MERGE supports copying a global into another global with different collation properties (alternative collation, numeric collation, null collation). Previously, in this circumstance, MERGE inappropriately stored the source global subscripts into the destination global without doing the needed subscript transformations.(GTM-5573)

Specifying collation with a missing or improperly specified collation library gives an appropriate error. Previously this resulted in a segmentation violation (SIG11). [UNIX] (GTM-5576)

GT.M defers the processing of signals, such as SIGTERM, while flushing dirty database buffers. Previously, a signal delivered to a GT.M process within a small time window during the buffer-flushing operation could leave shared resources in a state that delayed other GT.M process from making further progress for up to several minutes. FIS encountered this issue in testing and has not received a customer report of such an event. [UNIX] (GTM-7279)

GT.M handles concurrency among GT.M processes performing global buffer flushing and buffer pool recovery correctly. Previously, such concurrency could lead to out of design situation causing a GTMASSERT error. This problem caused a process loss, not database damage. Note that buffer pool recovery is an unusual event so such a concurrent event is quite unlikely. [UNIX](GTM-7600)

GT.M checks if DSKNOSPCAVAIL is configured in the custom errors file when handling out-of-space conditions. Previously, if any custom errors file was configured, GT.M always behaved as if DSKNOSPCAVAIL was in the custom errors, and indefinitely retried out-of-space conditions. [UNIX] (GTM-7645)

A replicated database access can fail with REPLINSTNOSHM instead of REPLINSTMISMTCH if the journal pool is unreachable. Previously, such a database access could fail with REPLINSTMISMTCH if the journal pool id was missing or invalid in the replication instance file. In this case, the REPLINSTMISMTCH error message displayed the instance file name as an empty string, and journal pool shared memory id as 4294967295. [UNIX] (GTM-7649)

GTM performs correct update serialization in MM mode on multi-CPU systems. Previously, GT.M did not protect against out-of-order memory accesses by some pSeries model hardware that occasionally produced inconsistent database states. This condition did not affect BG mode databases, appeared only in the GT.M development and test environment, and has not been reported by a customer. This change also improves throughput on MM mode databases. [AIX] (GTM-7756)

GT.M and its utilities more cleanly maintain the mechanism used to manage database critical sections. Previously a terminating process could inappropriately decrease the number of available entries in a queuing structure within the critical section logic and eventually cause less efficient critical section management. Also, MUPIP SET -MUTEX_SLOTS can be set between 64 and 32768. Previously, the lower limit was 1024. The default value of 1024 should be appropriate for most situations and FIS recommends changing it only in consultation with FIS GT.M Support. (GTM-7804)

Name-level $ORDER() appropriately deals with an extended reference argument. Previously, it ignored the extended reference. In addition, $VIEW("REGION",gname) always uses $zgbldir. Previously it improperly used the global directory used by the last global reference, which could be inappropriate if that was an extended reference. Finally, $REFERENCE includes any extended reference after a Naked Reference. Previously, a Naked Reference after an extended reference operated properly but removed the evidence of its target global directory from $REFERENCE. (GTM-7810)

GT.M appropriately handles a MUPIP STOP received by a process within a small window while processing a TRESTART while an performing an update involving a trigger. Previously, such an unlikely circumstance could cause the process to terminate with a segmentation fault (SIG-11). FIS encountered this issue in testing and has not received a customer report of such an event. This item was actually addressed in V6.0-003 but was missed in that release note, and so is included here for those who no longer have cause to look at revisions of the V6.0-003 release note.[UNIX](GTM-7811)

The TRIGNAMEUNIQUE error message correctly states that the maximum number of unique run-time trigger names is 3907. Previously it stated that this number was 3844. [UNIX](GTM-7825)

When run as root, GT.M components use the owner and group of the database file as the owner and group of newly created journal files, backup files, snapshot files, shared memory, and semaphores. In addition, they set the permissions on the resulting files, shared memory, and semaphores, as if running as the owner of the database file and as a member of the database file group. Note that FIS recommends against running GT.M components as root. Previously, a process running as root set the owner to root, and the group and permissions as for a normal (non-root) user. [UNIX](GTM-7846)

Replication filters, operating on a source or a receiving instance, apply global nodes containing null subscripts correctly to database regions with Standard Null Collation enabled. Previously, such nodes were stored with GT.M Null Collation which effectively made them inaccessible.(GTM-7854)

GT.M replication correctly applies global nodes containing null subscripts when the source and receiver sides have different null collation settings (Standard Null Collation vs. GT.M Null Collation). Previously on receiving instances, GT.M stored such nodes with the null collation setting of the source instance, which effectively made them inaccessible. In some cases, the nodes also had incorrect values. (GTM-7856)

GT.M optimizes non-TP database read operations by bypassing the critical section in many cases. Previously, in application patterns involving frequent updates, non-TP reads typically required a database critical section. (GTM-7865)

