GT.M now issues a LOCKSPACEFULL error to the application and the operator log when the environment attempts more concurrent M LOCKs than the configured LOCK_SPACE for the region can support. Previously under these conditions, GT.M gave a fatal GTMASSERT error. (S9C11-002251)

GT.M now reports an error if the first argument of a two argument $GET() is defined and the second is an undefined local variable. Prior versions returned the value of the first argument and ignored the error mandated by the standard. If the second argument was a global variable, this worked properly in earlier versions. Note that any code that relies on the prior behavior needs updating. (S9D08-002354)

GT.M now reports UNDEF errors when it encounters an undefined variable specified for a local variable subscript as it already does for a global variable and as defined in the language standard. In prior versions, KILL and its variants or functions including $DATA(), $GET(), $ORDER() and $QUERY() ignored these errors. Under NOUNDEF operation this change has no user impact. Note: code that relied on the earlier non-standard behavior may now give errors and require revision.

GT.M now supports VIEW "NEVERLVNULLSUBS" to issue the LVNULLSUBS error for any inappropriate use of empty string subscripts in local arrays.  The [new] gtm_lvnullsubs UNIX environment variable can have the values 0 - equivalent to VIEW "NOLVNULLSUBS", 1  (the default) - equivalent to VIEW "LVNULLSUBS" or  2 - equivalent to VIEW "NEVERLVNULLSUBS". In VMS, the initial state of this characteristic is controlled by the GTM$DEFAULTS file. An empty string as the last subscript in $ORDER() and $QUERY() has the semantic significance of requesting the next lexical item and is not subject to NULLSUBS errors. $VIEW("LVNULLSUBS") returns the corresponding numeric value. In prior versions, GT.M only supported NOLVNULLSUBS, which reported the LVNULLSUBS error for SET commands that would have instantiated a prohibited local array subscript, but not for KILL and its variants or functions including $DATA(), $GET(), $ORDER() and $QUERY().  $VIEW("LVNULLSUBS") returned 0 for NOLVNULLSUBS and 1 for LVNULLSUBS. Note that code that expected $VIEW("LVNULLSUBS") to provide a truth value must be revised. Empty string subscript checking for global variables is controlled by a database region characteristic that was already comprehensive and is not affected by this change. Note that programs using empty subscripts for references where those references cannot possibly exist are wasting computer resources, so FIS recommends using LVNULLSUBS or NEVERLVNULLSUBS for local variables and NULLSUBS options ALWAYS or NEVER for global variables. (S9D10-002376) V5.4-000A

If the OPEN or USE for a SOCKET device does not specify a MOREREADTIME parameter, any READ uses an initial 200msec internal timeout to minimize CPU load. Subsequent attempts to acquire more characters to fulfill the same READ use a 10msec internal timeout. When a MOREREADTIME parameter specifies a value, READ uses that value exclusively. Previously READ used a default of 10msec when no MOREREADTIME parameter was specified and consequently could consume considerable CPU cycles, especially on a lightly loaded system. If your application currently uses the MOREREADTIME parameter, FIS suggest that you consider whether this new behavior provides better performance and/or reduces CPU usage and allows you to avoid specifying MOREREADTIME. (S9J07-002732)

The distribution tar file now provides all files owned by the innocuous user:group ids 40535:40535, with no executable or writable files. With the exception of AIX, no directories have write permissions. On AIX, unpacking the distribution file no longer requires root permissions. In prior versions, the configure script adjusted the authorizations, but the distribution had various settings, and, on AIX, required root access to unpack the distribution. [UNIX] (S9J07-002737)

GT.M now handles the OPEN of a PIPE device better in certain situations. In prior versions, specifying various device parameters such as ESCAPE for the $PRINCIPAL device when it was a terminal could result in delays or other problems. [UNIX] (S9J11-002749)

GT.M now ensures a proper value for $QUIT on x86 and x86-64 GNU/Linux. In V5.4-000 under certain circumstances, it could return an inappropriate result. [x86-64 Linux, x86 Linux] (S9K02-002755) V5.4-000A

On compiling an M program, GT.M now omits reports showing the point in the line of code that caused the warning or error for lines longer than 1023 characters; it still reports the line number and column number. In prior versions, attempts to report such long lines could overrun an internal buffer and cause the process to hang or fail. (S9K02-002756) V5.4-000A

The TRUNCATE parameter on USE for sequential file devices now works correctly for Linux. In prior versions, USE could truncate at the wrong position. [Linux] (C9E02-002513)

The GT.M SOCKET device behavior has improved in a number of ways.

