Changes since labca_3_7_2: Changes since labca_3_7_1: 2018/03/02 (TS): - tarball rules must include submodule (sciclean) - fixes (perlification) for non-cygwin windows builds Changes since labca_3_7_0: 2018/02/12 (TS): - Added Makefile rules to create released tarball Changes since labca_3_5_5: 2018/02/12 (TS): - removed mexLock from initialization code (breaks matlab 2017; see email thread with mathworks/J. Sebek, 12/9/2017) *** Note: changes that went into R3.6 were rolled back; there is no 'official' 3.6 branch/release. *** Changes since labca_3_5_3: 2017/08/18 (HS): - Removed RELEASE_SITE file and added it to .gitignore. Changes since labca_3_5_2: 2017/08/11 (TS): - replaced CVS $Id$/$Name$ in manual.tex and glue/ini.cc by git description strings 2017/08/11 (TS): - fixed manual.tex (warnings due to new html.sty) 2017/08/11 (TS): - removed CVS $Id$ keywords 2017/08/11 (TS): - added test case for recent bugfix 2017/08/10 (TS): - merged changes from labca_3_5_branch Changes since labca_3_5_1: 2017/08/10 (TS): - use psed/sed everywhere (SED defined in configure/CONFIG) Changes since labca_3_5_0: 2017/08/08 (TS): - use sed on linux, psed on windows (perl 5.22 stopped shipping psed) 2017/08/08 (TS): - BUGFIX (thanks to Jesus Vasquez for reporting). When retrieving DBF_LONG in native format, result was inadvertently converted to float and then promoted to double which lead to a loss in precision. 2015/09/08 (TS): - fixed test script (didn't catch all partial enum string mismatch) - fixed test script (matlab compatibility) Changes since labca_3_4_2: 2015/09/08 (TS): - replaced #include by - replaced obsolete TS_STAMP by epicsTimeStamp (TS_STAMP is a macro for the latter under reasonably recent base) 2015/06/04 (TS): - RENAMED lcaGetEnumStates -> lcaGetEnumStrings 2015/06/04 (TS): - matlab/lcaGetEnumStates.c, matlab/Makefile: added lcaGetEnumStates() - matlab/lcaDebugOn.c, matlab/lcaDebugOff.c, matlab/Makefile: added lcaDebugOn()/lcaDebugOff(). - ezca/ezca.c, ezca/ezca.h: added ezcaGetEnumStates() - glue/labca.sce.in, glue/ezcaSciInt.c: changed names from ezca->labca. Added helper to unload labca. Added lcaGetEnumStates() - testing/lcaTest.sce, testing/lcaTest.m: added tests for lcaGetEnumStates() 2015/06/03 (TS): - Reorganized ezcaGetPrecision(), ezcaGetUnits(), ezcaGetNelem(), ezcaGetStatus(), getLimits() (which is now getInfo()) to share more code and eliminate duplication (of which there is still plenty). - added compile-time test to check consistency between CA size macros and EZCA's Changes since labca_3_4_1: 2015/03/12 (TS): - Added top-level rule to uninstall everything but documentation (for prep. of distribution tarball). - Silenced compiler warnings (changed format string arg of lcaSetError() to 'const char*') - Applied fixes for matlab 2014a contributed by Matthew Furseman. Changes since labca_3_4: 2013/06/25 (TS): - FIXED: build problems due to changed scilab libraries when going from scilab-5.3 to 5.4. Thanks to Takashi Obina for reporting and suggesting a fix. - documentation/xmlhelp/en_US/Error.xml no longer passed through xml_to_jar. Removed and replaced by . Changes since labca_3_3: 2012/01/13 (TS): - glue/ezcaSciInt.c: removed epicsShareAPI for C2F(ezca). This caused a leading underscore (when building w/o .def file as we now do) which is not understood by scilab's addinter. - glue/Makefile: (WIN) add dependency libraries to TARGET_LDFLAGS instead of LIB_LIBS + _DIR. This works around problems if the library path contains blanks. - README.win32pathwithblanks: updated info. - documentation/Makefile: use portable workaround for mkdir -p (w/o causing errors if target dir is already present). 2012/01/12 (TS): - glue/Makefile: re-enabled mx_DIR + friends (non-windows only) - matlab/lcaSetMonitor.c, matlab/lcaNewMonitorWait.c, matlab/lcaNewMonitorValue.c: removed unused 'i' variable. - matlab/Makefile: enable warnings for MEX compiler (linux). - testing/Makefile: always build lcaTest.m (even under scilab) so that it is present for distribution. - ReleaseNotes: updated. - glue/Makefile: o win32/64 scilab-5 needs 'output_stream' and 'intersci' libraries in addition to LibScilab :-(. Manual intervention is needed to build for scilab-4 (define SCILAB_V4 makefile variable). o Commented unused library paths (libmx_DIR + friends). o Use 'psed' (perl) instead of 'sed' and use double quotes for sake of win-non-cygwin portability. o Use 'vpath' instead of symlink to find sciclean.c for sake of win-non-cygwin portability. 