MERGE issues GVSUBOFLOW and MAXNRSUBSCRIPTS error as appropriate in case the target of the MERGE is a global node. Previously, MERGE did not issue these error messages but created beyond-limit global nodes, resulting in a DBMAXNRSUBS error from MUPIP INTEG, or GVSUBOFLOW error from global accesses inside GT.M. (GTM-7867)

GT.M now issues a single restart of a transaction which encounters an instance freeze in the database file extension process. Previously the transaction would be restarted repeatedly, leading to unnecessary locking contention and CPU consumption. [UNIX] (GTM-7902)

$INCREMENT() handles the case where its first argument is a spanning node that another process concurrently KILLs. In prior versions starting with V6.0-000 this unusual circumstance could cause a segmentation violation (UNIX SIG11 or OpenVMS ACCVIO) or an unintended result. FIS is not aware of any occurrences of this issue in the production systems. Because spanning nodes are always string values and thus unlikely to be arguments to $INCREMENT() and the concurrency window for the issue is small, we do not expect users to encounter this problem. (GTM-7920)

Under conditions of high contention, exiting processes more efficiently acquire critical sections needed in order to release resources, and release any slots they occupy in "mutex queues," which are part of the control mechanism processes use to acquire critical sections. Previously, under conditions of high contention for a resource, it was possible for an exiting process to fail to acquire needed critical sections in a timely manner, and to not release queue slots, causing slots to "leak" and become unavailable for future processes, which in turn could cause spikes of seconds in response times. The workaround for stuck processes and mutex queue slot depletion was to monitor queue slot usage and shut down all mumps processes accessing the database file, and start new processes when the number of queue slots became small or when occasional response time spikes occurred. Note that processes occupying a mutex queue slot that terminate due to fatal errors, external actions like a kill -9, or OS resource starvation, continue to result in that slot becoming unavailable to subsequent processes. (GTM-8070)

SOCKET devices now accept LISTEN as an OPEN deviceparameter, treating it the same as a ZLISTEN. Previously, the LISTEN deviceparameter was used with devices in the TCP mnemonicspace. Also, GT.M produces an error - INVMNEMCSPC on UNIX and LIB-E-ACTIMAGE on OpenVMS - in response to an attempt to OPEN a device with the "TCP" mnemonicspace, which was previously obsolete and has long been deprecated. The "SOCKET" mnemonicspace provides access to TCP/IP sockets. On UNIX, other invalid mnemonicspaces also produce an INVMNEMCSPC error, where previously they produced an UNIMPLOP error; on OpenVMS, other invalid mnemonic spaces continue to produce a LIB-E-ACTIMAGE. (GTM-2040)

USE of a SOCKET device with just a "SOCKET=" device parameter specifying a listening socket checks for an incoming connect request. It accepts the request and creates a new connected socket. $KEY provides information on the new socket in the format described below for WRITE /WAIT: CONNECT|<handle>|<remote>. If the listening socket has no request pending, $KEY has the value of an empty string. Previously USE of a listening socket just made the SOCKET device the current device.

WRITE /WAIT on a SOCKET device checks for sockets with new connections pending, which it accepts by creating a new connected socket. WRITE /WAIT also checks for data in the internal buffers of already connected sockets as well as for new data (not yet read into the internal buffer) when determining whether a socket has data to READ. When a SOCKET device has more than one socket attached, WRITE /WAIT prioritizes connection requests on listening sockets ahead of data available for reading on already connected sockets, and, within each event type, giving highest priority for the longest ready down to least priority for the most recently ready. WRITE /WAIT selects the socket at the head of the priority list. Previously, WRITE /WAIT did not check the internal buffers and always returned the oldest socket (lowest index in the ZSHOW "D" output) in the current SOCKET device with an available event.

After a WRITE /WAIT completes successfully (that is: without timing out if a timeout is specified), the $KEY Intrinsic Special Variable (ISV) contains information describing the selected socket in the form: CONNECT|<handle>|<remote or READ|<handle>|<remote>. For TCP sockets, <remote> shows the port number, and for LOCAL sockets it shows the path to the socket file. If the WRITE /WAIT times out or there are no additional ready sockets beyond the one selected by the WRITE /WAIT, $KEY has the value of an empty string.

