Changes since version 2.0.0 --------------------------- - Added support for g++ builds on Solaris. - Fixed IcePack node and registry database environment names. They have been changed to be the name of the node for IcePack nodes and 'Registry' for the IcePack registry. This means that you can now configure the database environment using the properties prefixed with Freeze.DbEnv. or Freeze.DbEnv.Registry. - Fixed a bug where the endpoint cache for indirect proxies wasn't always cleared upon a communication failure with an object. - Fixed a race condition in the IcePack on-demand server activation code which could cause clients to hang under rare conditions if the server couldn't be activated. - Fixed a small memory leak in the IcePack node that would occur each time a process is forked (Unix only). - Added 'object list' and 'object describe' IcePack admin commands. - Changed the signature of IcePack::Admin::removeObject() to take the identity of the object instead of its proxy. - Connection validation now checks for the timeout specified by the property Ice.Override.ConnectTimeout and will raise Ice::ConnectTimeoutException if the connection validation times out. - Fixed an assert that could occur during connection establishment if no more file descriptors are available. - Improved support for on-demand server activation in IcePack. - Improved the IcePack registry to prevent potential deadlocks caused by thread starvation. - The Ice manual stated that the IcePack configuration property IcePack.Registry.Internal.Endpoints was only necessary when nodes were being used, but in fact the property is required regardless of whether nodes are used. The manual has been updated accordingly. - Fixed a bug where Ice::ObjectAdapter::activate() could throw Ice::ObjectAdapterDeactivatedException if the object adapter was deactivated concurrently. - Fixed a bug where setting the locator proxy on an indirect proxy could cause an assert. - Fixed a bug in IceSSL when multiple communicators are used. - Added an optimization to the Freeze evictor for the ice_ping operation. The object is no longer loaded into the cache by this operation; Freeze simply checks if the object is in the cache or the database. - Fixed a slice2cpp failure when the --output-dir option specified a directory containing backslashes. - Changed the signature of Ice::Service::main() and Ice::Service::run() to accept a reference to argc (int & argc) rather than passing argc by value (int argc). This is to ensure that, if Ice::initialize() or Ice::Service::initializeCommunicator() rewrites argv, argc reflects the number of elements of the rewritten argument vector. - Fixed an encoding bug in IceSSL. - Fixed incorrect logging message if no more endpoints are available for retry. - Added setDefaultContext() and getDefaultContext() to the Ice::Communicator interface. This allows a default context to be established on a communicator-wide basis. See section 29.9.3 in the doc. Added ice_defaultContext to Ice::ObjectProxy. This creates a new proxy that uses the default context established on the communicator. - Fixed a rare proxy marshaling bug that could cause the receiver to encounter an UnmarshalOutOfBoundsException. - Overloaded the checkedCast member function of the generated Prx classes to allow a trailing argument of type Ice::Context. This makes it possible to do stateful things that require a context in a servant locator's activate() method. - Added catalogs for Freeze: each database environment used by Freeze now contains a special dictionary (named __catalog) that keeps track of all evictor and dictionary databases in this environment. - Fixed a bug in slice2cs and slice2vb: incorrect code was generated if an interface was derived from a base interface in a different module and if the base interface contained an AMD operation. - Changed IceUtil::GCShared to derive from IceUtil::Shared. (Previously, these two classes were unrelated.) This change allows servants to be derived from IceUtil::Thread, which was impossible previously. Changes since version 1.5.1 --------------------------- - Fixed a bug in IcePack which could cause a core dump of the IcePack registry or node when updating an application deployment. - The proxy returned by the object adapter operations addFacet and addFacetWithUUID did not contain the facet. This required the application to make an extra call to ice_newFacet in order to obtain the correct proxy. This has been fixed. - Added the object adapter property .PublishedEndpoints, which specifies endpoints to advertise in proxies created by the adapter. - For WIN32, Thread::getThreadControl() set the thread id of the ThreadControl object to the current thread instead of the thread the control object belongs to. This has been fixed. - Changed Ice::Service to use _exit() in the daemon parent. - Fixed a bug with AMD exception handling, where it was possible that certain exceptions propagate out of ice_response() or ice_exception(), and therefore such exceptions were not handled properly. - Exceptions raised while marshaling data for batch requests resulted in a connection deadlock. This has been fixed. - Fixed a bug in slice2cs: the generated code was incorrect for dictionaries with sequence value types, if that sequence value type was mapped to an array. - Fixed bug in IcePack that would prevent dynamically added adapters from being used again after the IcePack registry was restarted. - Fixed tracing of operation mode. - Fixed bug in slice2cpp: for statically linked binaries under Windows, exceptions and classes could cause an UnmarshalOutOfBoundsException. - Fixed bug in slice2java: with the --impl-tie option, incorrect code was generated for local interfaces and classes. - Tightened semantic check for all Slice compilers: previously, a local interface or class could be derived from a non-local interface or class. This is now correctly diagnosed as an error. - Changed code generation in slice2java: for local interfaces, only an OperationsNC (but no Operations) interface is generated now. (Previously, both interfaces were generated, with one of them being empty.) - The IceSSL plug-in was using regular mutexes in some places where static mutexes are necessary. This has been fixed. - Fixed a bug in the garbage collector that caused a crash in rare circumstances. - Freeze dictionaries now support indices. You can index on the full value of the dictionary, or on a member (when the value is a struct or a class). When you index on a