Firebird 歷史版本列表 Page4

更新時間：2018-02-02
更新細節：

What's new in this version:

Improvements:
- Engine response has been improved on cancel/shutdown signals when scanning a long list of pointer pages
- The name of the encryption key is not top secret information. It can be read using the gstat utility or service, for example. However, for working with that name from a program it was desirable to access the key name using the API call Attachment::getInfo(). Hence, this facility is now provided via the information item fb_info_crypt_key
- Some clauses of the ALTER DATABASE statement require updating of the single row in RDB$DATABASE: SET DEFAULT CHARACTER SET, SET LINGER, DROP LINGER. Others, such as BEGIN|END BACKUP, ENCRYPT, DECRYPT, et al., do not need to touch that record
- In previous versions, to prevent concurrent instances of ALTER DATABASE running in parallel transactions, the engine would run an update on the RDB$DATABASE record regardless of the nature of clauses specified by the user. Hence, any other transaction that read the RDB$DATABASE record in READ COMMITTED NO RECORD VERSION mode would be blocked briefly, even by a “dummy update” that in fact did not update the record
- In some cases, such as with an ALTER DATABASE END BACKUP the blockage could last 10 minutes or more. A user would seem to be unable to connect to the database with isql, for example, while ALTER DATABASE END BACKUP was running. In fact, isql would connect successfully, but it would read RDB$DATABASE immediately after attaching, using a READ COMMITTED NO RECORD VERSION WAIT transaction and then just wait until the work of ALTER DATABASE END BACKUP was committed
- From this sub-release forward, the update of the RDB$DATABASE record is avoided when possible and an implicit lock is placed to prevent concurrent runs of the ALTER DATABASE statement
- All queries that are semantically the same should have the same plan. However, until now, the optimizer understood only an explicit reference inside an ORDER BY clause and would ignore sorts derived from equivalent expressions. Now, it will consider equivalence classes for index navigation. Refer to the Tracker ticket for an example
- Common Table Expressions are now allowed to be left unused
- Flushing a large number of dirty pages has been made faster
- Measures have been taken to avoid serialization of isc_attach_database calls issued by EXECUTE STATEMENT
- Output from gstat now includes the date and time of analysis
- The merge stage of a physical backup stage could run too long, especially with huge page cache. Changes have been made to reduce it
- Message “Error during sweep: connection shutdown” now provides information about the database that was being swept
- Improvement in performance of ALTER DOMAIN when the domain has many dependencies
- Compression details and encryption status of the connection (fb_info_conn_flags) have been added to the getInfo() API call
- Restoring a pre ODS 11.1 database now correctly populates RDB$RELATION_TYPE field in the metadata
- Speed of backup with nBackup when directed to NAS over SMB protocol has been improved
- The optimizer can now estimate the actual record compression ratio

更新時間：2018-01-08
更新細節：

What's new in this version:

- Populate RDB$RELATION_TYPE while restoring pre ODS 11.1 databases
- GSTAT output now includes datetime of analysis
- Improve concurrency of new connections being established
- Allow unused Common Table Expressions
- 19 bugs fixed

更新時間：2017-03-26
更新細節：

What's new in this version:

New Feature:
- Support was added for fast/low-latency “TCP Loopback Fast Path” functionality introduced in Windows 8 and Server 2012. This feature is said to improve the performance of the TCP stack for local loopback connections, by short-circuiting the TCP stack for local calls. The details of the feature can be found in this Microsoft Technet blog

Improvements:
- It is now possible to filter out info and warnings from the trace log
- Enhanced control capability when sharing the database crypt key between Superserver attachments
- The physical numbers of frequently used data pages are now cached to reduce the number of fetches of pointer pages
- A read-only transaction will no longer force write the HeaderTIP page content to disk immediately after a change. This improvement gives a significant performance gain where there are numerous light read-only transactions. At this stage, it affects only servers in SS mode. For CS and SC it is more complex to implement and should appear in Firebird 4.0
- The database name was made available to an encryption plug-in
- libfbclient.so was compiled for Android (x86/x86-64/arm64)
- Nesting of keys in a plug-in configuration was enabled
- The Linux code is now built with --enable-binreloc and an option was included in the installer script to install in locations other than /opt/firebird
- For Trace, a filter has been provided to INCLUDE / EXCLUDE errors by their mnemonical names
- Android port (arm32)
- A port was done and tested for Linux on the ancient Motorola 680000 CPU platform to satisfy some requirement from Debian
- Support has been added to enable SELECT expressions to be valid operands for the BETWEEN predicate

