Release Notes for APS Release 3.11.6
			May 11, 1994

New and/or Updated Documentation:

EPICS Application Source/Release Control
EPICS Add On Products Source/Release Control
Channel Access Security Design
Archiving Reference Manual
Channel Access Reference Manual
Application Developer's Guide
Synchronized Time Stamp Support

A. Directions for switching to production APS release R3.11.6

    NOTE:	IOC Memory

		This release will use more memory on your iocs than previous
		releases (mainly because of access security). John Winans
		has modified the board support package to support > 8 MegByte
		memory boards. See accessSecurity Document for details.

    PART I.	All developers and users

		Add the following early in your .cshrc file:

		    source /usr/local/epics/cshrc

		This file creates definitions for Unix environment variable
		EPICS_ADD_ON and changes your path so that it is searched in
		the order: .epicsUnix, $PVT_ADD_ON, $EPICS_ADD_ON, $path.
		If you are in an epics application area the epics release bin
		directory will be searched first, then PVT_ADD_ON (if you have 
		defined this variable), then the epics add on area, then the
		previously defined path.
    PART II.	Application System Area Managers
           	Follow the directions given in the Application Source/Release
		Control document, Chapter 3 - Procedures for Application System
		Area, Section 4 - Integration.

    PART III.  	Application Shadow Area Developers
               	The following directions are identical to the direction
               	given in the Application Source/Release Control document
               	Chapter 4 - Shadow Node Procedures, Section 2 -
		Synchronizing an application shadow node.

    PART IV.	Changes to Record Support

		All ascii record definitions have to be changed to add
		the access security level to each field description. The
		following script can be used to perform the conversion
		(It assumes that the files are named <type>Record.ascii)

		#! /bin/sh
		for file in *Record.ascii
		ex - $fdile <<\EOF
		1,$s/DBF_/ASL1 DBF_/

		The source file for each record support module must be edited
		to change all calls of the form




B. Unbundling of Channel Access Clients

	The channel access clients as well as GDCT have been unbundled
	from epics. See the EPICS Add On document for details

C. Access Security

	Access Security is now implemented.
	See the Channel Access Security Document for details

D. Channel Access Additions

		Caching and Blocking Channel/Database Access

	The requirements resulted from discussions between
	Jeff Hill, Claude Saunders, and Marty Kraimer.

	We agreed that the following changes should be make to channel access
	and/or database access:
	1) Cached db_puts. 
	2) Locked db_gets
	3) ca_puts with notification of record processing completion.
	4) Allow tracking of multiple batches of synchronous Channel Access Requests.

		Cached  database puts

	Previously the following was true. If
	1)a dbPut is directed to a passive record
	2)the put is supposed to cause the record to be processed.
	3)The record is already active (PACT is true)
	then the request is just ignored.

	Changes were  made so that the following is true
	1) If a dbPutField, which is a request from outside the database such
   	as from Channel Access, is directed to a field with Process Passive
   	True and the record is already active, then, after the record finishes
   	processing, it will again be processed.
	2) If a dbPutLink with Process Passive true is directed to a record
   	that is already active because of a dbPutField request to that
   	record then after the record finishes processing it will again be

		Locked db_gets

	Previously a record was not always locked while data is fetched.
	If an array record is being read this could result in part of the data
	coming from one scan and the remainder from the next scan.
	Records are now locked during channel access gets.

		ca_puts with confirmation of record processing.
		Batched synchronous Channel Access Requests

	Claude requested that channel access clients be able to receive
	notification of completion from ca put requests.  By completion it
	means that if the put results in a request for record processing
	(the record is passive and the field is PP) then a completion
	message is generated only when that record and all records processed
	as a result of that record being processed complete processing.

	In addition Claude has requested that blocking 
	for completion of a group of put requests be possible.
	This request is satisfied by a new CA IO synchronization 
	facility. The new facility allows the programmer to create
	any number of synchronization groups. The programmer can
	interleave IO requests within any of the groups. Once The
	IO operations are initiated then the programmer is free to
	block for IO completion within any one of the groups as needed.

E. Synchronous Time Support

	New software has been added to IOC core to maintain time stamps.
	The new software has the ability to maintain time stamps over all
	IOCs on a network. This software replaces the previous ioc
	time stamp support. The main differences are that the new version
	support multiple hardware boards and allows multiple time stamp
	events. See the manual listed above for details.