SOCKET CONNECT operations now respond to MUPIP INTRPT; in prior versions they did not.

If the command does not specify a timeout, a SOCKET OPEN now waits for the connection to complete or an interrupting event that terminates the attempt. After an interrupt that does not terminate the connection attempt, such as the default value of $ZINTERRUPT ("IF $ZJOBEXAM()"), the process resumes the wait when the XECUTE of the $ZINTERRUPT completes. Previously, an untimed SOCKET OPEN <device>:CONNECT would make a single attempt to connect and return whether or not it was successful, as if it had a timeout of 0.

SOCKET CONNECT status values EINPROGRESS, EWOULDBLOCK and, on some platforms, EALREADY now wait for the connection to complete or for the specified timeout to expire.  Previously, GT.M treated these conditions as errors.

If IOERROR="TRAP" is specified on a SOCKET OPEN or USE, it now invokes any active EXCEPTION, $ETRAP, or $ZTRAP if an error occurs on a CONNECT.  The above status values as well as ECONNREFUSED and ETIMEDOUT are not considered as errors.  Previously, status values other than ECONNREFUSED or ETIMEDOUT would always be treated as errors regardless of the IOERROR value. (C9D07-002356, C9H06-002868)

GT.M now allows only a single set of gtm_init/gtm_exit per process and any attempt to call gtm_init/gtm_ci after a gtm_exit returns the code associated with the CALLINAFTERXIT error. Earlier GT.M versions also did not support invoking gtm_init or gtm_ci after a gtm_exit but did not enforce the restriction.[UNIX] (C9H07-002875)

READ from a sequential disk file in UTF-8 mode read buffers the file reads for higher throughput. In prior versions, single character reads limited performance. [UNIX] (C9J04-003118)

The UNIX installation script for GT.M now sets the executable bits for files such as mumps and mupip only after a successful installation. Previously, the executable bits for these files were sometimes set after an unsuccessful installation. [UNIX] (C9J06-003140)

ZSHOW and ZWRITE now display information about alias variables with no associated data provided at least two of the joined variables are selected by the command argument. In prior releases these commands displayed alias containers with no data, but not alias variables with no data. When an alias exists with no data for the command to display, GT.M holds the variable so that it can become the right-hand side of a proper SET * command. However, if the command selection criteria excludes any other alias, GT.M has insufficient information to form a valid SET * argument and does not display that alias. Alias container variables always have a default value (the empty string), so they are handled differently and display as they did before. (C9J07-003148)

GT.M now handles very large object files correctly on z/OS. V5.3-003[A] produced invalid object files if the generated code exceeded 32KB. [z/OS] (C9J08-003170)

In UTF8 mode, when the minor version is not specified in the environment variable gtm_icu_version, e.g. 3 instead of 3.6, GT.M issues an ICUSYMNOTFOUND error. Previous versions of GT.M used to issue the ICUVERLT36 error even when the major version specified was greater than or equal to 4. [AIX, GNU/Linux, HP-UX, Solaris, z/OS] (C9J08-003174)

^%XCMD is a utility program that XECUTEs input from the shell command line and returns any error status (truncated to a single byte on UNIX) generated by that code. For example in UNIX: mumps -run %XCMD 'write "hello world",!' would produce "hello world" (C9J08-003179)

MUPIP and GT.M now clean up temporary snapshot files from INTEG under a wider variety of conditions. In V5.4-000, stopping the INTEG with a <CTRL-C> or a signal could leave the snapshot file orphaned. [UNIX] (S9K02-002757) V5.4-000A

MUPIP INTEG -REGION now supports the -ONLINE qualifier. This allows checking database structural integrity to run concurrently with database updates. Previously, INTEG froze updates on each region while checking it. Online integ is not supported with the -FILE qualifier. The -NOONLINE qualifier indicates that the database should be frozen during the MUPIP INTEG operation.

The -ONLINE qualifier is now the default for MUPIP INTEG, except for databases containing V4 blocks, for which the default is -NOONLINE. Note that databases containing V4 blocks should exist only in databases that are in the process of being migrated from V4 to V5; please complete your migration to the V5 format before using MUPIP INTEG -ONLINE.

