The current maintenance kits are Caché and Ensemble 2009.1.6. This document is only part of the available maintenance kit documentation. The full set of documents is available in the docs subdirectory of the product installation directory. Release notes and related material are available as follows:

• Late-breaking information about the release: <install-dir>\docs\cache\en-us\relnotes\cache\readme.1st or <install-dir>\docs\cache\en-us\relnotes\ensemble\readme.1st
• Release notes for 2009.1.6: <install-dir>\docs\cache\en-us\relnotes\cache\relnotes.htm or <install-dir>\docs\cache\en-us\relnotes\ensemble\relnotes.htm
• Release notes for the previous maintenance kits since 2009.1: <install-dir>\docs\cache\en-us\relnotes\cache\prenotes.htm or <install-dir>\docs\cache\en-us\relnotes\ensemble\prenotes.htm (this document)

This document is a historical archive of changes since the original 2009.1 release, but prior to the current maintenance kit. Changes unique to the current maintenance kit are described in relnotes.htm.

This document includes the following sections:

• General Announcements
• InterSystems News, Alerts, and Advisories
• Online Documentation
• Change Reports for Previous Maintenance Kits

InterSystems advises all users to recompile their application classes as part of the upgrade process so as to take advantage of all the performance improvements as well as other changes that may affect them in this release.

Anyone upgrading from a Caché release at or before 5.0 will need a new license key.

From time to time, InterSystems publishes items of immediate importance to users of our software. These include alerts, mission critical issues, important updates, fixes, and release announcements.

The most current list can be obtained from the InterSystems web site.

Users should check this list periodically to obtain the latest information on issues that may have an effect on the operation of their site.

As a convenience to our users, InterSystems provides online access to documentation for recent product versions at the InterSystems documentation web site.

The following are the previous maintenance releases of Caché 2009.1:

This change stops the Windows installation so it no longer reset passwords in the CSP Gateway for server connections in CSP.ini for external web servers on upgrades.

This change ensures that the Gateway responds gracefully to cases where requests for additional memory are denied by the operating system. This scenario occurs particularly when buffering huge amounts of response data in order to generate a HTTP content-length header field. Previous versions did not always handle this event cleanly and an access violation followed by memory leakage could occur.

This build will ensure that the terminating CRLF is discarded from the final output stream where response chunking is in operation between CSPnsd and mod_csp, in cases where a chunked response is not expected by the client.

This change protects against a problem in the IIS v7 Native Modules interface that could cause a memory access violation to occur if an invalid query string was submitted.

In particular, URLs containing empty query strings after the introducing '?' character could cause access violations in the hosting application pool.

This change resolves an access violation that could occur if a connection reserved for a state-aware session closed unexpectedly. This error is reported in the Event Log as:

Error Condition
Exception caught in f:cspTCPIPOpenSession: c0000005|62; etc ...

In addition to this fix, the logic for testing the health of state-aware connections has been strengthened. The periodic test will be suppressed if there is heavy user activity against the connection in question. If a connection reports a network error, the Gateway will perform the full round-trip test before marking the connection as defunct and closing its (client) side of the connection.

This change corrects a defect that could prevent data from being properly exported to Microsoft Excel spreadsheets.

This change fixes an error where the statement handle of a Stored Procedure call through the SQL Adapter is not correctly dropped. This leads to resource leakage in the target JDBC or ODBC driver.

This change fixes several systemic bugs in HL7 and other VDoc classes, and makes the DTL compiler generate a PokeDocType() for new target objects instead of setting DocType directly. A new object generally is missing segments it needs to pass schema validation so there's no point in triggering a BuildMap() on it immediately. The DTL change arranges this.

The changes are:

• Make CopyValues() generate cloned segment objects in the target, not just copy object references or segment ID's. This prevents a bug where if the target object produced by a DTL is deleted without ever being saved, it could delete the segment storage referred to by the source object.
• Make buildRuntimeIndexFromPath() populate the "bidirectionalLink" map. This prevents duplicate segment object allocation when retrieving segments by path and by index.
• Fix code that was intended to clear BuildMapStatus whenever a new segment is added or removed, but was mis-coded so that it didn't. This could cause the segment map to fail to be re-built in the course of a DTL when it should have been.
• Specify pKeepParsingAfterError=0 in calls to BuildMap() that are invoked internally in the course of adding or removing segments in a document. This should make DTL operations faster by avoiding looping. It also means that BuildMapStatus values in the final target object from a DTL may be less verbose than those from Validation or HTML display of a complete message.

This change fixes the method PokeDocType() of the various VDoc Segment classes to be the same as DocTypeSet(). PokeDocType() is useful for Document classes to avoid the overhead of the BuildMap() invocation that is implicit in a normal Set of the DocType property. However, for Segment classes there is no BuildMap(), but some helper shadow properties do need to get set.

This change fixes two bugs in the XML HL7 schema loader:

1. Allow wildcard segment references named 'Any'.
2. Do not hang the job in an infinite loop if someone saves a schema with base referring to itself.

Under extreme connection situations on multiple threads (50+ simultanious connections), resource issues at the socket layer cause failures which can end up causing a freeze in the connection pooling logic, blocking all threads. The failure is machine dependent in terms of the number of connections needed to cause this, but the fix is the same for all platforms.

This change corrects a situation where %XML.TextReader fails to get very long CDATA value. This will also apply to any character, comment or cdata.

This change corrects a situation where an SQL parsing error was reported for the case where it expected an EXISTS clause but none was found.

This change fixes a problem specific to the Microsoft SQL Server. Stream retrieval for multiple rows caused JVM memory to grow. Now, we do not rely on garbage collection but instead explicitly remove large objects.