F. Description of major differences between  R3.11.3 and R3.11.6

    1. IOC Software Components

		Added definitions for READ_ACCESS and WRITE_ACCESS

		The ARR_kbd and AR_kbd programs are removed.
		Use ARR_cmd and AR_cmd.

	    replaced gcvt by sprintf with e format.
	    cvtFast.h - removed support for non-ansi prototypes
	    The ExpToString routines now just call sprintf. Bug for .1 fixed

	    removed iocFloatLib.c

            It is now ok if DTYP does nor have associated INP or OUT
            Check for record names too long.
	    Added dbReportDeviceConfig

	    Added dbhcr (Hardware Configuration Report)
	    To use type (on ioc)
		dbhcr > report
	    Then use the Unix sort command
		sort < report > reportSort

	    dbPutLink no longer calls recGblRecordError (error message storms)

	    Now allows callback to call taskwdRemove

	    Fixed bug that caused message to be suppressed
	    if status=0 and errno=0.
	    Now calls logMsg instead of printf.
        Scan Tasks
            added scanOnce.
            If scanAdd fails set precord->scan=SCAN_PASSIVE
            Fixed bug that causes failure if first record in database
            	has 28 character name.
	    Will only call taskwdRemove if restarting
	Analog Output Record
	    Added AOFF and ASLO

	Binary Output Record
	    When the binary output record support module processes the record
	    because of the HIGH field, it calls dbProcess rather than the
	    process entry itself. This if the record becomes disabled between
	    the time VAL is set to 1 and the time specified by HIGH occurs then
	    the record will not be processed. Before it would get processed.

        Event Record and soft device support.
            Made the VAL field a prompt field. The constant link value is 
            now moved to the VAL field only if VAL is zero.
        Histogram Record
            Fixed bug in callback mechanism
        Calc Record
            Modified postfix.c so it returns a non zero value when it can
            detect improper usage of a comma in a calc record expression.

	    Added support for constants CONST_PI CONST_R2D and CONST_D2R
	    ( PI, Deg to Radians, Radians to Degrees)

	    Calc record now allows usage of lower case anywhere in expression.

	Sequence Record
            Added monitor release code for the values that are fetched and
        then put when the links are processed.

	compress record

	    Fixed th updating of the inx field for the AVERAGE algorithm

            Float/Double conversions to string were incorrect for value>1e7.
        Allen Bradley IXE Millivolt Input
            This was not correctly performing linear conversions. It is
	Hardware Error Messages and Error Message Storms:
	    A problem arose because Allen Bradley Analog Outputs caused
	    error messages to be generated every time a record was processed
	    if the Analog Output did not exist or if the Adapter was down.
	    A check was made of many of the existing device support modules.
	    They were changed to implement the following algorithm:
	    1) If a hardware error was detected during record processing the
		associated record is put into alarm. This was true before.
	    2) If errVerbose is TRUE and the alarm status and/or sevrity is
	  	being changed then an error message is issued.
	    NOTE: errVerbose has the value False. It can be set true by issuing
		the vxWorks shell command:
		This can be done in an ioc startup file just before iocInit or
		at any other time.
	    The device support modules that use this new algorithm are:

            Replaced itob by cvtLongToString
        Subtool, dbLoadRecords, and dbLoadTemplate
            Fixed a bug in the variable substitution software.  Subtool,
            dbLoadRecords, and dbLoadTemplate had a buffering problem.

        VXI resource manager
            A bug was fixed in the VXI resource manager. The bug only effects
            systems with multiple VXI crates interconnected by the MXI bus.
            The resource manager was allowed to discover the presence of another
            MXI bus repeater while searching for devices in a MXI bus repeater. 
            This resulted in a bus addressing conflict. The solution was to
            search for and discover the presence of all devices on the MXI bus
            prior to opening the LA window of the first MXI found and searching
            for devices underneath of it.

        Bitbus driver
            Both the PEP and Xycom bitbus drivers were updated such that
        a possible race condition is eliminated in WdTask.

    2.  Channel access bug fixes

        A 32k memory leak was discovered in the CA client library. The leak 
        occurred each time that a communication linkl with an IOC was lost.

        A bug was fixed in the vxWorks version of the CA client library
        relating to connection cleanup. Symptoms of the problem are
        bus errors when a vxWorks CA client exits.

        It was observed that the CA server would stop accepting new
        CA client connections if an IOC reached its maximum file 
        descriptor limit even if steps were taken to close exiting CA client 
        connections to free up exiting file descriptor allocations. The
        bug was fixed.

        A <50 byte memory leak was fixed in the CA server. The leak occurred 
        each time a client disconnected.