The $ZKEY ISV contains semi-colon (";") separated list of entries detailing any waiting sockets for a current SOCKET device. Each $ZKEY entry has the vertical-bar separated form: LISTENING|<handle>|<remote> for an incoming connection or READ|<handle>|<remote> for data available to READ. (GTM-3731)

GT.M now begins listening immediately on the OPEN or USE of a SOCKET device with the LISTEN or ZLISTEN deviceparameter set. The listen queue length defaults to 1. The "WRITE /LISTEN(numexpr)" may still be used to set the queue length. Previously GT.M only started listening when the application issued a WRITE /LISTEN(numexpr). The socket state as observable through $KEY and ZSHOW "D" is now set to "LISTENING" after a successful LISTEN/ZLISTEN; previously it was set to a "BOUND" state. The "BOUND" state now only occurs in error conditions. (GTM-4797)

A SOCKET device reports an appropriate error for an improperly formatted argument to the [Z]LISTEN deviceparameter Previously, such an error caused the process to terminate. (GTM-4800)

GT.M allows arguments present in a function definition in the C/Java external call table to be missing from a corresponding invocation. All non-trailing missing and output-only arguments translate to default values on the receiving side: false in case of gtm_jboolean_t (Java only), 0 in case of all numeric types (Java and C), empty string in case of gtm_char_t * and gtm_char_t ** (C only), a structure with 'length' field matching the preallocation size and 'address' field being a NULL pointer in case of gtm_string_t * (C only), an empty string in case of gtm_jstring_t (Java only), and an empty array in case of gtm_jbyte_array_t (Java only). Previously, omitted non-trailing non-output-only arguments were disallowed; omitted non-trailing output-only arguments of certain types resulted in arbitrary values passed to the target C or Java routine; and the 'length' field of output-only gtm_string_t * arguments was erroneously set to the length of the corresponding M variable. [UNIX] (GTM-6419)

The CLOSE command for a PIPE device that is not OPEN'd with the INDEPENDENT deviceparameter uses a timed check on termination status of the PIPE co-process. The TIMEOUT=intexpr specifies the timeout in seconds, defaulting to 2 seconds. Devices other than PIPE ignore the TIMEOUT deviceparameter. The $ZCLOSE Intrinsic Special Variable (ISV) provides termination status of the last PIPE CLOSE as follows: -99 when the check times out, -98 for unanticipated problems with the check, the negative of the signal value if a signal terminated the co-process, and under normal circumstances, the exit status returned by the last co-process. Previously, GT.M did not capture the termination status, and the process executing the CLOSE could hang if the co-process did not terminate. [UNIX](GTM-6666)

A DETACH operation on a socket now succeeds the first time it is done in a process. Previously, the first DETACH got a SOCKNOTFOUND error. (GTM-6867)

The JOB command now allows DETACHed sockets (i.e., sockets in the socket pool) to be passed as INPUT, OUTPUT, and/or ERROR,by specifying "SOCKET:<handle>" as the file name, where "<handle>" is the socket handle. On successful completion of theJOB command, any passed sockets are closed and no longer available to the parent process. To pass a socket as the $PRINCIPAL of a child process, pass it as both INPUT and OUTPUT parameters of the JOB command. Previously only file names could be specified for INPUT, OUTPUT, or ERROR. [UNIX] (GTM-7322)

The ZLINK command accepts and relinks routines on the GT.M invocation stack. Previously, attempting this would produce a LOADRUNNING error. The new behavior can be enabled with VIEW "LINK":"RECURSIVE", or by setting the environmental variable gtm_link to "RECURSIVE". When enabled, whenever an entryref includes a routine name, GT.M accesses the most recently ZLINK'd version of the routine. As usual, GT.M performs auto-ZLINK if no version has yet been linked, and, when an entryref includes no routine name, GT.M accesses the currently executing version. DO, GOTO, ZGOTO, and extrinsic functions invoke the appropriate version; meanwhile, older versions remain referenced by the M virtual machine stack until they execute a GOTO, explicit or implicit QUIT, or ZGOTO. Similarly $TEXT and ZPRINT retrieve the appropriate version's source when available; ZBREAK sets and clears breakpoints in the appropriate version. VIEW "LINK":"NORECURSIVE" and any other value of gtm_link provide the prior LOADRUNNING behavior; $VIEW("LINK") returns the current setting. In addition, ZSHOW offers a new information code, R, and ZSHOW "*" is now equivalent to ZSHOW "IVBDLGRC". Previously, ZSHOW "*" was equivalent to ZSHOW "IVBDLGSC". ZSHOW "R" displays the GT.M invocation stack and, for each routine on the stack, displays an MD5 checksum of its M source code. $VIEW("RTNCHECKSUM",<rtnname>) returns the source check-sum for the most recently ZLINK'd version of <rtnname>. [x86-64 Linux, AIX, Solaris] (GTM-7551)

GT.M SOCKET devices support LOCAL (also known as UNIX domain) sockets. Except as noted below, LOCAL sockets usage and behavior is the same as TCP sockets.