member, you can define several indices (for different members). See the Freeze bench demo and the Freeze dbmap test for examples. - The Ice::Service class no longer resets the umask to 0, but rather uses the inherited umask. - Fixed a code generation problem in slice2cpp. This problem showed up only for the --impl option. - Changed the way C++ keywords are escaped. Previously, a Slice identifier that is a C++ keyword was escaped everywhere. For example, Slice "while" was mapped to "_cpp_while" and "_cpp_whilePrx". With the change, Slice "while" is mapped to "_cpp_while" and "whilePrx". - Fixed the slice2java and slice2cs compilers for operations with multiple exceptions in an exception specification: if an exception was a base of one or more other exceptions in the same exception specification, the code generator sometimes emitted the catch blocks in the marshaling and dispatch code in the wrong order. (This applied only to interfaces with an ["amd"] metadata directive.) - Fixed a bug in the slice2cs code generator: ["cs:collection"] sequence S1; ["cs:collection"] sequence S2; Sequences of Object or Object* caused incorrect code to be generated if they were marked as "cs:collection". - Fixed a bug in the slice2cs code generator: for nested sequences, incorrect code was generated if both the inner and the outer sequence were mapped as arrays. - Fixed a number of bugs in slice2cs that caused incorrect code to be generated if C# keywords were used as Slice identifiers. - Fixed a bug in slice2cpp that permitted impossible exceptions to be received by a client if client and server used Slice definitions with mismatched exception specifications. The client now correctly receives UnknownUserException if the server throws an exception that, according to the client's view of the operation, is impossible. - The documentation has always stated that same-named constructs cannot be directly nested inside each other. (For example, a module `M' cannot contain a constant named `M'.) The Slice compilers did not enforce this correctly up to now for modules containing constructs with the same name as the enclosing module. This has been fixed and now results in a diagnostic. - The slice2cpp compiler now deprecates Slice definitions at global scope: only modules can be defined at global scope. Everything else (constants, classes, interfaces, etc.) must be defined inside a module. For the time being, the compiler issues a warning for each global definition but continues to compile the code. Global non-module definitions will elicit a hard error two releases from now. - Several demos used Slice classes where interfaces were more appropriate. This has been fixed. - Fixed a bug in the Windows IcePack registry project that would cause a compilation error if the registry was compiled before the node. - Fixed a hang in the IcePack registry that would occur if multiple nodes were started simultaneously. - Added new properties Ice.StdErr and Ice.StdOut to redirect the standard error and output streams of a process. Reimplemented IcePack's error/output redirection using these properties. Changes since version 1.5.0 --------------------------- - The relevant Windows project files were fixed to generate Slice checksums. - Fixed harmless warnings emitted by the IcePatch client. - Fixed a bug in IceStorm that caused a subscriber to no longer receive events after unsubscribing and resubscribing to the same topic. - Fixed a bug in the slice2cs code generator: classes with multiple sequence members caused incorrect code to be generated. - Added work-around for clients "hanging" at exit when using IcePack on Windows. Changes since version 1.4.0 --------------------------- - Added support for generating checksums of Slice definitions, enabling peers to verify that they share the same client-server contract. See the manual for more information. - Fixed a bug that could cause an assert or exception in some rare circumstances, if an operation is invoked after the object adapter for the Ice object has been deactivated. - Fixed a bug in the C++ translator that caused compile errors when two Slice classes with the same name appeared in different modules in the same file. - Changed the names of Windows executables to lower case, for consistency with non-Windows platforms. The following executables have been renamed: DumpDB.exe -> dumpdb.exe GlacierRouter.exe -> glacierrouter.exe GlacierStarter.exe -> glacierstarter.exe IcePackNode.exe -> icepacknode.exe IcePackRegistry.exe -> icepackregistry.exe IcePatchCalc.exe -> icepatchcalc.exe TransformDB.exe -> transformdb.exe All library names have also been renamed accordingly. - Changed the default behavior of the IcePatch client to dynamically calculate the unique signatures of local files in order to reduce the number of cache files, and added the property IcePatch.Dynamic to control this behavior. - AIX 5.2/VisualAge 6.0 port. - Changed the --depend option of slice2java. If you get dependencies for a file x.ice that includes y.ice, the dependency line that is written now is "x.ice: y.ice". (Previously, it was "x.cpp: x.ice y.ice".) If a Slice file does not include any other Slice file, no dependency is printed. With some post-processing, this is sufficient to generate dependencies for tools such as ant and make. - Removed IceBox::ServiceBase and IceBox::FreezeService. There's now a single IceBox::Service interface. - IcePack descriptors have been redesigned. See the manual for more information on the new descriptor format. - Some IcePack improvements: * Descriptors are now parsed only by the icepackadmin tool, therefore it is no longer necessary to store the descriptor files on each IcePack node or registry host. * It is now possible to update an application or server deployment by modifying its XML descriptor and using the `application update' or `server update' commands in icepackadmin. * It is now possible to define variables from the icepackadmin command line when deploying or updating an application or server. * An icepackadmin `node remove' command has been added to clean up resources associated with a given node. * A command line option `--checkdb' has been added to the icepacknode executable. When the node is started with this option, it performs a consistency check on its database to remove servers that are no longer registered with the IcePack registry. Changes since version 1.3.0 --------------------------- - Added a fix to the mutex tests in test/IceUtil/thread for FreeBSD. - Modified