Bazaar revision control system we have removed all the keywords from the Base source code.
Building this version of Base on a Linux system creates a systemd service file suitable for starting the Channel Access Repeater under systemd. The file will be installed into the target bin directory, from where it can be copied into the appropriate systemd location and modified as necessary. Installation instructions are included as comments in the file.
A new "getenv" device support for both the stringin and lsi (long string input) record types can be used to read the value of an environment variable from the IOC at runtime. See base/db/softIocExit.db for sample usage.
A new order-only prerequisite build rule has been added to ensure that
library files (and DLL stubs on Windows) get installed before linking any
executables, which resolves parallel build problems on high-powered CPUs. There
are some (rare) cases though where a Makefile has to build an executable and run
it to be able to compile code for a library built by the same Makefile. With
this new build rule GNUmake will complain about a circular dependency and the
build will probably fail in those cases. To avoid this problem the failing
Makefile should set DELAY_INSTALL_LIBS = YES before including the
$(TOP)/configure/RULES file, disabling the new build rule.
The IOC now sets a number of environment variables at startup that provide the
version of EPICS Base it was built against (EPICS_VERSION_...) and its build
architecture (ARCH). In some cases this allows a single iocBoot/ioc directory to
be used to run the same IOC on several different architectures without any
changes.
There are also 3 new environment parameters (EPICS_BUILD_...) available that
C/C++ code can use to find out the target architecture, OS class and compiler
class it was built with. These may be useful when writing interfaces to other
languages.
The mechanism behind the promptgroup() field property inside a record type
definition has been changed. Instead of using a fixed set of choices,
the static database access library now collects the used gui group names
while parsing DBD information. Group names should start with a two-digit number
plus space-dash-space to allow proper sorting of groups.
The include file guigroup.h that defined the fixed set of choices
has been deprecated. Instead, use the conversion functions between index number
and group string that have been added to dbStaticLib.
When a DBD file containing record-type descriptions is expanded, any
old-style GUI_xxx group names will be replaced by a new-style
string for use by the IOC. This permits an older record type to be used with
the 3.15.4 release, although eventually record types should be converted by
hand with better group names used.
RSRV now honors EPICS_CAS_INTF_ADDR_LIST and binds only to the provided list
of network interfaces. Name searches (UDP and TCP) on other network interfaces
are ignored. For example on a computer with interfaces 10.5.1.1/24, 10.5.2.1/24,
and 10.5.3.1/24, setting EPICS_CAS_INTF_ADDR_LIST='10.5.1.1 10.5.2.1' will
accept traffic on the .1.1 and .2.1, but ignore from .3.1
RSRV now honors EPICS_CAS_IGNORE_ADDR_LIST and ignores UDP messages received
from addresses in this list.
Previously, CA servers (RSRV and PCAS) would build the beacon address list using
EPICS_CA_ADDR_LIST if EPICS_CAS_BEACON_ADDR_LIST was no set. This is no
longer done. Sites depending on this should set both environment variables to
the same value.
libca, RSRV, and PCAS may now use IPv4 multicasting for UDP traffic (name search
and beacons). This is disabled by default. To enable multicast address(s) must
be listed in EPICS_CA_ADDR_LIST for clients and EPICS_CAS_INTF_ADDR_LIST for
servers (IOCs should set both). For example:
EPICS_CAS_INTF_ADDR_LIST='224.0.2.9' EPICS_CA_ADDR_LIST=224.0.2.9
Please note that no IPv4 multicast address is officially assigned for Channel Access by IANA. The example 224.0.2.9 is taken from the AD-HOC Block I range.
Since EPICS Base 3.15.0.2 on Posix OSs the initialization of the epicsThread
subsystem has called mlockall() when the OS supports it and thread
priority scheduling is enabled. Doing so has caused problems in third-party
applications that call the CA client library, so the functionality has been
moved to a separate routine epicsThreadRealtimeLock() which will be
called by the IOC at iocInit (unless disabled by setting the global variable
dbThreadRealtimeLock to zero).