(GTM-7554)

GT.M processes appropriately handle standard input directed to a file and standard output directed to a socket. This situation could arise when an inetd invoked process in turn invoked MUMPS after redirecting its standard input to a file. Previously, such a configuration caused a KILLBYSIGSINFO1 error. [UNIX] (GTM-7603)

In UTF-8 mode, using both the FIXED and FOLLOW deviceparameter appropriately maintain buffered input when interrupted by an INTRPT, appropriately maintain $X and $Y, and detect proper lengths for fixed length READs (READ x#n). Previously, when a READ resumed from an INTRPT, it lost any input received before the INTRPT and fixed length reads sometimes improperly maintained $X and $Y, which could lead to incorrect READ length and/or hangs. [UNIX](GTM-7709)

If GT.M detects an error in using a system service to start or stop a system-maintained timer (used internally to schedule timed events, such as HANGs and dirty buffer flushes), it sends a SETITIMERFAILED diagnostic message to the operator log. In some cases, these errors are self-correcting but might cause some actions to take longer than intended; in those cases GT.M tags the operator log message as warning and lets the process continue. In other cases, the process may be subject to indefinite hanging, possibly holding shared resources, and GT.M terminates the process with a fatal SETITIMERFAILED error. Previously, GT.M did not verify the status of timer-setting operations. [UNIX] (GTM-7719)

A SOCKET OPEN or USE with the LISTEN device parameter properly handles errors while assigning a IP address and port, or path for the socket file to the socket (perform a bind system call). Errors other than EADDRINUSE ("address is already in use") set $DEVICE and, if IOERROR="TRAP" is enabled, raise an error condition. For EADDRINUSE errors on TCP sockets, GT.M reties the operation until the timeout, if specified, expires and then treats it as for other errors. GT.M does not retry EADDRINUSE errors on LOCAL sockets but treats them immediately as an error. Previously, $DEVICE was not set for errors, EADDRINUSE returned without raising an error even if no timeout was specified, and other errors always raised an error regardless of the IOERROR mode. (GTM-7739)

Auto-ZLINK and ZLINK commands without a .m or .o file extension in their argument determine the need to recompile based on whether the object file was more recently modified than the source file using time in nanoseconds, as provided by the underlying system call. Note that, although the format of the file modification timestamps provides a nanosecond granularity, many supported OSs currently update the file timestamps with an accuracy of one second. Previously, GT.M compared the modification timestamps of source and object files using time in seconds. Also, the GT.M configure installation script now preserves the original timestamps for all files that are not changed by the installation. Preserving timestamps on files, particularly GT.M source and object files, is generally a better practice. Previously, all files in the installation target directory had the installation timestamp.[UNIX] (GTM-7750)

In UTF-8 mode, READs for a sequential disk (SD) device appropriately detect and report bad character errors (BADCHAR), including correctly setting $DEVICE, $ZA, $KEY and $ZB. Previously, some bad character conditions incorrectly caused SYSTEM-E-UNKNOWN errors and did not appropriately maintain $DEVICE, $ZA, $KEY and $ZB. Also READs for a sequential disk using the FOLLOW deviceparameter appropriately handle timeouts. Previously, such READs could timeout prematurely. [UNIX](GTM-7763)

GT.M now sets $TEST for "WRITE /WAIT(timeout)" on SOCKET devices. Previously GT.M did not update $TEST in this case. (GTM-7773)

GT.M handles restart of SOCKET connect operations properly. In V6.0-003, out-of-band operations which occurred during a socket connect caused GT.M to exit with a fatal GTMASSERT2 message while trying to restart the connect. (GTM-7835)

The sorts-after operator (]]) works correctly with NOUNDEF enabled. Previously, with an operand which was NEW'd after some local variable had been KILL'd, the operator could sometimes produce an incorrect result. (GTM-7836)

ZGOTO <n>:<entryref> discards local variables NEW'd in levels n+1 and up. Previously, ZGOTO 0:<entryref> did not discard variables NEW'd in level 1.[UNIX](GTM-7842)

When failing to load ICU libraries on AIX, GT.M now reports the correct library and message in the output error message. Previously, GT.M produced an error message that incorrectly identified the problematic ICU library.[AIX](GTM-7886)

In UTF-8 mode, compiling a SET $ZPIECE() command with a literal piece separator that is an invalid UTF-8 string, such as $ZCHAR(255), does not raise a BADCHAR error. GT.M does raise BADCHAR error for SET $PIECE(), and the error shows the failing routine, line-number and the line of source that caused the error as for other compilation errors. Previously, GT.M gave inappropriate BADCHAR errors for SET $ZPIECE(), and BADCHAR errors gave no indication of which routine or line caused the error. [UNIX] (GTM-7896)

The JOB command adds the child process ID to the OUTPUT and ERROR filenames for JOB'd off process after a request to do so by a VIEW command with keyword JOBPID. Previously, if the JOB command occurred in an object residing in a shared library with this VIEW setting, it produced a Segmentation Violation (SIG-11). The workarounds were to use VIEW "NOJOBPID" or to avoid putting routines issuing JOB commands with VIEW "JOBPID" into shared libraries. [UNIX] (GTM-7905)