更新時間：2017-03-26
更新細節：

What's new in this version:

New Feature:
- Support was added for fast/low-latency “TCP Loopback Fast Path” functionality introduced in Windows 8 and Server 2012. This feature is said to improve the performance of the TCP stack for local loopback connections, by short-circuiting the TCP stack for local calls. The details of the feature can be found in this Microsoft Technet blog

Improvements:
- It is now possible to filter out info and warnings from the trace log
- Enhanced control capability when sharing the database crypt key between Superserver attachments
- The physical numbers of frequently used data pages are now cached to reduce the number of fetches of pointer pages
- A read-only transaction will no longer force write the HeaderTIP page content to disk immediately after a change. This improvement gives a significant performance gain where there are numerous light read-only transactions. At this stage, it affects only servers in SS mode. For CS and SC it is more complex to implement and should appear in Firebird 4.0
- The database name was made available to an encryption plug-in
- libfbclient.so was compiled for Android (x86/x86-64/arm64)
- Nesting of keys in a plug-in configuration was enabled
- The Linux code is now built with --enable-binreloc and an option was included in the installer script to install in locations other than /opt/firebird
- For Trace, a filter has been provided to INCLUDE / EXCLUDE errors by their mnemonical names
- Android port (arm32)
- A port was done and tested for Linux on the ancient Motorola 680000 CPU platform to satisfy some requirement from Debian
- Support has been added to enable SELECT expressions to be valid operands for the BETWEEN predicate

更新時間：2016-09-27
更新細節：

更新時間：2016-09-27
更新細節：

更新時間：2016-07-04
更新細節：

更新時間：2016-07-04
更新細節：

更新時間：2016-04-19
更新細節：

What's new in this version:

- The primary goals for Firebird 3.0 were to unify the server architecture and to improve support for SMP and multiple-core hardware platforms. Parallel objectives were to improve threading of engine processes and the options for sharing page cache across thread and connection boundaries.
- Alongside these aims came new strategies to improve performance, query optimization, monitoring and scalability and to address the demand for more security options. A number of popular features were introduced into the SQL language, including the long-awaited support for the Boolean data type and the associated logical predications.

BUGS FIXED:

Core Engine:
- An assertion would occur in the cloop dispatcher when trying to save exception information in the status interface.
- Failure to find a record on re-fetch would produce inconsistent effects on subsequent statement execution.
- Unique key violation on a BOOLEAN column was returning an erroneous message.
- HAVING COUNT(*) NOT IN ( ) would erroneously exclude a record from the outer result set. ( represents a result set without nulls)
- It was possible to create a unique index on non-unique data.
- Regression: the engine could throw transliteration errors when running from a non-ASCII system path.
- The server could go into an infinite loop when data was being sent from a client to the server in small packets.
- Regression: LEFT JOIN was incorrectly pushing COALESCE into the inner stream, causing wrong results.
- CREATE TRIGGER was failing with the error "Ambiguous field name between table B and table A".
- Join order was sub-optimal if one table had a selective predicate and MIN was calculated for the other one.
- Deadlock could occur when a database was being encrypted or decrypted under high parallel load.
- ALTER SEQUENCE RESTART WITH was failing to change the initial value.
- Minor regression: wrong error message was being returned on an attempt to compile a view with both a subquery and the WITH CHECK OPTION clause in its definition—an illegal combination.
- The optimizer could avoid using an expression index if the query was running in a different connection character set to the one that was in effect when the index was created.
- Internal Firebird consistency check (cannot find tip page (165), file: tra.cpp line: 2375) would occur if too many transactions were created for one connection to a read-only database. This was a very old bug, dating back to InterBase 6.

Server Crashes:
- Regression: The server would crash when aggregate functions were used together with the NOT IN predicate.