the Ice::Service class to support the Ice.Nohup property, rename the win9x function to checkSystem, and add more flexibility in configuring services. See the reference manual for more information. - Added unsupported FreeBSD port. - Ice now builds with GCC 3.4.0. - Fixed a bug when making asynchronous invocations on a routed proxy. - Removed slice2wsdl. - Changed the way sequences are unmarshaled to protect against malicious messages with faked very large sequence count values. While even with previous versions of Ice, buffer overruns were impossible, malicious messages could cause large amounts of memory to be preallocated, causing the receiving process to run out of memory. This is no longer possible -- the total amount of memory preallocated for sequences during unmarshaling is now capped at Ice.MessageSizeMax. - Freeze Evictor changes to adapt to the facet redesign, and make several improvements: * The Freeze Evictor now has add, addFacet, remove and removeFacet operations with the same signature and behavior as the Ice ObjectAdapter. * The createObject and destroyObject operations were kept for compatibility with Ice 1.2/1.3, but are now deprecated. * getIterator now takes a facet parameter, and no longer takes a loadServants parameter. * Indices can now be defined for facets (not just the "" facet). * The ServantInitializer initialize operation now takes a facet parameter. * The ServantInitializer is now provided to the Evictor in the createEvictor call. In this way, the Evictor can consider the servant initializer immutable and avoid locking. * Servant initialization is now performed without holding any Evictor locks. As a result, a servant initializer can perform any remote call, except one that comes back to the object being initialized (which would result in a deadlock). * The Evictor is now associated with a single object adapter; this object adapter is provided in the createEvictor call. * The Evictor is now "deactivate-safe": if you deactivate the evictor while other threads are performing other Evictor operations, deactivate() will wait until these operations have completed. * The new keep and release operations let you lock and unlock an object in an Evictor's cache. - Removed a bogus assert that could happen under certain race conditions when a connection timeout occurs. - Facets have been significantly redesigned. * Facets are no longer hierarchical. As a result, FacetPath (which was a sequence of strings) is now simply a string. This is reflected in all APIs that used FacetPath. * There is no longer an active facet map in Ice::Object. As a result, the following operations have been removed: + On proxies: ice_facets, ice_appendFacet. + On Object: ice_facets, ice_addFacet, ice_removeFacet, ice_updateFacet, ice_removeAllFacets, ice_findFacet, ice_findFacetPath. * The object adapter is now used to register facets instead of the active facet map. The following operations have been added to Ice::ObjectAdapter: addFacet, addFacetWithUUID, removeFacet, removeAllFacets, findFacet, and findAllFacets. Please see the reference manual for more information. For naming consistency, the following object adapter operations have been renamed: + identityToServant has been renamed to find. + proxyToServant has been renamed to findByProxy. * This object model design change means that facets are no longer transferred with an object sent by value. * If your application receives an object with facets from an older version of Ice, a MarshalException is raised. This is unavoidable because it is a change to the object model, and not just to the way objects are transferred or encoded. * If your application receives a request or reply with a facet path with more than one element, the run time throws a MarshalException. * If your application receives a proxy with a facet path with more than one element, the run time throws ProxyUnmarshalException. - Ice no longer retries operation calls on RequestFailedException. (ObjectNotExistException, OperationNotExistException, and FacetNotExistException are derived from RequestFailedException.) - Fixed link errors when building Ice without BerkeleyDB in the shared library search path. - Added ConnectionRefusedException as a specialization of ConnectFailedException, to indicate if a connection fails because a server actively refuses the connection. - The documentation claimed that the Ice.ProgramName property was initialized to the value of argv[0], but that initialization was in fact not happening. As of this version, Ice.ProgramName is initialized correctly. - Fixed the slice2cpp compiler for operations with multiple exceptions in an exception specification: if an exception was a base of one or more other exceptions in the same exception specification, the code generator sometimes emitted the catch blocks in the marshaling and dispatch code in the wrong order. Changes since version 1.2.0 --------------------------- - Added FreezeScript, which consists of the tools dumpdb and transformdb for inspecting and migrating Freeze databases, respectively. - The IceStorm server now uses a single database for all topics, which makes topic creation much faster. - Added the virtual member functions ice_preMarshal and ice_postUnmarshal to Ice::Object. The default implementations do nothing, but subclasses may override them to take special action before marshaling and after unmarshaling, respectively. - Added the demo/IcePack/simple example. - Fixed a bug in IcePack that could cause the node to crash if a server failed to start. - IceBox services deployed with IcePack are now loaded in the order they appear in the XML IceBox descriptor. - Connections are no longer closed when the last proxy using the connection is destroyed. Doing so is error prone: * Quite often, proxies are created on the fly, resulting in connections being opened and closed all the time. This is especially true for services that receive a proxy and data, and then forward this data using the proxy. * Often, connections are stay open for too long, because proxies are kept even though they are never used again. * It doesn't work well with AMI requests, because the AMI callback objects keep track of connections directly. This would mean that if a process only uses AMI requests, a connection is opened and closed for each request, as each request typically has its own AMI callback object. Instead, ACM (Automatic Connection Management) is now enabled by default, with a default value of one minute. This means that idle