    3.  Application Source/Release Control

        makefile (top)
            Changed logic -  The three major invocations are:
            command:    make
                just invokes make in each application directory
            command:    make world
                invokes doGets to get any out-of-date sccs controlled files
                then invokes makesdr  to bring sdr files up-to-date
                then invokes make world in each application directory
            command:    make dotar
                builds a compressed tar file of this whole structure
            command:    make tarinfo
                explains how to unpack the compressed tar file.
        makefile (app)
            Changed logic -  The two major invocations are:
            command:    make
                just does makes in the target arch, Db's
                and if present invokes makefile.pvt
            command:    make world
                invokes doGets to get any out-of-date sccs controlled files
                then rebuilds Makefile(s) from Imakefile(s)
                then invokes make on the rebuilt Makefile(s)
                then invokes make in each <*>Db directory
                then invokes make -f makefile.pvt (if present )
        makefile (dct)
            Makefile has been changed to use default.dctsdr, default.sdrsum
            and <file>.rpt dependencies to determine when to create
            <file>Db.database.  It has been changes to use atdb/dbta instead
            of DCT to create the database and the short form reports. The
             makefile now also allows usage "make <file>Db.database" to make
            only one database.
        makefile (gdct)
            This makefile does nothing unless there is a makefile.pvt present
            in the same directory. It is invoked by makefile.gdct

            Makesdr has been changed to use dependencies on <epics>/share/
            ascii/*.ascii, cat_ascii/*.ascii, rep_ascii/*.ascii and recursively
            on any files included by any of the ascii files to determine
            when to create default.dctsdr and default.sdrsum.
        apSccsScan, doGets
            apSccaScan has been replaced by new tool doGets which
            performs the same function as apSccsScan and now works
            recursively from any directory in which it is invoked.
            New file to convert DCT reports to atdb/dbta format.

            Now puts .epicsUnix links in all ioc directories.

            Reworked to allow creation of a plain Db directory.

            New file for list of target architectures for the build
            objects in an application.

            Changed to use archList for list of target architectures to 
            be built.

            Corrected handling of makefile.pvt.

            Added op/doc directory to application template directory 

            Added epicsH/rec to INCLUDE path.
            Changed default definintion of SNCFLAGS to null n the ImakefileVx
            Added rules MakeObjRuleForStateCpp and MakeObjRuleForStateCppFlags
            for using the c pre-processor, cpp, prior to the state notation 
            compiler, snc.

            Added dbsccs create and replaced DCT usage with atdb/dbta calls.

    4. Unix Tools

            Added startup file selection boxes for log files and changed alh 
            to use dbDefs.h to define pvname and field name lengths.

            For CT_HEX fields a leading 0x is now printed.
            Added the ability to put substitution variables into ".db" files
            using gdct.  Gdct now allows substitution variables to be put in
            to any dct editable field.
            The snaps program has been modified to use the file description
            manager to control the pending events, all the input data
            channels are monitored, the timestamp is also returned for the
            monitored channels, the time scale for the plot with accuracy to
            milli-seconds for pulse mode.
            Removed the restriction of putting non null string array to IOC.
            Fixed the problem of entering multiple put enum waveform records.
            Allowed an option of automatic flush pend events before checking
            for wait events. Monitored channel also returns timestamp.
            A new function CaWaitFlush is added to CaMath which allows the
            user to set automatic event flush to on or off before performing
            the check event.
            A new tool, xmsnaps, will be available to EPICS users in production
            release R.3.11.4. Xmsnaps is an integrated Motif application which 
            can provide a view of the most current history for a number of 
            monitored process variables(PVs). Any changes in the monitored 
            process variable values will be automatically captured by the 
            snaps program. It uses an adjustable circular buffer to store the 
            most current snapshots of the moitored process variables.
            Xmsnaps can operate in three different modes: pulse mode, periodic 
            scannning mode, and periodic event-driven scannning mode.  It can 
            snap the values of the monitored PVs, export captrued data to 
            disk files, display text data files, and plot the results.

            The user guide for "xmsnaps" is available in the user drawer of 
            ApsDoc Interleaf cabinet.

    5.  Motif-based Editor/Display Manager

            a private colormap can be specified for use in MEDM (complete
            with flashing) so that MEDM can run even when color-greedy
            applications like PV-WAVE run.   this option is enabled by a
            command-line option  "-cmap".   the default is not to use a
            private colormap...
            trigger channels for Cartesian plots has been implemented and
            is available for testing.  existing (non-triggered) plots should
            work as they currently do, but, by specification of a trigger
            channel PV name, plot visual updates occur only when the trigger
            channel changes.   this can be used to throttle the visual
            updates to some modest rate, although the internal data updating
            will occur at whatever the plotted channel data rate is...
            The EPICS_DISPLAY_PATH environment variable can now use
            colon-separated format for multiple search directories. In other
            words, setting the EPICS_DISPLAY_PATH environment variable to a
            string of the format a:b:c:d...  will result in file access
            attempts in the directory order specified, and, upon successful
            access, will use the first accessible directory/file pathname.
            The macro-substitution mechanism in place for related displays
            has been extended for "top level" displays.  A new command line
            option  "-macro"  has been introduced which allows passing of
            a string of macro substitutions for display startup time, to
            be applied to the specified displays on the MEDM command line.
            This option is only honored for EXECUTE mode MEDM startup (-x)
            and applies only to command line specified displays.

