Release 9.2

Release date: 2012-09-10

E.96.1. Overview

This release has been largely focused on performance improvements, though new SQL features are not lacking. Work also continues in the area of replication support. Major enhancements include:

  • Allow queries to retrieve data only from indexes, avoiding heap access ( index-only scans )

  • Allow the planner to generate custom plans for specific parameter values even when using prepared statements

  • Improve the planner's ability to use nested loops with inner index scans

  • Allow streaming replication slaves to forward data to other slaves ( cascading replication )

  • Allow pg_basebackup to make base backups from standby servers

  • Add a pg_receivexlog tool to archive WAL file changes as they are written

  • Add the SP-GiST (Space-Partitioned GiST) index access method

  • Add support for range data types

  • Add a JSON data type

  • Add a security_barrier option for views

  • Allow libpq connection strings to have the format of a URI

  • Add a single-row processing mode to libpq for better handling of large result sets

The above items are explained in more detail in the sections below.

E.96.2. Migration to Version 9.2

A dump/restore using pg_dump , or use of pg_upgrade , is required for those wishing to migrate data from any previous release.

Version 9.2 contains a number of changes that may affect compatibility with previous releases. Observe the following incompatibilities:

E.96.2.1. System Catalogs

  • Remove the spclocation field from pg_tablespace (Magnus Hagander)

    This field was duplicative of the symbolic links that actually define tablespace locations, and thus risked errors of omission when moving a tablespace. This change allows tablespace directories to be moved while the server is down, by manually adjusting the symbolic links. To replace this field, we have added pg_tablespace_location() to allow querying of the symbolic links.

  • Move tsvector most-common-element statistics to new pg_stats columns (Alexander Korotkov)

    Consult most_common_elems and most_common_elem_freqs for the data formerly available in most_common_vals and most_common_freqs for a tsvector column.

E.96.2.2. Functions

  • Remove hstore 's => operator (Robert Haas)

    Users should now use hstore(text, text) . Since PostgreSQL 9.0, a warning message has been emitted when an operator named => is created because the SQL standard reserves that token for another use.

  • Ensure that xpath() escapes special characters in string values (Florian Pflug)

    Without this it is possible for the result not to be valid XML .

  • Make pg_relation_size() and friends return NULL if the object does not exist (Phil Sorber)

    This prevents queries that call these functions from returning errors immediately after a concurrent DROP .

  • Make EXTRACT(EPOCH FROM timestamp without time zone ) measure the epoch from local midnight, not UTC midnight (Tom Lane)

    This change reverts an ill-considered change made in release 7.3. Measuring from UTC midnight was inconsistent because it made the result dependent on the timezone setting, which computations for timestamp without time zone should not be. The previous behavior remains available by casting the input value to timestamp with time zone .

  • Properly parse time strings with trailing yesterday , today , and tomorrow (Dean Rasheed)

    Previously, SELECT '04:00:00 yesterday'::timestamp returned yesterday's date at midnight.

  • Fix to_date() and to_timestamp() to wrap incomplete dates toward 2020 (Bruce Momjian)

    Previously, supplied years and year masks of less than four digits wrapped inconsistently.

E.96.2.3. Object Modification

  • Prevent ALTER DOMAIN from working on non-domain types (Peter Eisentraut)

    Owner and schema changes were previously possible on non-domain types.

  • No longer forcibly lowercase procedural language names in CREATE FUNCTION (Robert Haas)

    While unquoted language identifiers are still lowercased, strings and quoted identifiers are no longer forcibly down-cased. Thus for example CREATE FUNCTION ... LANGUAGE 'C' will no longer work; it must be spelled 'c' , or better omit the quotes.

  • Change system-generated names of foreign key enforcement triggers (Tom Lane)

    This change ensures that the triggers fire in the correct order in some corner cases involving self-referential foreign key constraints.

E.96.2.4. Command-Line Tools

  • Provide consistent backquote, variable expansion, and quoted substring behavior in psql meta-command arguments (Tom Lane)

    Previously, such references were treated oddly when not separated by whitespace from adjacent text. For example 'FOO'BAR was output as FOO BAR (unexpected insertion of a space) and FOO'BAR'BAZ was output unchanged (not removing the quotes as most would expect).