connections are closed after one minute, regardless of whether there are proxies using the connection or not. This closing of connections is transparent, i.e., closed connections are automatically reestablished in case they are needed again. Please see the description of the Ice.ConnectionIdleTime property for details. - ACM has been completely reworked. It now works properly with respect to retrying failed requests. - Added the IceSSL.Client.IgnoreValidPeriod and IceSSL.Server.IgnoreValidPeriod properties. See the description in the manual for details. - Added the IceBox.LoadOrder property, which specifies the order in which IceBox services are loaded. - Added support for the Intel C++ compiler (v8.0) on Linux x86. - Removed the Ice.Daemon, Ice.DaemonNoChdir and Ice.DaemonNoClose properties. Support for Unix daemons and Win32 services has been integrated into the Ice::Service class. - The default thread pool size is now just one thread. This is the fastest possible setting, still allows one level of nesting, and doesn't require that servants be thread safe. (Please see the description of the thread pool properties in the manual for information on how to increase the number of threads.) - HP-UX port. See INSTALL.HP-UX. - Added extra file integrity checks to IcePatch::Client. - Fixed a problem with servers not shutting down properly under certain circumstances. - IceUtil::Thread now calls srand() in each new thread (Windows only). - When unmarshaling a string sequence parameter, Ice was not clearing the vector before appending elements to it. This has been fixed. - Fixed a rare connection deadlock, that could happen if lots of long messages are sent rapidly in parallel, using separate threads or AMI. - Fixed a race condition that could lead to a deadlock if a thread tried to join with another thread from within the destructor of a servant or class. - Added support for IcePack and Glacier on Windows platforms. The IcePack Registry, IcePack Node and Glacier Starter programs can optionally be installed as Windows services. See the Ice manual for more information. - Added a new logger implementation that uses the Windows Event Log. To use it, set the Ice.UseEventLog property to a non-zero value. - Three new icepackadmin commands: * server signal NAME SIGNAL Causes the IcePackNode to send a signal to a server. (Unix only) * server stdout NAME MESSAGE Writes MESSAGE on server's stdout. * server stderr NAME MESSAGE Writes MESSAGE on server's stderr. - Two new IcePack.Node properties: * IcePack.Node.Output= If set, the IcePack node will redirect the stdout and stderr output of the started servers to files named .out and .err in this directory. Otherwise, the started servers share the stdout and stderr of the IcePack node. * IcePack.Node.RedirectErrToOut= If is set to a value larger than zero, the stderr of each started server is redirected to the server's stdout. - Added Slice interface Ice::Process in slice/Ice/Process.ice. This interface enables IcePack to properly shut down a process without relying on signals, which is necessary for successful operation on Windows. - Added setServerProcessProxy to Ice::LocatorRegistry. - Added new properties .RegisterProcess and Ice.ServerId. If RegisterProcess is defined, the object adapter creates a servant implementing the Ice::Process interface and registers it with the locator registry using the server id defined by Ice.ServerId. - Added the "register" attribute to the element in the IcePack descriptors. If true, the RegisterProcess property is defined for the object adapter. - Added getLocator to ObjectAdapter. - Eliminated Glacier's dependency on the crypt library. It now uses an equivalent function from OpenSSL. - Fixed a bug that could cause a crash if an application used Ice::Application and was interrupted by a signal, resulting in two calls to Communicator::destroy(). - Replaced interface Glacier::PasswordVerifier with Glacier::PermissionsVerifier, property Glacier.Starter.PasswordVerifier with Glacier.Starter.PermissionsVerifier, and changed Glacier::Starter::startRouter() to throw PermissionDeniedException instead of InvalidPasswordException. - Fixed a problem with the random selection of SSL endpoints. - Ice is now supported on Windows 98 SE. See INSTALL.WINDOWS. - Fixed an incorrect complaint in the Slice parser about a change of meaning for enumeration constant definitions. - The glacierrouter and glacierstarter processes used to abort if given a non-existent config file. - Changed Ice::Exception::ice_name() to return const std::string& instead of returning a string by value. - Added support for "global" metadata, which can only be specified once in a file and must appear before any definitions. The syntax is similar to that of local metadata, but with an extra set of brackets: [["metadata1","metadata2","etc."]]. The first use of this metadata is for specifying a package in Java. - The IcePatch server no longer changes the working directory. This avoids potential problems with core files being generated in the data directory. - Ice now uses gethostbyname_r() if available. - Errors during close() or shutdown() on connections now cause exceptions. - Fixed a deadlock that could happen if requests were sent from the exception() AMI callback. - Eliminated the need to define _UNICODE when building Ice on Windows. - Fixed problems with generated code in slice2cpp and slice2freeze when --header-ext is used. - The Slice parser and database transformer didn't parse negative floating-point numbers correctly. This has been fixed. - If the Glacier Starter couldn't start the Glacier Router, the starter terminated. This has been fixed. - Visual C++ 7.x builds can now be used by applications built with /Zc:wchar_t. - Fixed a bug in IceSSL that caused the error message "WRN unable to load certificate authorities" to be logged when no value was specified for the "path" attribute of . - Added property Ice.Override.ConnectTimeout. See the manual for details. - Fixed a rare deadlock in the object adapter, when a locator was used. - Fixed a bug that could cause an abort if communicators were created and destroyed in rapid succession. - Fixed transformdb to take separate arguments for specifying include paths. The option -I has been replaced with the options --include-old and --include-new. - Added twoway-throttling to Glacier. See the descriptions of the properties Glacier.Router.Client.Throttle.Twoways and Glacier.Router.Server.Throttle.Twoways for details. To trace