A JOB command appropriately handles situations in which the JOB'd process does not start, or terminates (for example with a SIGTERM), without notifying the process issuing the JOB command. In V6.0-002 and V6.0-003 the JOB command waited indefinitely for the nonexistent process to report its status. [UNIX](GTM-7908)

Certain complicated, deeply nested patterns raise a PATMAXLEN error. Previously, these patterns terminated the process with an segmentation violation (UNIX SIG11 or OpenVMS ACCVIO). Note that the patterns in question are very complex and seem unlikely to be encountered in real applications. FIS encountered this issue in testing and has not received a customer report of such an event. (GTM-7923)

MUPIP INTEG appends DBTN errors showing the block number following DBMBTNSIZMX errors that indicate a transaction number too large on a bit map block. Note that DBMBTNSIZMX is analogous to the DBTNTOOLG error for non-bitmap blocks. In addition, INTEG issues a DBTNLTCTN summary error with associated text including the count of DBMBTNSIZMX as well as DBTNTOOLG errors. Previously, INTEG did not report the bit map block number for a DBMBTNSIZMX or include this type of bit map error in the DBTNLTCTN summary. (GTM-7074)

JOURNAL BACKWARD RECOVERY or ROLLBACK detects previous interrupted JOURNAL RECOVERY or ROLLBACK correctly. Previously, in some cases MUPIP failed to detect previous JOURNAL RECOVERY or ROLLBACK interruption and incorrectly issued a JNLDBTNNOMATCH error. [UNIX](GTM-7453)

MUPIP BACKUP and MUPIP REPLIC operations only issue a CRITSEMFAIL error during journal and receiver pool initialization when they encounter a non-transient problem. Previously, these operations could fail if their shared resources were removed by another process in a small window at the journal and receiver pool initialization phase. FIS encountered this issue in testing and has not received a customer report of such an event. [UNIX](GTM-7626)

GT.M replication provides the option to encrypt communication using Transport Layer Security (TLS) version 1. The API of the encryption plugin has functionality required for TLS, and FIS tests the reference implementation of the encryption plugin against OpenSSL (http://www.openssl.org). As with database encryption, FIS neither recommends nor supports any specific cryptographic library, and you are responsible for selecting and procuring the library that best meets your needs, including support for that library appropriate to your use. FIS GT.M support includes the unmodified reference implementation of the plugin, but does not extend to any version of the plugin containing your modifications. Note: a basic understanding of TLS is assumed - see the Wikipedia article (http://en.wikipedia.org/wiki/Transport_Layer_Security) for more information.

For TLS support, the reference implementation requires a configuration file, pointed to by the gtmcrypt_config environment variable. This file supports both TLS and database encryption. To support the new configuration file format, the reference implementation also requires the popular libconfig library (http://www.hyperrealm.com/libconfig) to be installed and available; check for pre-packaged distributions for your operating system. Note that even if your environment is configured only for encrypted databases, the reference implementation requires libconfig library to be installed. The reference implementation continues to support the gtm_dbkeys environment variable for configurations that use encrypted databases but not TLS. For backward compatibility, if both $gtmcrypt_config and $gtm_dbkeys are defined, the reference implementation uses $gtm_dbkeys.

The Source and Receiver Server startup commands support the: TLSID, [NO]PLAINTEXTFALLBACK, and RENEGOTIATE_INTERVAL qualifiers:

Other configuration parameters required to establish a successful SSL/TLS connection with the reference implementation are specified in a configuration file, described below under Additional information for GTM-7743 - GT.M support for TLS encryption of replication streams using a plug-in, along with an example.(GTM-7743)

MUPIP JOURNAL -EXTRACT -GLOBAL=<key> works correctly when the key contains a string subscript with embedded double-quotes and the key specification terminates with the wildcard (*). Previously, in this combination of circumstances, JOURNAL -EXTRACT might miss some records the specification should have included. (GTM-7769)

MUPIP REORG -TRUNCATE -SELECT= does a thorough job of moving vestigial blocks of KILL'd global variables so that they do not linger after unused space, thereby making it possible to release unused space efficiently. Previously, GT.M did not move these vestigial blocks in all cases, and unused space before any block is not released to the file system. Note that even when an entire global variable is KILL'd, a database file retains two vestigial blocks: a root block and a single empty data block for that global variable. [UNIX](GTM-7820)

MUPIP EXTRACT robustly handles BINARY output format. Additionally, MUPIP EXTRACT, in BINARY mode, operates faster on encrypted databases. Previously, in certain rare cases involving heavy concurrent updates, MUPIP EXTRACT, in BINARY mode, generated corruptor empty extract files and performed unnecessary work on encrypted databases. [UNIX](GTM-7827)