This change corrects a <STORE> error caused by a misplaced curly brace internally in the ODBC driver.

A problem has been corrected where some schema names would not display properly in the Management Portal when Czech Unicode Locale was installed.

A problem has been corrected where a view would not compile even when the owner of the view has all the necessary privileges on the view query's table(s) granted to the owner via a role.

This change addresses how Caché handles the situation of having cached queries and being out of routine buffers.

If a process attempts to allocate another routine buffer but all system routine buffers are in-use, it will now reduce its own cached routines down to 10% of its allowed maximum. If still no routine buffer is available then Caché then drops the system wide number of routines cached per process.

While waiting for a job to start, this change allows Caché to handle various cleanup work to prevent possible future deadlocks involving resource shortages.

On Linux distributions where the zlib library (libz.so) was not installed in the searchable path (by default or by user's choice), one would get <DYNAMIC LIBRARY LOAD> error running Cache compression/decompression utility (e.g., $System.Util.Compress() etc.). This change addressed that by including a copy of the zlib library in the Caché installation.

This change corrects a rare problem on Windows systems where when the control process resumes a system that had been suspended due to lack of write daemon activitiy (for example, a long backup). In this circumstance, it was possible for a small number of jobs to be permenently left suspended.

This corrects an issues where duplicate console messages on which a newline has been supplied were not being suppressed.

In cases where CacheIDS<device #>.<inode #> has same value on both nodes accessing a shared disk, confusion can arise. One symptom of this was false status messages being sent which caused the partner node to crash.

A message transmission that requires a reply (such as JOBEXAM) may hang forever if the target of the message exits after receiving the message but before responding to the request. This has now been fixed.

The inter-cluster ECP session activity is not a regular session, and journal position is not tracked. But the session is started with the current journal position, and if the slave did not stay idle for more than 60 seconds (time enough for it to clear the old journal position), it prevented purging the old journal files.

This is now handled correctly.

A problem affecting OpenVMS Clusters and cluster mounted encrypted databases has been resolved which would have resulted in database degradation if blocks from an encrypted database were recovered from the pij during cluster failover or during cluster recovery at startup.

This change is compatible with the existing cluster code so that a running cluster can be upgraded one node at a time without shutting down the entire cluster. However, it is crucial that encrypted databases must not not be cluster mounted until ALL of clustered instances (for that Caché cluster) have been updated to this version.

This correction resolves a rare problem during cluster recovery after the cluster master shuts down or fails. This defect prevents subsequent attempts to join the cluster. The problem can occur any time the master leaves the cluster, gracefully or not and required restarting all of the cluster nodes to resolve it.

This change corrects an error that could occur during a journal restore of an ECP cluster. Under some circumstances, the last journal records in the restore would not be posted to the database because of an error in waiting for all the records to be written to disk.

This change fixed a transaction rollback problem when there are BITSET and Global SET journal records on the same global node. Under some circumstances, the SET journal record might not be rolled back when it should be.

This change corrects a situation that would cause anonymous ODBC connections originating from a Windows terminal server session to incorrectly consume a license unit when a web add-on license is used.

This corrects an issue that caused cluster shadows with a filter to apply an update twice in a row. This unnecessarily inflated the journal files on the shadow side and degraded shadowing performance. It did not affect data integrity, however, as the duplicate followed the original record immediately.

This corrects a defect that could cause a shadow to create a corrupted copy of source journal file and hang in certain circumstances involving stopping and starting journaling on the source side.

A previous change caused %SYS.Namespaces.GetGlobalDest() to fail if a single-level subscript was used as the third argument, and it was not enclosed in quotes. This change restores the previous capability.

Previously releases, when a new global subscript-level mapping was detected at startup, Caché validated that the collation of the physical global in the database matched the collation, if the global was populated with data. If the collation did not match, Caché reported an error, and did not save the global SLM mapping.

If the configuration file was edited by hand, this could cause problems at startup when Caché attempted to mount the database.

This has been changed so that Caché only checks the collation when it is entered via the Management Portal. Futhermore, it only checks the collation of the physical global when the mapping entered is a subscript-level mapping.

The "locksiz" entry in the CPF file contains the size (in bytes) of memory allocated for locks. The startup activity rounds up the value to the next multiple of 64 kilobytes. The default is 1,179,648 bytes. On the HP platform, the default is four times the usual default or 4,718,592 bytes.

The range is 65536 bytes up to the gmheap size. If a system needs more room for the lock table, increase the gmheap parameter as well.

The change takes effect immediately without the need for a restart.

When using LDAP for user authentication, if a user entered a username, with a null password, they would always be authenticated and allowed into the system. This is because the operating system LDAP libraries treat this as an "anonymous" bind, and uses the LDAP simple_bind_s function to succeed with a null password.

With this change, Caché tests to see if a null password has been entered, and reject the attempt.

This is a problem only on the following client to server connections:

• Unix or OpenVMS client to any LDAP server
• Windows client to non active directory LDAP server

It is not a problem on a Windows client to a Windows Active Directory LDAP server connection.

This change prevents the reset of the contents of the Gateway shared memory block in the case where an Apache web server re-reads its configuration file.

The reset only occurs on UNIX-based Apache installations.

This change fixes a problem that could result in the Apache DSO module hanging and erroneously timing-out when processing response messages containing no content body.

This change represents multiple corrections to DeepSee involving:

• Changes in roles causing established filters and subject areas to be lost
  
  
• Increases in the limitations on the number of globals that can be imported
  
  
• Pivot table handling that prevented details from being properly printed
  
  
• Premature session timeouts and subsequent logouts
  

and other miscellaneous fixes.