The environment variable GTM_BAKTMPDIR (unlike most GT.M environment variables, this is upper case; it is the same directory used for MUPIP BACKUP -ONLINE) can be used to indicate a directory where MUPIP should place the snapshot files (used by online integ). If GTM_BAKTMPDIR does not exist, MUPIP places the snapshot files in the current directory at the time you issue the INTEG command.

As it checks each database file, MUPIP INTEG -ONLINE creates a sparse file of the same size as the database.  As each GT.M process updates the database, it places a copy of the old block in the sparse file before updating the database.  For any database blocks with a newer transaction number than the start of the INTEG, MUPIP uses the copy in the sparse file.  Thus, analogous with MUPIP BACKUP -ONLINE, INTEG reports on the state of the database as of when it starts, not as of when it completes.  Note: a command such as ls -l command shows sparse files at their full size, but does not show actual disk usage. Use a command such as du -sh to see actual disk usage.

Temporary directory security settings must allow write access by the the MUPIP process and read access by all processes updating the database. MUPIP creates the temporary file with the same access as the database file so processes updating the database can write to the temporary file. If the database is encrypted, the updating processes write encrypted blocks to the snapshot file and the MUPIP INTEG process must start with access to appropriate key information as it does even -NOONLINE.

Only one online integ can be active per database region. If an online integ is already active, a subsequent one issues an error and immediately terminates. If an online integ does not successfully complete, GT.M cleans it up in one of the following ways:

MUPIP INTEG -TN_RESET now resets block transaction numbers and backup event recorded transaction numbers to (one) 1, and the current transaction number to two (2) which makes the backup event recorded transaction numbers more meaningful and useful. It also issues an advisory message to perform a backup. Previously the backup event recorded transaction numbers were not maintained by -TN_RESET, which meant they held inappropriate information. Note that there should never be a need for a -TN_RESET on a datbase with only V5 blocks, even when cleaning up after a runaway process. (C9B12-001842)

GT.M now provides a new monitoring facility for processes holding a resource for an unexpected amount of time. This facility runs a shell command or a script pointed to by the new environment variable gtm_procstuckexe when any of the following conditions occur:  

MUPIP now verifies that the byte ordering (endian-ness) of a journal file matches the byte ordering of the instance on which it is running and issues an error in case of a mismatch. In prior versions, such a case of operational cross-threading caused MUPIP to fail with various complaints about the journal file format or even a memory access violation (SIG-11 or ACCVIO). (C9J04-003110)

The replication source server now handles database files that are in WAS_ON replication state (and in turn a journaling state of OFF) in a safe manner. If it cannot find the journal records in the journal pool, it reads as many journal records as possible from the journal file chain and when it reaches the point where journaling got turned OFF, it periodically displays a REPL_WARN message until journaling gets turned back ON and then correctly issues a NOPREVLINK error. At all points, it behaves as a read-only process as far as journaling is concerned and never switches journal files. In previous versions of GT.M, it was possible for the source server to incorrectly attempt a switch of journal files in such WAS_ON regions eventually terminating with many types of errors (JNLFILOPN etc.). (C9J07-003156) V5.4-000A

The messages produced by a replicating instance when the originating instance encounters a REPLINSTNOHIST error now read:

Originating instance encountered a REPLINSTNOHIST error. JNL_SEQNO of this replicating instance precedes the current history in the originating instance file. uuuu exiting.

Where uuuu is either a rollback or the receiver server. [UNIX] (C9J08-003166)

MUPIP BACKUP -REPLINSTANCE now ensures its backup of the replication instance file is in sync with all databases backed up as part of the same command. This fixes a regression introduced in GT.M V5.3-000 (and present in V5.3-004A) that could in rare cases cause REPLINSTDBMATCH error when attempting to use the backed-up instance file because it held an incorrect sequence number for a database file. A workaround for this is to recreate the instance file, as the backed up databases are correct. (C9K01-003222)[UNIX]

GT.M now correctly handles an unusual set of circumstances that can occur if MUPIP BACKUP -ONLINE and MUPIP INTEG -ONLINE are running concurrently with M transaction processing. In V5.4-000, these unusual circumstances could cause a SIG-11. [UNIX] (C9K02-003239) V5.4-000A

DSE, LKE, and MUPIP now handle unusually long command input lines more gracefully. In prior versions, long commands could damage adjacent memory and lead to difficult to diagnose failures. [UNIX] (C9J09-003199)