Argumentless MUPIP RUNDOWN, MUPIP RUNDOWN -FILE and MUPIP RUNDOWN -REGION do not remove semaphore of a region in use. Previously, MUPIP RUNDOWN of a file with only read-only users or using differing global directories with RUNDOWN -REGION '*' could inappropriately remove the semaphore for a region, but leave its corresponding shared memory. In this state, a subsequent attempt by a process to first access a database resulted in REQRUNDOWN error. The workaround was to do an additional MUPIP RUNDOWN with no read-only processes active. [UNIX](GTM-7845)

MUPIP INTEG -SUBSCRIPTS works correctly if the argument to the -SUBSCRIPTS qualifier specifies a null subscript. Previously this terminated abnormally with a segmentation violation (UNIX SIG-11 or OpenVMS ACCVIO). (GTM-7853)

The Replication Receiver Server handles a case where a flow control message arrives at the receiver pool boundary. In V6.0-003, the work done for GTM-7631 introduced a new issue managing this case which caused a segmentation violation (SIG11). [UNIX](GTM-7890)

If an external replication filter fails to respond within just over a minute to delivery of a mini-transaction or TP transaction, a Source or Receiver Server issues a FILTERTIMEDOUT error, stops the filter, and exits. Previously, the servers could hang indefinitely waiting for a reply from the external filter. [UNIX](GTM-7892)

Replication processes and MUPIP RUNDOWN appropriately handle a journal pool deadlock check performed while waiting for access to the journal pool resource. Previously, under unusual circumstances, such a check could fail with a segmentation violation (SIG-11). FIS encountered this issue in testing and has not received a customer report of such an event. (GTM-7910)

MUPIP ONLINE ROLLBACK robustly handles cases where it fails to acquire standalone access on the Journal and/or Receive Pool. Previously, in such cases, ONLINE ROLLBACK terminated with a segmentation violation (SIG-11). FIS encountered this issue in testing and has not received a customer report of such an event.[UNIX](GTM-7916)

This release note is included in the Utilities - Other Than MUPIP section because the most visible changes to the user interface are in GDE. Global variables with the same unsubscripted name can span multiple database regions.

To effect this change, global directories support mapping at the level of subscripts, not just based on the unsubscripted global name.

Specify the subscripted name of root nodes to map entire subtrees to a regions, for example:

GDE> add    -name IMPL                             -region OTHERMUMPS  ! Map MUMPS implementations to OTHERMUMPS
GDE> add    -name IMPL("GT.M")                     -region MYMUMPS     ! While mapping GT.M to MYMUMPS

Specify ranges with a colon (:) to map ranges of subscripted names and their subtrees to a regions. for example:

GDE> add    -name PRODAGE(0:10)                    -region DECADE0     ! Ranges are closed on the left and open on the right
GDE> add    -name PRODAGE(10:20)                   -region DECADE1     ! PRODAGE(10) maps to DECADE1
GDE> add    -name PRODAGE(20:30)                   -region DECADE2

Ranges must be specific numbers or strings - GDE does not support wildcarding (using "*") in ranges.

You can use $CHAR() and $ZCHAR() to specify unprintable characters; the arguments must be positive integers (exponential - E syntax not allowed), and valid code points for $CHAR() or in range for $ZCHAR(), both with respect to the current $ZCHSET. Also, "" (an empty string) or no value (e.g. 20: or :20 or :) specify open-ended ranges, which span, on the left, from the first subscript ("") to, on the right, the last possible string , for example:

GDE> add    -name PRODAGE(:10)                     -region DECADE0     ! This line and the next are equivalent
GDE> add    -name PRODAGE("":10)                   -region DECADE0     ! numbers up to 10
GDE> add    -name PRODAGE(20:)                     -region DECADE2     ! 20 thru all numbers (> 20) + strings
GDE> add    -name PRODAGE(20:"")                   -region DECADE2     ! same as the add just above

Here is the syntax to map numeric subscripts and strings to separate regions

GDE> add    -name MODELNUM                         -region NUMERIC
GDE> add    -name MODELNUM($char(0):)              -region STRING

You can map global variables with the same unsubscripted name at multiple subscript levels, for example:

GDE> add    -name DIVISION("Europe","a":"m")       -region EUROPEAL
GDE> add    -name DIVISION("Europe","m":"z")       -region EUROPEM
GDE> add    -name DIVISION("Australia")            -region AUSTRALIA
GDE> add    -name DIVISION("USA","South","a":"m")  -region USSAL
GDE> add    -name DIVISION("USA","South","m":"{")  -region USSMZ
GDE> add    -name DIVISION("USA","WestCoast")      -region USWC

GDE merges overlapping ranges that map to the same region, for example:

GDE> add    -name X(1:10)                          -region=AREG
GDE> add    -name X(6:15)                          -region=AREG
GDE> show -name

     *** NAMES ***
 Global                             Region
 ------------------------------------------------------------------------------
 *                                  DEFAULT
 X(1:15)                            AREG

Specifying overlapping ranges that map to different regions produces an error, for example:

GDE> add    -name X(1:10)                          -region=AREG
GDE> add    -name X(6:15)                          -region=BREG
%GDE-E-NAMRANGEOVERLAP, Range in name specifications X(6:15) and X(1:10) overlap using collation sequence #0

Note that sub-ranges that map to different regions do not produce an error; for example: mapping X(1:15) and X(6:10) works because the second is entirely within the first. Only overlapping ranges (those that have a non-intersecting portion) that map to different regions produce this error.