This change removes the classes, %BI.SOAP, from the product as a fix to a security hole.

This change corrects problems in the SQL Adapter JDBC support arising from failure to handle return values from the underlying JDBC method invocations other than 0 or -1. In fact, these methods return values indicating row counts and the presence of Stored Procedure output ResultSet data.

This change corrects behavior in the SQL Adapters that resulted in erroneous errors when checking for multiple stored procedure results. Previously, the adapter would drop the current statement handle prematurely.

The SQL Adapter supports JDBC configurations either by specifying a JDBC URL in the DSN config value, or by configuring the DSN value as the name of a "SQL Gateway Connection" configuration block configured in the System Management Portal.

When calling a Stored Procedure via an adapter whose DSN value is configured using a named configuration block, the SP would fail with <INVALID OREF> zgetProcColumnsJ+3^EnsLib.SQL.CommonJ.1. This change corrects that problem.

This change fixes an erroneous but syntactically legal construction in the getReplyCode() method of the class EnsLib.HL7.Service.Standard. This bug resulted in an <ILLEGAL VALUE> error when an HL7 service is configured with NackErrorCode=ContentE.

This change corrects behavior by the upgrade installer that makes existing cconsole.log files rw-rw--- (no one but owner and cacheusr can access). The correct permission is rw-rw-r-- (everyone can read the file).

A problem has been corrected that occurred

• when a row was deleted from a table with a statement level AFTER DELETE trigger with TSQL code, and
  
  
• the trigger code deleted a row in other table which did not have a TSQL statement level AFTER DELETE trigger, but did have a TSQL statement level AFTER UPDATE trigger.
  
  

The error may have resulted in the DELETE in the trigger code returning an error, most likely SQLCODE=-29, field not found.

In a rare condition, ECP recovery dropped a series of client requests. It may happen if the server processed a partially received request buffer with many requests and the state synchronization between the app-server and the db-server was wrong.

This is now corrected.

Connection pooling was critically slowed down in multithreaded situations, due to excessive locking. The locking has been changed to improve performance.

This change corrects an error in the Java binding. Under some circumstances, the binding code could release locks that it should have held.

The semantics of FOR SOME %ELEMENT have an implicit DISTINCT BY %ID for the corresponding stream. The parser inserted the implied DISTINCT, but the location of this code caused it to apply only when the master map was also accessed for some reason. The code has now been moved, ensuring that it will be executed in this case, and whenever a single-collection-map strategy is chosen for a stream by the optimizer.

The Caché version 2008 server only supports version 1 of the class index cache which does not contain class timestamps in the class index. This means Caché is unable to detect a class has been changed on the server. This change forces a reload of the class definition once when an attempt is made to load a class definition. The change also causes Caché to properly handle locked classes.

GetClassDefinition now checks whether the class index is up-to-date before beginning processing of the class in Studio.

On a Unicode system, clicking on Call Stack to get variables causes Studio to hang forever. This is now fixed.

If a TRY/CATCH construct catches an error in a subroutine or method that is invoked as part of building the argument list for another subroutine or method, there would be a <SYSTEM> error when the subroutine with the CATCH quits. This has been corrected.

This change fixes a rare condition that could cause a memory access violation when a process halts after using many objects.

This change corrects a problem that could cause a crash when executing a MERGE command in low memory conditions, especially on 64-bit Windows platforms.

This change fixes a problem on 64-bit Windows that could cause a system crash when compiling a very large number of classes.

This change corrects a problem that occurs on OpenVMS where an asynchronous terminal disconnect was not handled properly possibly resulting in a process termination.

A bug has been resolved in loading a GIF file through a CSP interface, or code which emulates a CSP interface, which sometimes reported a <DATABASE> error. There was no underlying database error, the code improperly referenced a process-private global when a normal global shold have been accessed.

This change fixes a problem where an attempt to enlarge the symbol table size following a <STORE> error failed to increase the memory available to the process because there was no location to hold the new memory. The change ensures that there is always at least one location available for this purpose whenever memory is expanded.

This change fixed a problem that reverse $ORDER on a subscript-level mapped global could return a wrong answer.

A problem has been resolved which could result in some databases not being restored from the write-image journal during cluster failover. This does not affect the restore during Caché startup, just during the recovery performed by the cluster master during failover.

Since the journal data is also restored during cluster failover this problem does not result in data loss. However, it can result in database degradation which could show up during the journal restore phase of cluster failover. It also could show up some time later such as when the degraded block(s) later referenced in circumstances involving block splitting.

A problem has been corrected which could cause TCP communication with SSL to hang the system if the SSL connect or accept blocks while the process holds a global block, and the write daemon requires that block to continue.

In releases prior to 2009.1, if open or incomplete transactions were detected during a journal restore, the operator would be asked at the end of the journal restore whether to have the open transactions rolled back.

This behavior has been changed. Now, the operator is not given the rollback option. Transaction rollback is not performed, even if there are open transactions detected during the journal restore.

This fixes an undefined error in ^INSTALL when upgrading from version 5.0. It would only occur if the first database defined in the CPF file (in alphabetical order) was either an ECP or DCP database.

This change fixes an inconsistency in processing mapped %-globals. in an overlap between two range such as:

%Z:%ZeUTT -> DBA
%ZeUTT -> DBB
%ZeUTTX:%a ->DBA

the mapping of %ZeUTT would be mapped to %SYS instead of DBB

If the user cancels in the server selection dialog, the error code was misreported. This has been fixed.

This change improves the protection in, and resilience of, the code responsible for determining the requested file type in the Apache DSO modules. It enables the CSP module to cleanly reject requests destined to be processed by the WebLogic module (provided the 'CSP On' directive is not specified for the path).