            The form of the option is:   -macro "a=b,c=d"
            with the double quotes being a required part of the syntax,
            and, again is only honored if -x is also specified.

            This option also applies for remote startup of displays.
            Hence, if an instance of MEDM was started in EXECUTE mode to
            a display, and another instance is invoked via

            % medm -x -macro "a=b,c=d" macro1.adl macro2.adl macro3.adl
            then the display requests for macro[123].adl  will be forwarded
            to the remote MEDM along with macro substitution information
            for creation in that context.
          MAIN MENU
            de-sensitize Edit pulldown menu and File->New,Save,SaveAs 
            buttons when in execute mode (prevents saving in execute mode,
            for instance)
        -CMAP (private cmap) changes
            the use of BlackPixel(dpy,scr)/WhitePixel(dpy,scr) by xrt/graph,
            and the xc/ derived widgets causes problems with the private
            cmap. the private cmap now allocates a black and white pixel
            which maps to the same value as in the default colormap, hence
            circumventing the problem for Pseudo-Color visuals...  a fix was
            also introduced into the xc/ widgets to properly get colors out
            of the widgets colormap not the default colormap
            Controller Object updates now allowed while cursor is in

            The handling of controller objects as simultaneous monitors
            has been changed to guarantee accuracy with underlying channels
            (within the limits of CA's monitor mechanism).
            All controllers now will update regardless of presence of
            cursor in that object. The previous "enter-don't-update" feature
            was introduced to make valuators predictable to use and was
            applied to all controllers.  In retrospect this was not
            necessary or sensible for "discrete" controllers like message
            buttons, etc.

            Controllers now update even when the cursor is within the
            confines of the controller object.  However, for valuators, the
            updates are disabled during DRAG operations, and re-enabled
            upon conclusion of DRAG.  This behavior is appropriate for
            the valuator and gets around earlier problems.  Note that
            DRAG is initiated by selecting the slider AND MOVING IT; merely
            selecting the slider via MB1 will not disable updates until
            it is also moved.
            remove <Key>space translations for push buttons/radio box/toggle
            buttons (per Ned - to prevent unintentional pushing of buttons,
            e.g., after screen saver kicks in)
            alarm modified colors for choice button, message button and
            menu now consistent with all other monitors/controllers
            MEDM now carefully honors file permissions for display files
            under SAVE operations.  Various combinations of directory and
            file permissions are possible and are now correctly handled.
        BUG FIXES:
            bug fix in valuator expose/redraw mechanism
            properly handle window parent chaining mechanism at root window

            the "LinacSequence.adl" maximize problem has been fixed
            the "cannot allocate colors" colormap/X device-ran-out-of-colors
                problem has been fixed
            the "realloc-ing input fd array" problem has been fixed
            don't do monitor list traversal if in EDIT mode
            add a single caHeartBeat() timer proc in main to handle misc. CA
            processing (remove code which added an XtTimerProc caHeartBeat()
            for each transition to EXECUTE mode in modeCallback -- this was
            clearly a bug and probably is responsible for problems when
            people do lots of Edit <-> Execute  transitions...)
            An alarm color mode problem in private colormap (-cmap)
            instances of MEDM has been identified and fixed.
            Controllers as Monitors:
            There was a bug in the monitor list traversal (updateMonitors.c)
            code that treated controllers as monitors, including not doing
            updates when (data->value == data->displayedValue).  Note that
            controllers are different than strict monitors in this regard,
            because the user can change the visual state of the object
            such that data->displayedValue is not accurate.
            E.g., for choice buttons, the user can depress a button which
            sets a toggle which initiates a ca_put() which eventually
            gets back to the monitor and updates its state.  But if a
            another mechanism updates that state rapidly such that the
            traverseMonitorList() routine doesn't get around to the update
            in time, the value can be reset to its old value which
            is inconsistent with the visual state, yet the visual state
            is not updated since (data->value == data->displayedValue).
            This problem has been fixed, and controller's monitors are
            keyed off of data->modified alone which should guarantee
            the correct visual state regardless of user interactions.