Because string subscripts are subject to collation (the unsubscripted portion of a global variable name and numeric subscripts are not), GDE needs to know the collation sequence number associated with each unsubscripted global variable name. Zero (0) is the number for default M collation. As a consequence, when you use alternative collation(s) (other than 0), the collation transforms must be available to GDE in the same way as they are to other GT.M components. All of a global (all nodes sharing the same unsubscripted global name) must have a single collation, which is implicitly the case for globals that do not span multiple regions. Globals that do not span multiple regions and do not have any collation characteristics defined in the GBLNAME section of the global directory take on the default collation characteristics defined in the database region to which they map. On the other hand, globals that span multiple regions have their collation implicitly (collation 0), or explicitly, established by the GBLNAME section of the global directory and cannot adopt a differing collation based on the region collation characteristic. Because GT.M determines collation for globals spanning multiple regions by the GBLNAME characteristic, which cannot change once the database files are created, GDE reports collation on many error messages. In the above example, although the overlapping ranges are numeric, GDE reports the collation sequence number as part of the error message.

GDE merges and splits ranges as it reads each NAME command. Therefore, the following does not generate an error because the second command merges the ranges and the third splits an existing range, whereas reversing the second and third would trigger the NAMRANGEOVERLAP error.

GDE> add    -name X(1:10)                          -region=AREG
GDE> add    -name X(6:15)                          -region=AREG
GDE> add    -name X(6:15)                          -region=BREG
GDE> show   -name

        *** NAMES ***
Global                             Region
 ------------------------------------------------------------------------------
 *                                  DEFAULT
 X(1:6)                             AREG
 X(6:15)                            BREG

A point specification overrides a range specification at the same level, for example:

GDE> add    -name a(1:5)                          -region=AREG
GDE> add    -name a(3:4)                          -region=BREG
GDE> add    -name a(3)                            -region=CREG
GDE> show -name

         *** NAMES ***
 Global                             Region
 ------------------------------------------------------------------------------
 *                                  DEFAULT
 a(1:3)                             AREG
 a(3)                               CREG
 a(3:4)                             BREG
 a(4:5)                             AREG

Adding a name and deleting it can change the mapping, for example, note the "hole" left between 80 and 90:

GDE> add    -name a(1:100)                        -region=AREG
GDE> add    -name a(80:90)                        -region=BREG
GDE> delete -name a(80:90)
GDE> show   -name

         *** NAMES ***
 Global                             Region
 ------------------------------------------------------------------------------
 *                                  DEFAULT
 a(1:80)                            AREG
 a(90:100)                          AREG

Specifying separate entries for a subtree provides a way to map a root to a different region, for example, note X(2) maps to BREG whereas the rest of of X is mapped to AREG:

GDE> add    -name X                               -region=AREG
GDE> add    -name X(2)                            -region=BREG
GDE> add    -name X(2,:)                          -region=AREG

Out-of-order ranges generate errors

GDE> add    -name a(20:10)                        -region=AREG
%GDE-E-NAMRANGEORDER, Range in name specification a(20:10) specifies out-of-order subscripts using collation sequence #0

GDE SHOW -NAME and GDE SHOW -MAP display longer lines if subscripted names exceed the 36 column limit, and display non-printable characters in string subscripts using $CHAR() or $ZCHAR() notation.

The SHOW -MAP command uses "++" as a suffix to mean "the next lexical subscript", for example:

GDE> add    -name a(1)                            -region=A1
GDE> show   -map

                                  *** MAP ***
   -  -  -  -  -  -  -  -  -  - Names -  -  - -  -  -  -  -  -  -
 From                            Up to                            Region / Segment / File(def ext: .dat)
--------------------------------------------------------------------------------------------------------------------------
 %                               a(1)                             REG = DEFAULT
                                                                  SEG = DEFAULT
                                                                  FILE = mumps.dat
 a(1)                            a(1)++                           REG = A1
                                                                  SEG = NONE
                                                                  FILE = NONE
 a(1)++                          ...                              REG = DEFAULT
                                                                  SEG = DEFAULT
                                                                  FILE = mumps.dat