This change adds protection to prevent the Gateway exceeding the current maximum file size in force for the Event Log file (CSP.log). Failure to do this check can result in the CSPnsd process terminating abnormally during normal operation and/or at initialization (since it cannot write the startup entry in the log).

This change prevents the removal of errors when ##class(Ens.Purge).PurgeAll() is called.

This change quotes the characters ".", "(", and ")" in Message Browser generated queries against message body fields. This allows sub-properties and array elements to be specified without generating invalid SQL field names.

This change prevents an infinite loop that would occur if HL7 input was malformed and random data in the message happened to contain the sequences "MSH", "BHS" or "FHS" which the HL7 parser looks for in attempting to recognize HL7 data with unspecified "Flexible" framing.

This change fixes an error that resulted in invalid time codes being constructed for timestamp fields in EDI reply documents. The 4-digit time was always generated as "1708" instead of the actual time.

This change adds support for SQL -> syntax in the Message Browser when using detailed filter conditions of type Body Property. When the selected message body class has properties whose type is a persistent object class, the Message Browser's Value Selector popup dialog box shows the values of all the nested sub-properties, with a dot ('.') separating them from their respective main parent properties.

Formerly, selecting one of these nested sub-properties would cause invalid SQL to be generated and thus cause a page error upon clicking the Search button.

This change causes the Message Browser to generate SQL arrow ('->')syntax for references to such sub-properties, resulting in SQL that is valid if the sub-property belongs to a persistent class type.

This change also generates a valid SQL 'AS' name for the column displaying the sub-property values. It does this whether the sub-property is named using a dot separator or if it is named using an arrow separator entered by hand.

Windows installation will now check for SSE2 support in the underlying CPU and will display a warning that only client components will be installed. It will enforce a "client" setup role if the SSE2 instruction set is not available.

Before this change, interrupting an install via Ctrl-C on some Unix platforms could corrupt the install kit if it was on a writeable file system. The interrupt behavior is now correct.

In previous releases, the "InProcServer32" default registry value for all classes exported by mscomctl.ocx was not populated when the "activex" feature was not installed. This isnow fixed and this value is populated whenever the mscomctl.ocx component is installed.

CP establishes its connection on the first reference to any remote global. While establishing a connection, if the ECP client daemon creation fails, it may leave the connection in "in progress" state.

This change detects the errors associated with this situation and attempts to recover from it. It also attempts to fix itself when daemon creation fails.

This change fixes a rare condition where a long string could be dropped or lost if the database server was restarted right after the last data transmission, but before the journal became durable.

It also fixes issues with long string SET command recovery across ECP. If the server or connection fails in the middle of a SET involving a long string, the assignment may fail with <NETWORK> error or the SET may be dropped altogether.

In rare cases, when the ECP work daemons cannot keep up with activity, and there are many activities queued, the session that originated the work in the background may go away. It may also reconnect with leftover activity from the previous session. This modification prevents any old or stale work from being processed after a disconnect (and/or reconnect) by removing it from the work queue during session cleanup.

Under very rare conditions, if

• The ECP server is a different endianness than the application server;
• There is a SET in the process of being propagated to the client;
• The SET involves a non-compact Unicode string;
• The block that the SET will modify has arrived at the client;

then the cache may be updated with the wrong endian value.

This change corrects a situation that caused memory corruption when the binding server was used by Java.

This change corrects a situtaion where the SQL processor would ignore fields that could be used to optimize the query processing when those fields occurred in subqueries.

This change fixes a situation where an <UNDEFINED> error could occur while running a query that involves collection fields or when row-level-security is used.

A problem has been corrected that would cause an <UNDEFINED> error if you did an OUTER JOIN into a table that had Row Level Security

This devlog fixes an error that caused a <SUBSCRIPT> error to be raised at prepare time when DISTINCT was used in combination with an ON clause, as exemplified by this query:

SELECT DISTINCT b.id AS id
FROM sample.person b
LEFT OUTER JOIN sample.employee v ON b.id = v.spouse
LEFT OUTER JOIN sample.employee uv ON v.id = uv.id
AND uv.home_state =
    (select office_state from sample.employee where v.id=uv.id)

This change fixes an issue internal to the SQL compiler where a string was stored that exceeded the maximum subscript length on an array.

This change fixes a defect in SQL optimization that, though rare, could potentially occur in any moderately complex query resulting in incorrect results.

Certain quantified or EXISTS subqueries involving aggregation within a FROM clause subquery of the given subquery could have much worse performance in later releases compared to Caché 5.0.4. This rare regression is fixed by this change.

This correction changes Studio behavior on start up to remove class cache files with different build numbers. These files was caused a problem if two Studios from different builds were run simultaneously in the same instance.

Studio passes the namespace the Studio template page should run in as a URL parameter on the initial page. This is then stored in the %session object and used to set the 'RunNamespace'.

However, the code that then switched this page into this namespace was not using the value passed in, but the value from the %session.RunNamespace which was incorrect at this point. This change uses the correct value.

It appears when you have password authentication this is not a problem, but it does effect systems that are using the Unknownuser.

This change fixes a problem in Caché for Windows/x64 Unicode that could cause the process to crash if the user tried to set a global whose name included a wide character. The correct behavior is to get a <WIDE CHAR> error.

Some internal OpenVMS data structures changed between OpenVMS 8.2 and 8.3. This in turn prevented the Caché handler which deals with process termination when a connection is lost (eg. telnet window is closed) from properly dealing with this condition. The result would be a process stuck in Caché in a COM state at priority zero. This change resolves this problem.

This corrects an issue where signalling a message for an undefined event, or undefined process id will cause the event to be allocated but never used. This could cause a memory leak.