throttling, set the property Glacier.Router.Trace.Throttle. - The properties Glacier.Router.Client.BatchSleepTime and Glacier.Router.Server.BatchSleepTime have been replaced with Glacier.Router.Client.SleepTime and Glacier.Router.Server.SleepTime. Please see the description of the properties in the manual for the new semantics of the sleep times. - Added STL functors for partial proxy comparison: * ProxyIdentityLess * ProxyIdentityEqual * ProxyIdentityAndFacetLess * ProxyIdentityAndFacetEqual - A DNSException could cause a deadlock. This has been fixed. - Changed FD_SETSIZE under Windows to 1024. (The default permits only 64 concurrent connections, which is too small for some applications.) - The IcePatch server now ignores Ice exceptions in the update thread instead of terminating the update thread. - Fixed a concurrency bug that could cause the IcePack registry to crash. Changes since version 1.1.1 --------------------------- - Added a garbage collector for class graphs that form cycles. (See the C++ mapping chapter in the documentation for details.) - Manual retry in the exception() callback of asynchronous requests didn't work. This has been fixed. - Fixed a crash that could happen during shutdown. - Fixed a deadlock that could happen during connection establishment. - Fixed deadlock during shutdown that can happen if a thread pool with only one thread is used. - Moved UserExceptionFactory from Ice namespace to IceInternal namespace because UserExceptionFactory is no longer an interface for use by applications. - Bi-directional connections are now handled by the client-side thread pool instead of the server-side thread pool. - Fixed a bug in the generated code that caused "at-most-once" semantics to be ignored for collocated invocations. - Removed all Ice dependencies on Xerces. Added a simple C++ wrapper around the expat XML parser as IceXML/Parser.h. - Implemented TwowayOnlyException. That exception is raised if an attempt is made to invoke an operation that has a return value, out parameters, or an exception specification via a oneway or datagram proxy. - Removed ice_flush() on the proxy base class. Batch requests are now flushed by calling Communicator::flushBatchRequests(). This flushes all requests that are currently batched in the communicator, (for all connections). - Added back the connection closure timeout, but only for misbehaving peers. If a timeout is set, and a peer doesn't react to a close connection message, the connection is forcefully closed after the timeout. However, it is never forcefully closed if requests are still outstanding. Doing so would violate ordering guarantees for finished() and deactivate() on servant locators. - Fixed a bug in the slice2java code generator: if a parameter was named "current", illegal code was generated. - Fixed a bug in the slice2java code generator: tie classes were not generated correctly for operations with an "amd" metadata directive. - On Windows, you can now build IceUtil in such a way that mutexes are used instead of critical sections. See IceUtil/Config.h. - Fixed a bug where Ice would print a dispatch warning for Ice::RequestFailedException even if the Ice.Warn.Dispatch property was set to 1. - Added per-proxy contexts. The change is source-code compatible with the previous approach, that is, it is still possible to pass an explicit Ice::Context to an operation call as an additional, trailing parameter. However, IceProxy::Ice::Object now contains two new operations: - ice_getContext() This returns the context currently associated with a particular proxy by value. (By default, the context associated with proxies is empty.) - ice_newContext(const ::Ice::Context& context) This creates a new proxy that is associated with the passed context. Thereafter, calls via the new proxy always pass the context that was passed to ice_newContext() when that proxy was created. The net effect of this is that it is now possible to establish the context for a proxy once and, thereafter, have that same context sent automatically whenever an operation is invoked via the proxy (instead of having to pass an explicit context parameter on every call). - Added Ice::Properties::parseIceCommandLineOptions(). This operation converts to properties all options that start with one of the following prefixes: --Ice, --IceBox, --IcePack, --IcePatch, --IceSSL, --IceStorm, --Freeze, and --Glacier. - Added ice_clone() member function to non-abstract classes. ice_clone() polymorphically copies a class instance. The copy is a deep copy for all types except for class members; class members are copied shallow, that is, if a class contains a class member, the original and the clone contain a Ptr that points at the same single class instance. - Improved marshaling performance for built-in types (byte, bool, short, int, long, float, double, and string). This is noticeable in particular when transmitting sequences of these types, sequences of structures containing these types, or large dictionaries containing these types. - Added menu with several options to throughput demo. - Major Freeze update: - Removed support for the XML encoding. Freeze now always uses the binary encoding. - Removed the lowest Freeze layer. You should now use Berkeley DB directly. - Replaced the database transformation tool with one that supports the binary encoding. - For Freeze Maps and Evictors, the underlying Berkeley DB environment is now always transactional. New configuration variables: - Freeze.DbEnv..DbHome provides the path to the DB home, and defaults to . - Freeze.DbEnv..DbRecoverFatal triggers "fatal recovery" when set to a value other than 0 (defaults to 0). - Freeze.DbEnv..DbPrivate when set to a value other than 0 tells Berkeley DB to use process private memory instead of shared memory. Defaults to 1. Set it to 0 when you want to run db_archive (or another Berkeley DB utility) on a running environment. - Freeze.DbEnv..DbCheckpointPeriod: each Berkeley DB environment created by Freeze has an associated thread that checkpoints this environment every DbCheckpointPeriod seconds. Defaults to 120 seconds. - Freeze.DbEnv..PeriodicCheckpointMinSize defines the minimum size in kbytes for the periodic checkpoint (see above). This value is passed to Berkeley DB's checkpoint function. Defaults to 0 (no minimum). - Freeze.DbEnv..OldLogsAutoDelete: when set to a value other than 0, old transactional logs no longer in use are deleted after each periodic checkpoint. Defaults to 1. - Freeze.Trace.Map: trace the Map operations - Freeze.Trace.DbEnv: trace the operations on the DbEnv, which is shared between Maps and Evictors. Such a Freeze-created Berkeley DB environment is created and opened when the first Freeze Map or Evictor using it is created, and is closed when the last Freeze Map or Evictor using it is closed. - Freeze Map update: Freeze Maps are now opened/created directly using their constructor: C++: Map(const Freeze::ConnectionPtr& connection, const std::string& dbName, bool createDb = true); Java: public Map(Connection connection, String dbName, boolean createDb); Connection is a new local interface similar to the DBEnvironment local interface (which was removed). However, Connection is not thread-safe: access to a given Connection object must be serialized. A connection can be used to create a Transaction (similar to the old DBTransaction); likewise, a Transaction is not thread-safe. To access concurrently the "same" map, you should create multiple connections and associated maps. In Java, it is necessary to call close() on the Map if you want to close the underlying Berkeley DB database (and indirectly the environment) before the Map is finalized. As a consequence of using a transactional Berkeley DB environment, all Map writes are now transactional: - Non-iterator writes use DB_AUTO_COMMIT. - Writable iterators (non-const C++ iterators, Java Iterators) have an associated transaction that is committed when the iterator is closed. Concurrent reads and writes to the same underlying Map can generate deadlocks, even with a single writer. Berkeley DB detects such deadlocks and "kills" one locker (iterator or transaction) to resolve it. For non-iterator access, Freeze handles this situation and retries transparently. When using iterators, a DeadlockException is raised to the application, which is responsible for closing the affected iterator(s) or transaction, and retrying. Serialized access or concurrent reads never produce deadlocks. In C++, the copy constructor and assignment operator of Freeze Map iterators duplicate the source iterator. When the source iterator is non-const, the duplicate iterator will use the same transaction. This transaction is committed only when its last associated iterator is closed (i.e., destroyed or assigned end()). In both C++ and Java, write operations that are not within a user- controlled transaction automatically close all open iterators in the target Map. This reduces the likelihood of self-deadlocks. - Freeze Evictor update: Freeze Evictors are now created directly using a static createEvictor function: C++: Freeze::EvictorPtr Freeze::createEvictor(const Ice::CommunicatorPtr& communicator, const std::string& envName, const std::string& dbName, const std::vector indices = std::vector(), bool createDb = true); Java (in class Freeze.Util): public static Evictor createEvictor(Ice.Communicator communicator, String envName, String dbName, Index[] indices, boolean createDb); The persistence strategies have been removed. Ice objects are now saved by a background thread, controlled by the following configuration variables. - Freeze.Evictor...SaveSizeTrigger: as soon as SaveSizeTrigger objects have been created, modified or destroyed, the background thread wakes up to save them. When set to a negative value, there is no size trigger. Defaults to 10. - Freeze.Evictor...SavePeriod: the saving thread wakes up every SavePeriod ms and saves any created, modified or destroyed object. When set to 0, there is no periodic wake up. Defaults to 1 minute (60,000 ms). - Freeze.Evictor...MaxTxSize: the saving thread saves objects (facets) using transactions. MaxTxSize sets the maximum number of objects saved by a single transaction. Defaults to 10 * SaveSizeTrigger; if the provided or default value is negative, the actual value is set to 100. And instead of saving full objects, Freeze now saves on a per facet basis -- but still loads complete objects (with all their facets). Once a persistent object has been created using the Freeze Evictor, it is possible to add or remove persistent facets to this object using the following new Evictor operations: void addFacet(Ice::Identity identity, Ice::FacetPath facet, Object servant); void removeFacet(Ice::Identity identity, Ice::FacetPath facet); void removeAllFacets(Ice::Identity identity); When saving an object (facet), the background saving thread locks this object. In Java, it locks the object itself; in C++, the object is expected to implement IceUtil::AbstractMutex. Naturally the application must use the same synchronization to access the persistent state of the object. The EvictorIterator also changed slightly: the destroy operation has been removed. More importantly, identities are now retrieved in batch (internally) and the application can request that the corresponding servants be loaded. This is achieved with the new parameters of Evictor::getIterator: EvictorIterator getIterator(int batchSize, bool loadServants); You can also define Evictor indices for fast secondary key lookups: see the Freeze/phonebook demo. - Freeze now requires Berkeley DB 4.1. It no longer compiles with older Berkeley DB releases. With Visual C++, you need to use the same project directory settings when building Berkeley DB and Ice. In particular, if you build Ice with STLPort, you need to build Berkeley DB with the same STLPort. - Added Freeze hot backup demo. - Added new C++ AbstractMutex class, and helper templates (AbstractMutexI, AbstractMutexReadI, AbstractMutexWriteI) to easily implement AbstractMutex using Mutex, RecMutex, Monitor or RWRecMutex. - Fixed a bug in the implementation of StaticMutex on Windows, which could result in a crash at exit. - Added Ice.UDP.RcvSize and Ice.UDP.SndSize properties. These properties control the size of the UDP receive and send buffers as well as controlling the maximum size of a datagram invocation. If a datagram exceeds the configured size, the Ice run time throws a DatagramLimitException. (Note that, on the receiving size, detection of this condition is dependent on the local UDP implementation -- some UDP implementations silently drop datagrams that are too large to fit into the receive buffer instead of reporting the error or delivering a truncated datagram.) - Added Ice.Warn.Datagrams. This property prints a warning on the server side if a DatagramLimitException is thrown. - Added Ice.MessageSizeMax property. This property controls the maximum message size accepted by the Ice protocol in kiloBytes. The