Utilities:
- gbak: gbak restore could fail when there was some SQL function accessing a table while the switch -O(NE_AT_A_TIME) was in use.
- gfix: An attempt to set the number of page buffers to a non-supported value would return the wrong error message.
- isql: An invalid COMPUTED BY definition was being generated by the 'extract metadata' operation.
- Monitoring: Database-level statistics (MON$STAT_GROUP = 0) could contain multiple rows under concurrent load.
- User Management: The codes of operation for user management plug-ins were missing from the public API.
- User Management: Users with the same name can co-exist in the security database if they are created by different plug-ins. However, tags or attributes for a user created using one plug-in were being displayed in the SEC$ virtual tables for the same-name user created by the other plug-in.
- User Management: [CREATE OR] ALTER USER statement would fail if the PASSWORD parameter was present and was not specified immediately after the USER parameter.

Builds:
- The build would fail if HAVE_MMAP was not defined.
- Root rights were unnecessarily required to build a Linux package.
- The samples of dbcrypt-related plug-ins contained references to internal files.

更新時間：2016-04-19
更新細節：

What's new in this version:

- The primary goals for Firebird 3.0 were to unify the server architecture and to improve support for SMP and multiple-core hardware platforms. Parallel objectives were to improve threading of engine processes and the options for sharing page cache across thread and connection boundaries.
- Alongside these aims came new strategies to improve performance, query optimization, monitoring and scalability and to address the demand for more security options. A number of popular features were introduced into the SQL language, including the long-awaited support for the Boolean data type and the associated logical predications.

BUGS FIXED:

Core Engine:
- An assertion would occur in the cloop dispatcher when trying to save exception information in the status interface.
- Failure to find a record on re-fetch would produce inconsistent effects on subsequent statement execution.
- Unique key violation on a BOOLEAN column was returning an erroneous message.
- HAVING COUNT(*) NOT IN ( ) would erroneously exclude a record from the outer result set. ( represents a result set without nulls)
- It was possible to create a unique index on non-unique data.
- Regression: the engine could throw transliteration errors when running from a non-ASCII system path.
- The server could go into an infinite loop when data was being sent from a client to the server in small packets.
- Regression: LEFT JOIN was incorrectly pushing COALESCE into the inner stream, causing wrong results.
- CREATE TRIGGER was failing with the error "Ambiguous field name between table B and table A".
- Join order was sub-optimal if one table had a selective predicate and MIN was calculated for the other one.
- Deadlock could occur when a database was being encrypted or decrypted under high parallel load.
- ALTER SEQUENCE RESTART WITH was failing to change the initial value.
- Minor regression: wrong error message was being returned on an attempt to compile a view with both a subquery and the WITH CHECK OPTION clause in its definition—an illegal combination.
- The optimizer could avoid using an expression index if the query was running in a different connection character set to the one that was in effect when the index was created.
- Internal Firebird consistency check (cannot find tip page (165), file: tra.cpp line: 2375) would occur if too many transactions were created for one connection to a read-only database. This was a very old bug, dating back to InterBase 6.

Server Crashes:
- Regression: The server would crash when aggregate functions were used together with the NOT IN predicate.

Utilities:
- gbak: gbak restore could fail when there was some SQL function accessing a table while the switch -O(NE_AT_A_TIME) was in use.
- gfix: An attempt to set the number of page buffers to a non-supported value would return the wrong error message.
- isql: An invalid COMPUTED BY definition was being generated by the 'extract metadata' operation.
- Monitoring: Database-level statistics (MON$STAT_GROUP = 0) could contain multiple rows under concurrent load.
- User Management: The codes of operation for user management plug-ins were missing from the public API.
- User Management: Users with the same name can co-exist in the security database if they are created by different plug-ins. However, tags or attributes for a user created using one plug-in were being displayed in the SEC$ virtual tables for the same-name user created by the other plug-in.
- User Management: [CREATE OR] ALTER USER statement would fail if the PASSWORD parameter was present and was not specified immediately after the USER parameter.

Builds:
- The build would fail if HAVE_MMAP was not defined.
- Root rights were unnecessarily required to build a Linux package.
- The samples of dbcrypt-related plug-ins contained references to internal files.