This correction fixes a possible access violation crash during Caché upgrade on OpenVMS alpha platform if an empty 'Z' routine exists in the CACHESYS database.

This changes correct a situation where a new ECP master could hang while waiting for slave to establish the ECP connection.

This changes corrects a problem where $ZF functions were calling out to external routines at the wrong privilege level.

This corrects a situation where a process that HALTs could generate an access violation error if there were XDEV devices open at the time the process stopped.

This submission addresses an issue that caused journal restore to fail if switch 10 is set (e.g., as continuation from database restore where switch 10 is set).

This change addresses an issue where journal restore with selected globals on Unicode systems fails to restore the specified globals.

This corrects a situation that could cause the %SYSTEM.License.LicenseServerKeyList() query to fail. The problem was reproducible the first time the query was run following system startup, but could not be reproduced on subsequent invocations.

This change corrects an ECP lock inconsistency during ECP recovery when ECP server restarted.

The [LAT] section of the config.cpf file is incomplete in Caché versions 2009.1 and 2010.1. The System Management Portal does not configure the advertised services (ServiceName, ServiceDescription, ServiceRating). This change alters the LAT daemon (lat.exe) so it reads its startup parameters from a lat.ini file in the installation directory instead of from the cache.cpf configuration file.

Users can copy the [LAT] section from an existing cache.cpf file into the new lat.ini file or you can enter the information with a text editor.

When using OS authentication in combination with delegated authentication and Caché Password authentication, if the ZAUTHENTICATE routine returns the error $$$OSDelegatedAuthenticationFailed, the security subsystem is supposed to NOT continue on and prompt for username/pw using Caché Authentication. It is supposed to audit the login failure, and halt. This was not happening, the user was then getting prompted for a username/pw.

This change implements the correct behavior.

This change allows a valid routine name range mapping to end in a ".". For example,

Map %zabc:. to DB1

When converting to 2009.1.x, it could cause an installation to fail as mappings could be generated in the CPF file to remove overlapped mapping ranges.

This change fixes a problem with using embedded SQL on OpenVMS only when using it with the %SYS.ProcessQuery class. For example, the embedded SQL command

&SQL(SELECT * FROM %SYS.ProcessQuery)

would fail on an OpenVMS system, but would work fine on Windows and Unix.

This fixes a problem when logging in on a com port, that the W /mnemonics commands would not work, and would cause a <MNEMONICSPACE> error to occur.

This fixes a bug in the CPF conversion routine where if there is an overlapping routine mapping as follows:

%ZDTOOL==>DB1
%Z*==>DB2

The mapping is not converted to

%Z:%ZDTOOL==>DB2
%ZDTOOL==>DB1
%ZDTOOL.:%a==>DB2

When Caché is configured with an invalid login limit set to something other than 0, if a user account hits this login failure limit, Caché would cause the process to HANG for invalidloginattempts-loginfailurelimit seconds.

Thus for every failed login attempt, Caché would hang 1 second longer, until a valid login attempt would occur. On an account that could generate hundreds of invalid login attempts, the hang could last a long time. The code has been updated to hang for a maximum of 20 seconds.

This updates the routine mapping conversion at the end of the "%" range so that if you have a routine mapping such as:

%zquiet* -> DB1
%z* -> DB2

It will convert this to

%zquiet* ->DB1
%z:%zquiet ->DB2
%zquietu:&->DB2

rather than

%zquiet* ->DB1
%z:%zquiet ->DB2
%zquietu:.->DB2

When the last mapping was loaded it would be ignored by the system, and the mappings would go to the CACHESYS database rather than DB2.

When converting to 2009.1, the configuration settings for the Windows LAT daemon are now stored in the lat.ini file rather than in the CPF file. Users who wish to customize their LAT setting should now edit this file.

When auditing is turned on, and direct mode commands are being audited, if the user entered a "!" at the command prompt, they would be blocked from entering the Unix or OpenVMS shell. This problem has been fixed.

This change fixes a problem where, when calling the API method Config.CPF.Activate(), not all the namespaces would get activated, only the one last modified.

This change fixes a situation where %-routine re-mapping could result in the wrong destination database.

This change corrects a situation where a report display that compiled in 2008.2 failed to compile in 2009.1.

This change modifies the compiler/link settings such that universal bundles will be built for the x86-64 Mac OS X platform. The previous settings prevented the Management Portal from working properly on 64-bit systems.

This change restores the transaction level in user error traps.

When application code errs with a transaction level > 0, Caché does a TROLLBACK in the error trap. Otherwise, the transaction level will never go down to 0 and this will cause inconsistencies in processing afterwards.

This change added TROLLBACK for the user callback in Business Services, Processes and Operations.

This change fixes the SQL queries used in the Ens.MessageHeader Purge() method when that method is called with the argument KeepIntegrity = 1. Because of a subtle SQL syntax problem the previous form of the query did not correctly identify rows to purge that should be purged.

This change improves the previous truncation algorithm used for turning query strings into cache table keys. It now uses a shorter truncation limit in order to avoid <SUBSCRIPT> errors even when the allowed subscript length is halved due to the text containing Unicode characters.

This change also fixes three other places where similar cache table subscripts are used.

This change fixes three issues with SQL Large Object support:

• Large Object detection was not recognizing certain SQL Server types properly.
• JDBC calls to the removeLOB() cleanup function were missing an argument and therefore failing.
• The Large Object byte position appendage to the field ID in GetData/GetDataW() was not being tested properly, causing a failure to fetch all field data.