default value is 1024 (1 MB). - Fixed a number of incorrect property settings in the config files for the demos. - Added new property: Ice.Trace.Slicing. When set to a value > 0, unknown exception and class types that are sliced emit a warning. - Added destroyOnInterrupt() to the Application class. destroyOnInterrupt() is now the default behavior, because shutdownOnInterrupt() only shuts down the server side of an application, and therefore doesn't work with pure clients. - ObjectAdapter::removeServantLocator() no longer exists. The life cycle of servant locators that are registered with an object adapter ends when the adapter is destroyed. - Changed Ice::ServantLocator::deactivate to be passed the category for which a servant locator is being deactivated. - Added --noindex option to slice2docbook to suppress generation of index pages for pdf version of documentation. - Changed the behavior of Freeze::Map::insert to match the insert behavior on associative containers in the C++ standard: if the element is already there, it is not updated. Added Freeze::Map::put function, to insert or replace an element, like the old insert. This corresponds to the behavior of Berkeley DB's DB->put(). - Added a test to the Slice parser to complain if an operation on a local interface or class has an exception specification. - Added a test to the property parsing code to print a warning on stderr if a property is not recognized. This prevents silly typos, such as "Ice.config=MyFile" (instead of "Ice.Config=MyFile") from slipping through undetected. - Changed the mapping of the Slice byte type: Ice::Byte is now a typedef for unsigned char. (Previously, Ice::Byte was a typedef to char.) This change guarantees that byte values are always in the range 0..255 (instead of either -128..127 or 0..255, depending on the CPU architecture). This change also permits function overloading for Ice::Byte and char (which can be useful if you use both strings and byte sequences). - Changed the python code for printing output from test clients, so that you get each line as it comes. Changes since version 1.1.0 --------------------------- - Fixed a bug in IceXML that prevented successful extraction of strings containing 8-bit characters. - Added support for a default object factory, similar to the semantics of servant locators. Specifically, a factory registered with an empty type id is invoked when a type-specific factory cannot be found. - Added Linux/SPARC port from Ferris McCormick. - Fixed a bug where a daemonized Ice::Application wouldn't shutdown properly. - Added support for user-defined variables in IcePack service and server descriptors. - Fixed server-side bug with batch oneways or batch datagrams. - Fixed a performance problem in Glacier which would occur when using message batching. - Portability fix in src/IceSSL/OpenSSLPluginI.cpp. - config/Make.rules now checks that mkshlib is defined. - src/IceUtil/CtrlCHandler.cpp now handles (ignores) EINTR returned by sigwait. - Added DB_PRIVATE flag to DBEnv->open() on Linux, allowing Ice to use the Berkeley DB library included with RedHat 9. Changes since version 1.0.1 --------------------------- - Added support for Visual Studio .NET 2003 aka VC 7.1 - In IceUtil, all "try" and "timed" functions now return false when the resource could not be acquired (i.e., an expected failure). The "Try" from "timedTry" functions was removed, and for consistency the functions trylock, readlock and writelock were renamed to tryLock, readLock and writeLock. - Update to the library versioning scheme: On Windows, only the major and minor version are now used, for example: ice11d.dll (11d means 1.1 debug). In 1.0.x, the patch level was included in the version string, for example ice101d.dll. On Unix, the internal name (soname) is now constructed as lib.so., for example libIce.so.11 (for Ice 1.1.x). This internal name identifies a symbolic link to lib.so., for example: libIce.so.11 -> libIce.so.1.1.0. In 1.0.x, the internal name was set to the actual shared library file name, lib.so.. - Ice now implements slicing for exceptions and classes: if a process receives an exception or class that is of a more derived type than the receiver understands, the exception or class is automatically sliced. (Previously, this situation resulted in an exception.) - Class factories are no longer necessary for non-abstract classes (that is, for classes that do not have operations). For non-abstract classes, the necessary factories are automatically installed by the Ice run time. If you have existing code that implements factories for non-abstract classes, you can simply delete that code. Note that for abstract classes (that is, classes with operations), it is necessary to install a factory for each abstract class, exactly as in version 1.0.x. - User exception factories and user exception factory managers no longer exist. The factories for user exceptions are now automatically installed by the Ice run time. If you have existing code that implements user exception factories, you can simply delete it. - Fixed a bug in the Slice parser: the semantic check to see whether an identifier introduced into a scope has changed meaning was too stringent. - If a client receives a derived exception for which it does not have a user exception factory installed, the client receives the exception sliced to a known base exception. In 1.0.x, the client would have received NoUserExceptionFactoryException, an exception that no longer exists in 1.1.x. - Fixed a bug with connections being closed even though they have outstanding batch requests. - Fixed bug in the WIN32 icepatch client. The removal of orphaned files didn't take into account that WIN32 file names are not case sensitive. - New Ice.Nohup property: when its value is >0, shutdownOnInterrupt() in Application will ignore SIGHUP on Unix and CTRL_LOGOFF_EVENT on Windows. - Updated Application class: * The default behavior is now shutdownOnInterrupt(). * Removed the member function defaultInterrupt(). - Removed Communicator::signalShutdown(): Ice is not using signal handlers anymore, so Communicator::shutdown() can be used all the time. - Added IceUtil::CtrlCHandler: a class that provides portable handling of CTRL+C and CTRL+C-like signals, - Added IceUtil::StaticMutex, for simple, non-recursive, statically allocated and initialized mutexes. - Updated Mutex, RecMutex and RWRecMutex APIs: lock and unlock now return void, and trylock returns a bool that indicates whether the mutex was