2012/01/11 (TS): - glue/ctrlC-polled.c: scilab-5 seems to properly export + declare 'basbrk' (even under windows) so we use it. - glue/multiEzca.c: silence compiler warnings (64-bit?) by using ptrdiff_t instead of int and explicit type conversions where appropriate. - glue/mglue.c, glue/mglue.h: silence compiler warnings (64-bit?) by using size_t instead of int where appropriate. - documentation/Makefile: removed run-time test for scilab version (too hard to get to work under windows). We just in stall the scilab-5 help files anyways. - configure/RELEASE: removed CONFIG-SLAC hacks - doesn't work as intended. - configure/CONFIG: added defaults for 'windows-x64' - configure/RULES: avoid complicated shell commands (which windows command.com is unable to unerstand). 2012/01/11 (TS): - ezca/ezca.c: silenced compiler warnings (signed/unsigned, size_t printf format). 2010/08/23 (TS): - Changed SED commands that transform lcaTest.sce -> lcaTest.m so that the script is (again) properly transformed for matlab. Thanks to Steph Allison for pointing out problems... Changes since labca_3_2_beta: 2010/06/04 (TS): - Added (partly hand-crafted) help files for scilab-5. 2010/05/28 (TS): - IMPROVEMENT. ezcaGetUnits(), ezcaGet_xxx_Limits(), ezcaGetPrecision(), ezcaGetStatus() used to transfer the native element count even if the PV value itself was discarded. Modified ezca.c so that only 1 element is requested by these calls. This enhances speed with slow connections :-). - several small changes to support scilab-5.2. Some header install locations have changed since 5.1; __PARAMS() macro has gone. 2009/10/13 (TS): - several small changes to support scilab-5.1. (Scilab-5 is completely re-vamped.) TODO: Since scilab-5 help files must be in their XML format (converted to javahelp jars though) we need to re-write the documentation, breaking out individual commands into some 'meta' format from which XML and tex can be generated. HELP DOES NOT WORK YET on scilab-5! Changes since labca_3_1: 2008/5/21 (TS): - added ezcaGetWarnLimits() and ezcaGetAlarmLimits() and the respective interfaces from scilab + matlab. Changes since labca_3_0_beta: 2007/9/5 (TS): - wrapped all malloc/calloc/realloc/free calls into 'lcaMalloc, lcaCalloc' etc. to help debugging memory issues. - special scilab cleanup code because many scilab sanity tests are macros which return from a subroutine w/o giving us a chance to clean up (see glue/sciclean/). 2007/9/5 (TS): - fixed memory leak (lcaPutNoWait) in ezca (copy of data made by ezcaPutOldCa() was never free()ed when ezcaPutOldCa() calls were grouped). Thanks to Greg Portmann for helping hunting this down. 2007/8/29 (TS): - fixed (some) memory leaks (scilab only). Scilab macros (stack-c.h) would jump out of our routines w/o giving a chance to clean things up. Use the new 'sciclean' facility to register cleanups as necessary. Also, C-string matrices (GetRhsVar(...,"S",...)) were never free()ed! - removed redunant printing of error/warning messages. This should fix a complaint that errors were still reported even from a try-catch-end block. 2007/6/19 (Jim Sebek) - extended fix that was introduced (4/07) to work around problems with mcc compilation due to .m files residing in the same directory as .mex files. Now both, the labca initializer and finalizer are skipped when invoked during mcc compilation. Apparently this solves the problem. Changes since labca_2 2007/6/10 (TS): - call _exit() from the finalizer if called from the caRepeater fork must not do any other cleanup since the forked process has no thread. This fixes the 'starved caRepeater fork' problem (hanging process would never die). 2007/6/7 (TS): - completely reworked the finalizer. Good shutdown now seems to work w/o hangs or crashes for scilab and matlab on linux/solaris + win in gui and nogui mode. 2007/6/4 (TS): - ezcaDelay() now polls the Ctrl-C flag like the other calls. - added EZCA_ABORTED return code to flag user abort (CTRL-C) condition. 2007/6/1 (TS): - Some changes to CONFIG/RELEASE/RULES for automatically finding the correct mex-file extension. Also, EPICS_HOST_ARCH-specific settings for the EPICS_BASE, MATLABDIR + SCILABDIR variables were implemented. 2007/6/1 (TS): - stupid win32 doesn't implement c99 types :-o Resort to 'epicsTypes.h'... 2007/5/23 (TS): - Added lcaLastError() to retrieve the last error that was recorded. The usual scilab/matlab error reporting scheme is not good enough because I cannot report a vector of errors. If a lcaXXX operation fails on a subset of PVs this routine can report the error status for individual PVs (to be used like and possibly in addition to 'lasterror' in scilab/matlab). - Added multi-ezca interface to lcaLastError and wrappers for scilab and matlab. - Added multi-ezca interface to ezcaNewMonitorWait and wrappers for scilab and matlab. - Added ezcaNewMonitorWait() which blocks until new monitor data becomes available. This routine is groupable. - Changed implementation of EzcaPendEvent(). There is really no need for ca_pend_event() anymore with multithreaded CA :-). This means that we can keep a count of outstanding operations and let EzcaPendEvent simply ca_flush_io() and block on a semaphore (occasional timeout so that we can poll the X loop and check for Ctrl-C). As requests are answered a CA background task executes the callbacks which decrement the 'outstanding' counter and eventually post to 'done' semaphore. --> lcaXXX calls NO LONGER take multiples of ezcaTimeout to complete :-) but complete as soon as possible (w/o having to resort to a high CA poll rate). 