This change also adds Java type definitions to EnsSQLTypes.inc and adds new methods each to EnsLib.SQL.GatewayConnection and EnsLib.SQL.Snapshot:

• GetColumnSQLType - Returns the SQLtype integer of column 'pColumn' in the result set.
• GetColumnIsLOB - Returns whether column 'pColumn' is a LOB (Stream) column.
• GetColumnIsBinary - Returns whether column 'pColumn' contains binary data.

This change fixes a bug that causes the SQL Adapter to eliminate some or all output result snapshots generated from a stored procedure call if the SP also had any scalar output parameters.

This changes forces Ensemble to respect the default CSP application timeout when opening DTL in Studio. Previously, this was overridden to be set to "never", which caused CSP session licenses to be held in perpetuity; only manual intervention or an Ensemble restart would release the licenses.

In version 2009.1, a Windows installation on non-English systems would create a Start Menu shortcut in the Startup folder as "Launcher" rather than <INSTANCENAME>. As a result, after a reboot users could have two cubes for the same instance, if it was an upgrade from 2008.2 or below. This change fixes the problem.

This change corrects a problem when adding two large numbers. If each number has 19 digits, and each has a single digit after the decimal point, and the sum will not fit in a 64-bit signed number, the application can get a <SYSTEM> or other error when trying to save or use the result.

This change corrects an issue where, in rare conditions, ECP daemons might be out of sync and search for locally cached blocks. This caused an unexpected delay in responding to various requests.

Sometimes, when processing a client request, the server did not send cache updates for properties that were modified on the client. Thus resulted in excessive calls to the server instead of getting the value from the cache.

This change forces the client to assume that after a roundtrip to the server the client cache must be valid, even if the server didn't send any updates. As a result, the client should not send extra messages to the server.

VisualBasic and C# have slightly different rules regarding how methods are shadowed and/or overloaded. This change adjusts for the differences in the two languages thus removing VisualBasic compiler errors.

Prior to this change, the .NET code generator did not report compiler errors when generating an assembly; it always reported a successful completion even when an assembly was not generated. With this change, any errors or warnings that occur are reported back to the user.

This change corrects a problem where connection pooling by multiple Java applications accessing the same Caché data could result in a deadlock.

This correction resolves an issue where large integer values were not being properly converted from NSNumbers into d_decimal.

In some cases $DOUBLE values could be stored in these datatypes. With this change, the proper datatype is used.

On HP-UX Itanium, virtual memory was not being deallocated correctly. This became apparent when using %XML.Adaptor::XmlExportToString(). This change corrects the deallocation issues.

The bug was fixed that the delimited ids setting was not honored in joins.

A problem has been corrected where the use of {%row} in %CacheSQLStorage definitions would cause a class compiler failure.

When inserting or updating a row into a table that has a foreign key, Caché acquires a lock on the referenced row that is released when the transaction ends. This lock is required to check the referential integrity of the foreign key constraint and ensure the row still exists from the time the lock is acquired until the time this row is inserted or updated.

If there are many rows inserted or updated in a single transaction and many foreign keys on the table, this could hold many locks in the lock table.

With this change, Caché will no longer hold the lock for the duration of the transaction, but will release the lock after the row has been inserted or updated. By this time the system has ensured referential integrity and the row in the referenced table cannot be updated or deleted during the course of this transaction because the referential action on the foreign key constraint will prevent the update or delete on the referenced table from occurring.

With this change, Windows installation will set the username and password to blanks when a minimal security setting is used to install a product.

This change corrects a memory leak that occurred under certain uses of $LISTBUILD, $LISTGET and $LISTGETNEXT.

This change corrects a problem that could cause a process to hang when trying to read text pasted into the input stream that uses a multi-byte encoding such as UTF-8 or JIS.

This change allows Caché to properly recognize when it is running on the MAC OS 10.6 ("Snow Leopard") kernel.

External programs started with $zf(-2,<path>) on Unix platforms now run with the standard file descriptors 0, 1 and 2 redirected to /dev/null. Previously these three descriptors were closed before starting the program.

This correction resolves a defect where a process could get blocked in a USE command by the operating system and cause Caché to hang if it had retained ownership of a global buffer.

This change fixes a possible <NAMESPACE> or <DIRECTORY> error that could occur while the system is re-activating the namespace table while a namespace is edited.

$ZUTIL(193,x,1) converts a date X to UTC. If X represents the first day of the year, where (year mod 4)=1, the result might be incorrect. This bug has been fixed.

This change fixes a problem where a core dump can occur when running the INTEGRIT utility.

A problem has been corrected that could cause the Caché cluster resource to fail.

The symptom is that the cluster manager would become unresponsive and the Caché cluster resource will not work. This problem has been seen on 64-bit Windows, but may also occur on 32-bit Windows clusters.

This correction addresses an issue that could cause READs on a device opened with /GZIP (i.e., with stream compression/decompression enabled) to fail to read data. The problem is most prominent when the READ length is 32767, the default for stream READS without terminators or length specifier.

Trying to OPEN a TCP device with the letter "Q" in the hostname was failing with an <UNIMPLEMENTED> error. This was due to a faulty check for a "pipe" on OpenVMS. It has now been corrected.

The Caché Diagnostic Report was using an unsupported option for the process list on HP-UX. This has been corrected.

This change corrects an issue that could cause a shadow server to produce corrupted copies of source journal files on OpenVMS if process default I/O table is not RAW (for example, the I/O table is EUC in locale 'jpuw').

The problem is believed to be present in all supported Caché releases.

When a namespace is now deleted, all the cached mappings for it are also deleted. Previously, when the same name mapping was re-added, the cached mappings would be applied to the namespace.

With this change, the comm port server process now start and will handle comm port logins. Previously, the comm port server process would not start.