Here is an example of a complex mapping:

GDE> add    -name x                               -region=REG1
GDE> add    -name x(5)                            -region=REG1
GDE> add    -name x(5,10:)                        -region=REG2
GDE> add    -name x(5:20)                         -region=REG2
GDE> add    -name x(20)                           -region=REG2
GDE> add    -name x(20,40)                        -region=REG2
GDE> add    -name x(20,40,50:)                    -region=REG3
GDE> add    -name x(20,40:)                       -region=REG3
GDE> add    -name x(20:)                          -region=REG3

The new object-type GBLNAME (existing object-types are NAME, REGION, and SEGMENT) with the COLLATION characteristic provides a mechanism to specify the collation to use for global variables sharing the same unsubscripted name.

GDE> add    -gblname a                            -collation=0
GDE> show   -gblname

         *** GBLNAMES ***
 Global                             Coll  Ver
 ------------------------------------------------------------------------------
 a                                     0    0

GDE requires the library for any non-zero collation to be available.

GDE> add    -gblname a                            -collation=1
%GDE-E-GBLNAMCOLLUNDEF, Error opening shared library of collation sequence #1 library for GBLNAME a

Regions that contain global variables sharing the same unsubscripted name that span regions must use standard null collation; attempting to use the deprecated original null collation produces an error, for example:

GDE> add    -name a(1)                            -region=areg
GDE> add    -region areg                          -dynamic=areg
GDE> add    -segment areg                         -file=areg
GDE> verify
%GDE-I-STDNULLCOLLREQ, Region DEFAULT needs Standard Null Collation enabled because global ^a spans through it
%GDE-I-STDNULLCOLLREQ, Region AREG needs Standard Null Collation enabled because global ^a spans through it
%GDE-I-VERIFY, Verification FAILED

Notes

Caution

As MUPIP EXTRACT processes all nodes in one region of a global variable sharing the same unsubscripted name and mapping across multiple regions before moving on to the next region, nodes in the output of MUPIP EXTRACT may not be in sequential order, especially if subscripts in one region bracket subscripts in another. In the edge case where blocks in a region contain nodes with subscripts mapped to that region as well as nodes with subscripts mapped to another region according the current global directory, the output of MUPIP EXTRACT (which extracts all the nodes in a block before moving on to the next block) includes those nodes, whereas the output of ZWRITE which steps subscript by subscript through a global variable having the same unsubscripted name as determined by the current global directory does not. On the other hand, if the current global directory does not map any of a global to a region, EXTRACT does not check that region for nodes in that global.

Caution

GT.M associates LOCKs with resource names starting with an up-arrow (^) with the database region holding the global with the same unsubscripted name as the resource. For globals spanning multiple regions and multiple global directories this means that up-arrow LOCK resource names may not address a locking protocol in the same way as for globals that do not span multiple regions. To address this, ensure that all overlapping global directories used with up-arrow LOCK resource names and globals spanning multiple nodes map the unsubscripted global name to the same region. Alternatively, you can use LOCK resource names with no leading up-arrow and control LOCK interactions with the LOCKS global directory characteristic.

In conjunction with this change:

[UNIX, except as noted] (GTM-2168)

GDE ignores spaces in file names supplied for scripted operation invoked with @; Previously, accidental trailing spaces could cause hard to diagnose errors. (GTM-5572)

DSE remains in the same region after a DUMP -BLOCK operation where records in the dumped block contained global names mapping to a different region. Previously DSE incorrectly switched to the region mapped, causing subsequent DSE commands to operate on the wrong database file. (GTM-7562)

GDE treats text after an exclamation point (!) as a comment, whether or not in a command file. Exclamation points within quoted strings do not create comments. Previously, GDE only treated command file lines starting with an exclamation point as comments. (GTM-7693)

GDE accepts the -MUTEX_SLOTS=<i> qualifier for segment objects with minimum of 1Ki and a maximum of 32Ki. This facilitates establishing an initial value for this database parameter. Previously, setting -MUTEX_SLOTS required MUPIP SET, which is still available for adjusting existing databases. (GTM-7742)

The bypass mechanism DSE and LKE use to open a busy database uses more robust logic to report their action. Previously the process could generate a series of misleading error messages and fail. The workaround was to retry the operation. [UNIX](GTM-7848)

gtmsecshr ignores requests to ensure that a non-existent process is not suspended. Previously, it issued a warning under these circumstances, which can arise when a process holds a shared resource just before terminating. [UNIX](GTM-7858)

GTMJI allows assigning ISVs to M gtm_string_t-type input/output arguments in call-in invocations. Previously, assigning the values of certain ISVs to gtm_string_t-type input/output arguments caused segmentation violations. [IBM System p AIX, Sun SPARC Solaris, x86_64 GNU/Linux, x86 GNU/Linux] (GTM-7866)