2007/5/22 (TS): - labca now 'throws' (ezca) error codes when an error occurs (via mexErrMsgIdAndTxt / Scierror) that can be picked up by the application via catch + lasterror. - lcaNewMonitorValue() now 'throws' errors for all cases that in previous versions returned < 0 flag values. I.e., if lcaNewMonitorValue() completes successfully then it returns either 0 or 1. Changes since labca_2_1_beta (T.S): 2007/5/9 (TS): - use c99 types int8_t, int16_t, int32_t rather than char, short, int. (multiEzca.c; see also change made 2006/9/1). 2007/5/8 (TS): - lcaPutNoWait() would not flush out commands (thanks to Greg Portman for reporting). 2007/4/11 (TS): - scilab got stuck when initializing labca in non-windowing mode (-nw). This was because we still checked the X loop in ctrlC-polled.c. Now we use checkevts_() to find out if we have should work the X loop. 2007/4/10 (Jim Sebek): - attempts to mcc-compile matlab scripts that use labCA had failed apparently because the help (.m) files are in the same directory as the respective mex object files. Jim found a work-around that tests if a compilation is in progress and which skips labca initialization if this is the case. 2006/12/19 (TS): - mexLock() glue and CA libraries so that EPICS is not shut down as a result of matlab 'clear' operations [epics shutdown -> exit()]. 2006/9/1 (TS): - ezcaLong is really 32bit, i.e., I fixed multiEzca.c to use 'int' instead of 'long' which is correct on 32bit and 64bit platforms. Changes since labca_2_0_alpha (T.S): 2006/4/14 (TS): - scilab now also uses polled 'ctrlC' handling (undocumented 'isbrk() and inibrk()' routines. - changed shutdown procedure. This is still a pain and I don't understand exactly what needs to be done to avoid crashing the app on exit. Current solution seems to work on all platforms for both scilab and matlab. 2006/4/11 (TS): - upgraded to 3.14(.8.2)-style Makefiles. 2006/4/10 (TS): - ezcaEndGroupWithReport() used to clear out unconnected channels assuming that the disconnected state was a result of a failing search_and_connect(). We now clear only channels that never were successfully connected and leave other ones alone. This allows monitors to pick up seamlessly after IOC reboots. - disallow implicit mapping of native type -> FLOAT for monitors. (The old behavior let disconnected channels look for a FLOAT type monitor. We now let lcaNewMonitorValue() report -10 if asking for the native type of a disconnected channel). 2006/2/17 (TS): - we now use (undocumented) MATLAB 'utHandlePendingInterrupt()' to poll for pending 'Ctrl-C' rather than hacking into the windows event loop. Our hack didn't work with matlab 7.2(?)/win32 anymore. 2005/6/7 (TS): - flush output buffers (lcaPutNoWait()) Changes since labca_1_5 (T.S): 2004/3/24 (TS): - EZCA modifications: o try to recycle work nodes marked as 'trashme'. o assume ca_clear_channel() does real cleanup (i.e. assume it cancels scheduled callbacks) and hence recycle channel structures from push_channel(). Also, have push_channel() remove deleted channels from the hash table. o call clean_and_push_channel() if search attempts timeout: probably, the PV name is invalid and we want to prevent 3.14 background tasks to continue searching. o provide 'ezcaClearChannel()' and 'ezcaPurge()' to let the user explititely remove unneeded channels. - provide 'lcaClear()'. - Provided 'lcaPutNoWait()' for setting values without waiting for record processing to complete on the server. Changes since labca_1_5pre_beta (T.S): - set shared library version numbers in anticipation of future releases. Changes since labca_1_4_beta (T.S): 2004/3/8 (TS): - Have the matlab version call 'errMsgTxt()' on error to trigger the standard way of error flagging. I finally found the reason for the linux crashes (matlab ships with outdated system libraries) ;-) - Implemented a workaround in ezErr() -- Scilab's mexPrintf() [i.e. sciprint] crashed when munging long strings (multi-line errors). We now break those messages up in smaller chunks... - multi_ezca_put() now returns a value to indicate error status (rval == number of successfully written PVs, -1 on error). Changes since labca_1_3_beta: 2004/2/20: T.S added matlab implementation of lcaGetStatus() 2004/2/20: T.S added this ChangeLog