When upgrading to 2009.1, if an ECP server name was defined to be an IP address of the form n.n.n.n (for example, 127.0.0.1, the loopback address), the installation would fail during the conversion of the CPF file. This has been fixed.

Previously, if there was a problem converting the CPF file during an upgrade, the installation would fail. The only recourse was to correct the CPF file, then re-install.

Now the installation will complete, allowing administrators to fix the CPF file, and restart to be upgraded to the new version.

This fixes a problem where the system would not start if it was using DDP, and the DDP device name was left as null (which means use the default device.)

This change fixes an issue where a user overrode a system default mapping such as %Z* routine resulting in the wrong destination for the %Z* mapping.

This change fixes a situation where an attempt to activate mappings resulted in an <UNDEFINED> error. This occurred when there were many mapping entries and internally Caché attempted to re-index the mapping array.

This change ensures that escaped Unicode characters are properly extracted and passed to the function that un-escapes them and converts to UTF-8. Previously, this issue could prevent query strings with embedded Unicode characters from being decoded correctly.

These changes update the following tutorials:

• Web Applications Tutorial (TWEB) to show Web Form Wizard launched from Studio Templates Menu
• Screen Captures for 2009.1 in Caché QuickStart Tutorial (TQKS)
• CSP Tutorial to show launching Web Form Wizard from Studio Templates
• CSP tutorial example to add warning so users do not use code as a CSP login page

The datatype Ens.DataType.Class has a parameter EXTENDS that allows the user to require that the value of the type extend the indicated class. Before this change, however, if you tried to use the EXTENDS class as the value, then the validation would fail because a class isn't considered to be derived from itself (which isn't really proper polymorphic behavior).

When upgrading Ensemble, the upgrade process overwrote the CSP application definitions for each of the Ensemble-enabled namespaces. In case the user had modified settings such as resource required, matching roles, description, timeout, these modifications would be lost upon an upgrade.

This change corrects that situation. Now, if the CSP application for a namespace's Ensemble Portal already exists, the upgrade process opens its definition and only sets the following properties, keeping the previous values for all other properties: CookiePath, LockCSPName, NameSpace, Path, Recurse, ServeFiles, UseCookies, LoginPage.

This change corrects a problem that can occur on a Spanish system during an Ensemble upgrade. The collation for Ensemble globals with subscript level mapping from ENSLIB was not set correctly and could cause a collation mismatch error depending on the default collation for the database.

This would only happen on systems whose default collation is not Caché Standard, and in namespaces created during installation or post-installation such as ENSEMBLE (but not ENSDEMO which is created later in the process).

This correction adds new entries for all Ensemble globals mapped from ENSLIB with settings Collation = 5 (Caché Standard) and IsKeep = 1. It also properly handles existing local data that may exist in the local database for the globals that are mapped at the subscript level only.

This change fixes a bug that causes a stream storage for low-order byte data retrieved by the SQL Adapters into a Snapshot object to be killed improperly. The stream data in question is stored in the ^CacheStream global.

This change fixes a defect introduced with IPv6 parsing in the TCP Inbound Adapter and its subclasses. The bug results in an invalid IP address when a port is specified in both the Port setting and as part of the AllowedIPAddresses setting.

The correct behavior when port is doubly specified is for the Port setting value to be ignored and the AllowedIPAddresses value to take precedence. This change restores the correct behavior.

This change fixes errors in the method ImportFromResultSet() in class EnsLib.SQL.Snapshot that prevented it from working, in particular the use of an undefined variable tGblRef.

This change enhances the parser used to determine which Stored Procedure arguments are being used as parameters in an invocation of ExecuteProcedure() on the SQL Inbound or Outbound Adapter.

This change also improves disconnect detection around calls to DropStatement()

This change fixes the accessor methods used by X12 to get TypeValue and Identifier calculated property values in the case where the document is an X12 Group, that is, starts with a GS segment. Formerly these fields were extracted as if the document were a TransactionSet starting with an ST segment; this yielded the wrong field values.

This change makes the initialization methods return status code values and makes the methods that call them check those return values consistently. This improves behavior when an associated JavaGateway configuration item cannot be located for an SQL Operation or Service configured for JDBC.

Internet Explorer 8 has a bug that causes it to acquire a wrong size in frameset.rows for the first frame in a frameset.

IE8 bug report can be found at: http://connect.microsoft.com/IE/feedback/ViewFeedback.aspx?FeedbackID=432864

It affects the Ensemble Management Portal as most of the pages contain frames.

There were other changes to workaround this situation. This new version also introduces a new behavior regarding resizing of the title frame, regardless of browser version.

In case the title frame size is reduced we will keep the chosen size. In case it's enlarged, it will be set back to its original size of 100 pixels. There's nothing below so it's pointless to have it larger than 100 pixels. This makes the workaround easier and also it seems to be better than the previous behavior.

The behavior is also in effect for IE6 and IE7 in order to have all users see the same regardless of browser version.

Before this change, a previous correction modified the queries generated by the Message Browser page when there is only one simple equality SearchTable condition. This change fixes that modification to work when there is only one *enabled* SearchTable condition (and it is a simple equality condition). Disabled conditions no longer count.

This change fixes a case where the variable nActions is defined after it is used. This change also removes an obsolete property %ReplyActions.

This change now enables the Perl/Python bindings to be installed on MACX64

The regular expression used by the CSP gateway configuration in Apache's httpd.conf file has been modified to work around a bug in the Mac OS X 10.4 provided Apache.

This change alters the custom setup accelerator key in setup role dialog to "u" to avoid conflict with "Server" setup type key.