  • No longer treat clusterdb table names as double-quoted; no longer treat reindexdb table and index names as double-quoted (Bruce Momjian)

    Users must now include double-quotes in the command arguments if quoting is wanted.

  • createuser no longer prompts for option settings by default (Peter Eisentraut)

    Use --interactive to obtain the old behavior.

  • Disable prompting for the user name in dropuser unless --interactive is specified (Peter Eisentraut)

E.96.2.5. Server Settings

  • Add server parameters for specifying the locations of server-side SSL files (Peter Eisentraut)

    This allows changing the names and locations of the files that were previously hard-coded as server.crt , server.key , root.crt , and root.crl in the data directory. The server will no longer examine root.crt or root.crl by default ; to load these files, the associated parameters must be set to non-default values.

  • Remove the silent_mode parameter (Heikki Linnakangas)

    Similar behavior can be obtained with pg_ctl start -l postmaster.log .

  • Remove the wal_sender_delay parameter, as it is no longer needed (Tom Lane)

  • Remove the custom_variable_classes parameter (Tom Lane)

    The checking provided by this setting was dubious. Now any setting can be prefixed by any class name.

E.96.2.6. Monitoring

  • Rename pg_stat_activity .procpid to pid , to match other system tables (Magnus Hagander)

  • Create a separate pg_stat_activity column to report process state (Scott Mead, Magnus Hagander)

    The previous query and query_start values now remain available for an idle session, allowing enhanced analysis.

  • Rename pg_stat_activity . current_query to query because it is not cleared when the query completes (Magnus Hagander)

  • Change all SQL -level statistics timing values to be float8 columns measured in milliseconds (Tom Lane)

    This change eliminates the designed-in assumption that the values are accurate to microseconds and no more (since the float8 values can be fractional). The columns affected are pg_stat_user_functions . total_time , pg_stat_user_functions . self_time , pg_stat_xact_user_functions . total_time , and pg_stat_xact_user_functions . self_time . The statistics functions underlying these columns now also return float8 milliseconds, rather than bigint microseconds. contrib/pg_stat_statements ' total_time column is now also measured in milliseconds.

E.96.3. Changes

Below you will find a detailed account of the changes between PostgreSQL 9.2 and the previous major release.

E.96.3.1. Server

E.96.3.1.1. Performance

  • Allow queries to retrieve data only from indexes, avoiding heap access (Robert Haas, Ibrar Ahmed, Heikki Linnakangas, Tom Lane)

    This feature is often called index-only scans . Heap access can be skipped for heap pages containing only tuples that are visible to all sessions, as reported by the visibility map; so the benefit applies mainly to mostly-static data. The visibility map was made crash-safe as a necessary part of implementing this feature.

  • Add the SP-GiST (Space-Partitioned GiST) index access method (Teodor Sigaev, Oleg Bartunov, Tom Lane)

    SP-GiST is comparable to GiST in flexibility, but supports unbalanced partitioned search structures rather than balanced trees. For suitable problems, SP-GiST can be faster than GiST in both index build time and search time.

  • Allow group commit to work effectively under heavy load (Peter Geoghegan, Simon Riggs, Heikki Linnakangas)

    Previously, batching of commits became ineffective as the write workload increased, because of internal lock contention.

  • Allow uncontended locks to be managed using a new fast-path lock mechanism (Robert Haas)

  • Reduce overhead of creating virtual transaction ID locks (Robert Haas)

  • Reduce the overhead of serializable isolation level locks (Dan Ports)

  • Improve PowerPC and Itanium spinlock performance (Manabu Ori, Robert Haas, Tom Lane)

  • Reduce overhead for shared invalidation cache messages (Robert Haas)

  • Move the frequently accessed members of the PGPROC shared memory array to a separate array (Pavan Deolasee, Heikki Linnakangas, Robert Haas)

  • Improve COPY performance by adding tuples to the heap in batches (Heikki Linnakangas)

  • Improve GiST index performance for geometric data types by producing better trees with less memory allocation overhead (Alexander Korotkov)

  • Improve GiST index build times (Alexander Korotkov, Heikki Linnakangas)