When loading database files, macros get expanded even on comment lines. If a comment contains an undefined macro, the load still continues but an error message gets printed. For this release the error message has been changed to a warning, but even this warning can be made less verbose by setting this new variable to a non-zero value before loading the file, like this:
var dbQuietMacroWarnings 1 iocsh
dbQuietMacroWarnings=1 VxWorks
This was Launchpad bug 541119.
Dirk Zimoch provided code that allows the NTP Time provider (used on VxWorks and RTEMS only) to adapt to changes in the OS clock tick rate after the provider has been initialized. Note that changing the tick rate after iocInit() is not advisable, and that other software might still misbehave if initialized before an OS tick rate change. This change was back-ported from the 3.15 branch.
When a CA client gets data from an IOC record using a compound data type such
as DBR_TIME_DOUBLE the value field is fetched from the database in a
separate call than the other metadata, without keeping the record locked. This
allows some other thread such as a periodic scan thread a chance to interrupt
the get operation and process the record in between. CA monitors have always
been atomic as long as the value data isn't a string or an array, but this race
condition in the CA get path has now been fixed so the record will stay locked
between the two fetch operations.
This fixes Launchpad bug 1581212, thanks to Till Strauman and Dehong Zhang.
The 'make runtests' and 'make tapfiles' build targets normally only run the
self-tests for the main EPICS_HOST_ARCH architecture. If the host is
able to execute self-test programs for other target architectures that are being
built by the host, such as when building a -debug version of the host
architecture for example, the names of those other architectures can be added to
the new CROSS_COMPILER_RUNTEST_ARCHS variable in either the
configure/CONFIG_SITE file or in an appropriate
configure/os/CONFIG_SITE.<host>.Common file to have the test
programs for those targets be run as well.
An additional check has been added at build-time for the contents of the
configure/RELEASE file(s), which will mostly only affect users of the Debian
EPICS packages published by NSLS-2. Support modules may share an install path,
but all such modules must be listed adjacent to each other in any RELEASE
files that point to them. For example the following will fail the new checks:
AUTOSAVE = /usr/lib/epics
ASYN = /home/mdavidsaver/asyn
EPICS_BASE = /usr/lib/epics
giving the compile-time error
This application's RELEASE file(s) define
EPICS_BASE = /usr/lib/epics
after but not adjacent to
AUTOSAVE = /usr/lib/epics
Module definitions that share paths must be grouped together.
Either remove a definition, or move it to a line immediately
above or below the other(s).
Any non-module definitions belong in configure/CONFIG_SITE.
In many cases such as the one above the order of the AUTOSAVE and
ASYN lines can be swapped to let the checks pass, but if the
AUTOSAVE module depended on ASYN and hence had to appear
before it in the list this error indicates that AUTOSAVE should also be
built in its own private area; a shared copy would likely be incompatible with
the version of ASYN built in the home directory.
Two buffer overflow bugs that can crash the IOC have been fixed, caused by initializing a string field with a value larger than the field size (Launchpad bug 1563191).
The C++ interface to the epicsThread API could corrupt the stack on thread exit in some rare circumstances, usually at program exit. This bug has been fixed (Launchpad bug 1558206).
On RTEMS the NTP Time Provider could in some circumstances get out of sync
with the server because the osdNTPGet() code wasn't clearing its input socket
before sending out a new request. This
(Launchpad bug 1549908)
has now been fixed.
The bitwise operators in the CALC engine have been modified to work properly with values that have bit 31 (0x80000000) set. This modification involved back-porting some earlier changes from the 3.15 branch, and fixes Launchpad bug 1514520.
On process exit, don't try to stop the worker thread that makes DNS lookups asynchronous. Previously this would wait for any lookups still in progress, delaying the exit unnecessarily. This was most obvious with catools (eg. cainfo). lp:1527636
Simpler versions of the epicsTime_gmtime() and epicsTime_localtime()
routines have been included in the Windows implementations, and a new test
program added. The original versions do not report DST status properly. Fixes
Launchpad bug 1528284.