acquired or not. Also added new member functions to LockT, TryLockT, RLockT, TryLockT, WLockT, TryWLockT: acquire(), tryAcquire(), release(), acquired() and more. - Removed dependency on E2FSPROGS for UUID generation. On Unix, UUIDs are now generated directly with /dev/urandom. - Added support for `env' XML element in IcePack descriptors to allow setting environment variables. - Fixed a bug in the Slice parser: the semantic check to see whether an identifier introduced into a scope has changed meaning was too stringent. - Fixed integer parsing bugs in the C++ XML stream class. - Changed glacier router to send all requests from a separate thread. Previously, only requests forwarded as batch would be sent from a separate thread. - Added support for SPARC/Solaris with the Sun C++ 5.3, 5.4 and 5.5 compilers (32 and 64 bit). See INSTALL.SOLARIS for details. - Replaced the Freeze::EvictorPersistenceMode enumeration with the interface Freeze::PersistenceStrategy. - Added support for GCC 2.95.3. - Several incompatible IceStorm changes: * Moved the subscribe and unsubscribe operations to the Topic interface. * Added an object adapter to host publisher objects, and hence a new configuration property IceStorm.Publish.Endpoints. - Added dynamic thread pools, i.e., thread pools that grow and shrink based on a calculated load factor. Please see the section about thread pools in the manual for further details. - Structs and sequences which hold other structs or sequences are now legal dictionary key types, as long as such nested structs or sequences are (recursively) legal. - The connection timeout is now also used when connections are closed. This avoid hanging processes if the peer misbehaves, or if asynchronous methods invocations don't return. - In some cases, communicator destruction could result in a NullHandleException. This has been fixed. Now a CommunicatorDestroyedException is raised. - A monitor now holds the mutex lock while signaling a condition variable. Otherwise it's possible that the condition variable is destroyed before the signaling call returns, resulting in undefined behavior. The performance difference is minimal, hardly measurable. - Further performance improvements with respect to latency. - Added demo/Ice/throughput, which can be used to measure the transfer time of large sequences. - Fixed a bug in the slice2cpp code generator: if a class was forward declared and used as a class member, but the definition of the class was not supplied in the same Slice compilation unit, invalid code was generated. - Inlined most operators in IceUtil::Time for performance reasons. - Added IceUtil::ThreadControl::isAlive(). This function is useful for implementing a non-blocking join(). - Added IceUtil::ThreadControl::id() member function. - The scope of IceUtil::Thread::ThreadId has been changed to IceUtil::ThreadId. This is the type returned by both Thread::id() and ThreadControl::id(). - Joining with a thread more than once, joining with a detached thread, or detaching a detached thread now have undefined behavior. (Previously, doing any of these things raised an exception.) - Added timestamps to the default Ice logger. You can enable timestamps by setting the `Ice.Logger.Timestamp' property to a value larger than zero. By default timestamps are disabled. - Added several conversion functions to IceUtil::Time: toString, toSeconds, toMilliSeconds, toMicroSeconds. - Fixed a bug in Freeze::DBEnvironment which caused the sync method to hang. - Improved IceBox service deactivation. Instead of successively stopping a service and destroying its communicator, services are now all stopped first and then their communicators are destroyed. - Modified the Ice protocol to marshal sizes more efficiently. Sizes less than 255 now require a single byte whereas previously, sizes between 127 and 254 required five bytes. - Modified the Ice protocol to fix a problem with compression. The compressed message types have been removed, and a field has been added to the message header indicating compression status. - Added version negotiation to the Ice protocol. This permits future protocol extensions to be added without breaking interoperability with older deployed clients and servers. Unfortunately, doing this broke compatibility with the existing protocol so, when upgrading to this version, you must make sure that clients and servers use the same version of libIce.so. - Added a magic number to the Ice protocol. All Ice protocol messages have 'I', 'c', 'e', 'P' (encoded as ASCII) in the first four bytes. This is useful to allow protocol analyzers to recognize Ice messages. - Changed the encoding of encapsulations. An encapsulation now looks like: size (4 bytes) major version number (1 byte) minor version number (1 byte) data (n bytes) The size includes the count of bytes for the size and versions, so the value of size is n + 6. - Added -v and -e options to stringified UDP endpoints. -v sets the protocol version and -e sets the encoding version. (The current version for both protocol and encoding is 1.0.) - Compatibility fixes for GCC 2.96. - Added `option' and `vm-option' elements to IcePack server descriptors. These elements are used to pass command line options to server processes. - Added scandir() and alphasort() for Sun-OS and other Unix systems that do not have these BSD calls. Changes since version 1.0.0 --------------------------- - Added --header-ext and --source-ext command-line options to slice2cpp and slice2freeze. These permit setting header and source file extensions for C++ other than the default `h' and `cpp' file extensions. - Added the macro ICE_INT64 as a portable wrapper for long long constants. - Added support for Xerces 2.2.0. - Added support for OpenSSL 0.97a. - Added OPENSSL_FLAGS to several Makefiles that were missing it. - Changed signature of IceUtil::Time::milliSeconds() to IceUtil::Int64, to avoid overflows. - Renamed Lock and TryLock templates to LockT and TryLockT, to avoid compilation problems with Sun Forte 6.2. - Fixed a bug with throwing exceptions in AMD calls after invoking ice_response(). - Fixed a bug with throwing exceptions in AMI from within ice_response(). - For posix threads, PTHREAD_MUTEX_RECURSIVE_NP is now only used if __linux__ is defined. Otherwise, PTHREAD_MUTEX_RECURSIVE is used. - The EPROTO error code is now only used if it is defined (such as on Linux). - Several other changes that make Ice for C++ more portable. - Updated SSL test certificates. The old ones have expired.