This change corrects an access violation error in the setup bootstrapper that was happening when instance name was longer than installation directory name.

This change updates the Microsoft VC2008 SP1 redistributable packages to include ATL security update.

If a breakpoint was set on the line immediately following a ;;comment, the breakpoint would be ignored. This is now corrected.

This change makes the MultiValue spooler more robust when it comes to handling printing with transactions, notably transaction rollbacks.

All MultiValue print job updates are now performed outside of a transaction. This means that any transaction rollbacks that occur in the application will not affect the print job in any way and the data in the print job will remain intact and uncorrupted.

The PICK emulation for KEYIN() was incorrect in that it was echoing according to the ECHO ON/OFF status. In this emulation, it should never echo. This change implements that behavior.

This changes corrects problems in the Management Portal page:
[Home] > [Configuration] > [Device Settings] > [Devices]
namely,

• When editing device, some of the property values may not be loaded when using IE8.
• The "SubType" drop-down list may not be populated, and the value not shown as selected.
• After deleting a device item, it returned to the Device Settings page. It should stay on the same page and just refresh the tables for devices, magnetic tapes, and subtypes.

This change corrects an error in the TQSL parser that interpreted a name beginning legally with "%" as an expression involving the "%" operator.

This corrects a typographic error in the IsString method of %SAML AttributeValue that negated a validity check.

This change fixes a problem where a javascript error may occur when the wizard tries to display a warning message and, (A) the user clicked the Finish button without first selecting a row, or (B) the user selected more than 10 rows.

Another change added support for a new d_decimal type for %Numeric and %Decimal types to be projected via the C++ binding.

Since the Objective-C binding uses the C++ binding for its underlying connection to Caché, it was necessary to make additional changes to the Objective-C binding to project d_decimal types as NSNumber (actually, a subclass NSDecimalNumber).

This changes corrects an error where an expected value from a method call was not returned.

This change corrects a Role element of SOAP 1.2 fault which was not properly terminated.

The nThreadsPerCore method of the $SYSTEM.CPU class was misspelled as nTheadsPerCore. This fixes the misspelling.

This change alters the %XML.Reader to return SAX parser validation errors. The SAX parser was reporting the error, but there was a mismatch between %XML.Reader and the Caché interface to the parser.

This change corrects an error introduced in 2009.1 where return value of executeUpdate was assigned to %ROWCOUNT. The error manifests itself as the absence of error messages for an unsuccessful operation.

An optimization was added identifying cases where fields in certain views mentioned in a query were not actually essential to processing the query, resulting in improved run-time performance for those queries.

A correction was made to the code-generation of certain queries that use a multi-index access strategy, for example queries that use the %INLIST comparison operator, or queries of the form:

    SELECT id FROM t WHERE x=:a OR (x IS NULL AND :a IS NULL)

where t.x has an index and t has a multi-field IDKEY.

A problem has been fixed where an <undefined> could occur during execution of query that had two or more WHERE clauses, each of which has an OR that is subject to an OR to UNION optimization.

For the problem to occur, one of these OR conditions must be present in subquery of the query/subquery containing the other OR.

It is also possible this problem could arise with only one OR to UNION candidate. In that case, the query with an OR would have to also contain a subquery against an irreducible FROM clause subquery that had outer references to the query containing the OR.

This change implements the following: when a user disables syntax checking (Options->Editor->Syntax Check and Assist->Enable Syntax Checking "unchecked"), all syntax check parser related actions, including studio assist, tooltips and context sensitive menu for particular syntax element will be disabled.

Turning syntax checking in this way avoids unnecesaary work by Studio when scrolling through a large document.

This change correctly identifies when HyperThreading Technology is activated on recent Intel x86 chips (Nehalem-based), and how many cores are present.

This change corrects the behavior of object export. The multidimensional flag is only set if the property size is greater than zero. This change allows the xDBC connections to optimize their handling of properties only when they actually change.

This change disables the feature that returns unused space from a database to the filesystem. An attempt to invoke this functionality will have no effect, that is, no space will be returned.

This will remain disabled until there is a correction for a condition where an attempt by different processes to truncate and to expand the same database at the same time results in a hung system.

This change corrects a situation, present only in 2009.1, that could cause Caché to halt with an access violation error when a process halts leaving an XDEV in an open state.

The use of inet_ntop() to translate IP addresses was introduced in the system code when IPV6 support was introduced in Caché 2009.1. Checks for a zero return value were not made everywhere, which could lead to an access violation if inet_ntop() could not translate the binary address to display format. This problem has been corrected.

This change now makes starting up Caché work with a threaded Callin application.

This change corrects a situation where an attempt to restore a journal involving an unjournaled database could result in an <UNDEFINED> error.

The problem was introduced in 2009.1.0.

This change fixes a problem where the delock (L -^A) command might skip its operation and leave the lock in lock table.

This changes addresses a defect that resulted in an error like the following:

<UNDEFINED>zSourceEventsExecute+29^SYS.Shadowing.Shadow.1 *req("4"))

when the user started a non-cluster shadow after specifying the name of the shadow to start, but before giving a list of Source Events to choose to start shadowing at.

The problem is believed to be present in Caché 5.1 and forward (all releases containing HYY1119).

These changes properly handle the conversion of mappings present in a 2008.2 (or earlier) configuration file to an equivalent set of mappings in 2009.1. The major features of this conversion are:

1. All mappings (globals, routines, subscript-levels) are subject to the same rules.
2. Any overlaps found are converted into an equivalent, order-independent sequence of non-overlapped mappings.

This change corrects an error in end-of-file checking when reading the CPF file that could result in an infinite loop.

© 2009, 2010, InterSystems Corporation. All rights reserved.