  • Allow hint bits to be set sooner for temporary and unlogged tables (Robert Haas)

  • Allow sorting to be performed by inlined, non- SQL -callable comparison functions (Peter Geoghegan, Robert Haas, Tom Lane)

  • Make the number of CLOG buffers scale based on shared_buffers (Robert Haas, Simon Riggs, Tom Lane)

  • Improve performance of buffer pool scans that occur when tables or databases are dropped (Jeff Janes, Simon Riggs)

  • Improve performance of checkpointer's fsync-request queue when many tables are being dropped or truncated (Tom Lane)

  • Pass the safe number of file descriptors to child processes on Windows (Heikki Linnakangas)

    This allows Windows sessions to use more open file descriptors than before.

E.96.3.1.2. Process Management

  • Create a dedicated background process to perform checkpoints (Simon Riggs)

    Formerly the background writer did both dirty-page writing and checkpointing. Separating this into two processes allows each goal to be accomplished more predictably.

  • Improve asynchronous commit behavior by waking the walwriter sooner (Simon Riggs)

    Previously, only wal_writer_delay triggered WAL flushing to disk; now filling a WAL buffer also triggers WAL writes.

  • Allow the bgwriter, walwriter, checkpointer, statistics collector, log collector, and archiver background processes to sleep more efficiently during periods of inactivity (Peter Geoghegan, Tom Lane)

    This series of changes reduces the frequency of process wake-ups when there is nothing to do, dramatically reducing power consumption on idle servers.

E.96.3.1.3. Optimizer

  • Allow the planner to generate custom plans for specific parameter values even when using prepared statements (Tom Lane)

    In the past, a prepared statement always had a single "generic" plan that was used for all parameter values, which was frequently much inferior to the plans used for non-prepared statements containing explicit constant values. Now, the planner attempts to generate custom plans for specific parameter values. A generic plan will only be used after custom plans have repeatedly proven to provide no benefit. This change should eliminate the performance penalties formerly seen from use of prepared statements (including non-dynamic statements in PL/pgSQL).

  • Improve the planner's ability to use nested loops with inner index scans (Tom Lane)

    The new "parameterized path" mechanism allows inner index scans to use values from relations that are more than one join level up from the scan. This can greatly improve performance in situations where semantic restrictions (such as outer joins) limit the allowed join orderings.

  • Improve the planning API for foreign data wrappers (Etsuro Fujita, Shigeru Hanada, Tom Lane)

    Wrappers can now provide multiple access "paths" for their tables, allowing more flexibility in join planning.

  • Recognize self-contradictory restriction clauses for non-table relations (Tom Lane)

    This check is only performed when constraint_exclusion is on .

  • Allow indexed_col op ANY(ARRAY[...]) conditions to be used in plain index scans and index-only scans (Tom Lane)

    Formerly such conditions could only be used in bitmap index scans.

  • Support MIN / MAX index optimizations on boolean columns (Marti Raudsepp)

  • Account for set-returning functions in SELECT target lists when setting row count estimates (Tom Lane)

  • Fix planner to handle indexes with duplicated columns more reliably (Tom Lane)

  • Collect and use element-frequency statistics for arrays (Alexander Korotkov, Tom Lane)

    This change improves selectivity estimation for the array <@ , && , and @> operators (array containment and overlaps).

  • Allow statistics to be collected for foreign tables (Etsuro Fujita)

  • Improve cost estimates for use of partial indexes (Tom Lane)

  • Improve the planner's ability to use statistics for columns referenced in subqueries (Tom Lane)

  • Improve statistical estimates for subqueries using DISTINCT (Tom Lane)

E.96.3.1.4. Authentication

  • Do not treat role names and samerole specified in pg_hba.conf as automatically including superusers (Andrew Dunstan)

    This makes it easier to use reject lines with group roles.

  • Adjust pg_hba.conf processing to handle token parsing more consistently (Brendan Jurd, Álvaro Herrera)

  • Disallow empty pg_hba.conf files (Tom Lane)

    This was done to more quickly detect misconfiguration.

  • Make superuser privilege imply replication privilege (Noah Misch)

    This avoids the need to explicitly assign such privileges.

E.96.3.1.5. Monitoring

  • Attempt to log the current query string during a backend crash (Marti Raudsepp)

  • Make logging of autovacuum I/O activity more verbose (Greg Smith, Noah Misch)

    This logging is triggered by log_autovacuum_min_duration .

  • Make WAL replay report failures sooner (Fujii Masao)

    There were some cases where failures were only reported once the server went into master mode.

  • Add pg_xlog_location_diff() to simplify WAL location comparisons (Euler Taveira de Oliveira)

    This is useful for computing replication lag.

  • Support configurable event log application names on Windows (MauMau, Magnus Hagander)

    This allows different instances to use the event log with different identifiers, by setting the event_source server parameter, which is similar to how syslog_ident works.

  • Change "unexpected EOF" messages to DEBUG1 level, except when there is an open transaction (Magnus Hagander)

    This change reduces log chatter caused by applications that close database connections ungracefully.

E.96.3.1.6. Statistical Views

  • Track temporary file sizes and file counts in the pg_stat_database system view (Tomas Vondra)

  • Add a deadlock counter to the pg_stat_database system view (Magnus Hagander)

  • Add a server parameter track_io_timing to track I/O timings (Ants Aasma, Robert Haas)

  • Report checkpoint timing information in pg_stat_bgwriter (Greg Smith, Peter Geoghegan)

E.96.3.1.7. Server Settings

  • Silently ignore nonexistent schemas specified in search_path (Tom Lane)

    This makes it more convenient to use generic path settings, which might include some schemas that don't exist in all databases.

  • Allow superusers to set deadlock_timeout per-session, not just per-cluster (Noah Misch)

    This allows deadlock_timeout to be reduced for transactions that are likely to be involved in a deadlock, thus detecting the failure more quickly. Alternatively, increasing the value can be used to reduce the chances of a session being chosen for cancellation due to a deadlock.

  • Add a server parameter temp_file_limit to constrain temporary file space usage per session (Mark Kirkwood)

  • Allow a superuser to SET an extension's superuser-only custom variable before loading the associated extension (Tom Lane)

    The system now remembers whether a SET was performed by a superuser, so that proper privilege checking can be done when the extension is loaded.

  • Add postmaster -C option to query configuration parameters (Bruce Momjian)

    This allows pg_ctl to better handle cases where PGDATA or -D points to a configuration-only directory.

  • Replace an empty locale name with the implied value in CREATE DATABASE (Tom Lane)

    This prevents cases where pg_database . datcollate or datctype could be interpreted differently after a server restart.

E.96.3.1.7.1. postgresql.conf

  • Allow multiple errors in postgresql.conf to be reported, rather than just the first one (Alexey Klyukin, Tom Lane)

  • Allow a reload of postgresql.conf to be processed by all sessions, even if there are some settings that are invalid for particular sessions (Alexey Klyukin)

    Previously, such not-valid-within-session values would cause all setting changes to be ignored by that session.

  • Add an include_if_exists facility for configuration files (Greg Smith)

    This works the same as include , except that an error is not thrown if the file is missing.

  • Identify the server time zone during initdb , and set postgresql.conf entries timezone and log_timezone accordingly (Tom Lane)

    This avoids expensive time zone probes during server start.

  • Fix pg_settings to report postgresql.conf line numbers on Windows (Tom Lane)

E.96.3.2. Replication and Recovery

  • Allow streaming replication slaves to forward data to other slaves ( cascading replication ) (Fujii Masao)

    Previously, only the master server could supply streaming replication log files to standby servers.

  • Add new synchronous_commit mode remote_write (Fujii Masao, Simon Riggs)

    This mode waits for the standby server to write transaction data to its own operating system, but does not wait for the data to be flushed to the standby's disk.

  • Add a pg_receivexlog tool to archive WAL file changes as they are written, rather than waiting for completed WAL files (Magnus Hagander)

  • Allow pg_basebackup to make base backups from standby servers (Jun Ishizuka, Fujii Masao)

    This feature lets the work of making new base backups be off-loaded from the primary server.

  • Allow streaming of WAL files while pg_basebackup is performing a backup (Magnus Hagander)

    This allows passing of WAL files to the standby before they are discarded on the primary.

E.96.3.3. Queries

  • Cancel the running query if the client gets disconnected (Florian Pflug)

    If the backend detects loss of client connection during a query, it will now cancel the query rather than attempting to finish it.

  • Retain column names at run time for row expressions (Andrew Dunstan, Tom Lane)

    This change allows better results when a row value is converted to hstore or json type: the fields of the resulting value will now have the expected names.

  • Improve column labels used for sub- SELECT results (Marti Raudsepp)

    Previously, the generic label ?column? was used.

  • Improve heuristics for determining the types of unknown values (Tom Lane)

    The longstanding rule that an unknown constant might have the same type as the value on the other side of the operator using it is now applied when considering polymorphic operators, not only for simple operator matches.

  • Warn about creating casts to or from domain types (Robert Haas)

    Such casts have no effect.

  • When a row fails a CHECK or NOT NULL constraint, show the row's contents as error detail (Jan Kundrát)

    This should make it easier to identify which row is problematic when an insert or update is processing many rows.

E.96.3.4. Object Manipulation

  • Provide more reliable operation during concurrent DDL (Robert Haas, Noah Misch)

    This change adds locking that should eliminate "cache lookup failed" errors in many scenarios. Also, it is no longer possible to add relations to a schema that is being concurrently dropped, a scenario that formerly led to inconsistent system catalog contents.

  • Add CONCURRENTLY option to DROP INDEX (Simon Riggs)

    This allows index removal without blocking other sessions.

  • Allow foreign data wrappers to have per-column options (Shigeru Hanada)

  • Improve pretty-printing of view definitions (Andrew Dunstan)

E.96.3.4.1. Constraints

  • Allow CHECK constraints to be declared NOT VALID (Álvaro Herrera)

    Adding a NOT VALID constraint does not cause the table to be scanned to verify that existing rows meet the constraint. Subsequently, newly added or updated rows are checked. Such constraints are ignored by the planner when considering constraint_exclusion , since it is not certain that all rows meet the constraint.

    The new ALTER TABLE VALIDATE command allows NOT VALID constraints to be checked for existing rows, after which they are converted into ordinary constraints.

  • Allow CHECK constraints to be declared NO INHERIT (Nikhil Sontakke, Alex Hunsaker, Álvaro Herrera)

    This makes them enforceable only on the parent table, not on child tables.

  • Add the ability to rename constraints (Peter Eisentraut)

E.96.3.4.2. ALTER

  • Reduce need to rebuild tables and indexes for certain ALTER TABLE ... ALTER COLUMN TYPE operations (Noah Misch)

    Increasing the length limit for a varchar or varbit column, or removing the limit altogether, no longer requires a table rewrite. Similarly, increasing the allowable precision of a numeric column, or changing a column from constrained numeric to unconstrained numeric , no longer requires a table rewrite. Table rewrites are also avoided in similar cases involving the interval , timestamp , and timestamptz types.

  • Avoid having ALTER TABLE revalidate foreign key constraints in some cases where it is not necessary (Noah Misch)

  • Add IF EXISTS options to some ALTER commands (Pavel Stehule)

    For example, ALTER FOREIGN TABLE IF EXISTS foo RENAME TO bar .

  • Add ALTER FOREIGN DATA WRAPPER ... RENAME and ALTER SERVER ... RENAME (Peter Eisentraut)

  • Add ALTER DOMAIN ... RENAME (Peter Eisentraut)

    You could already rename domains using ALTER TYPE .

  • Throw an error for ALTER DOMAIN ... DROP CONSTRAINT on a nonexistent constraint (Peter Eisentraut)

    An IF EXISTS option has been added to provide the previous behavior.

E.96.3.4.3. CREATE TABLE

  • Allow CREATE TABLE (LIKE ...) from foreign tables, views, and composite types (Peter Eisentraut)

    For example, this allows a table to be created whose schema matches a view.

  • Fix CREATE TABLE (LIKE ...) to avoid index name conflicts when copying index comments (Tom Lane)

  • Fix CREATE TABLE ... AS EXECUTE to handle WITH NO DATA and column name specifications (Tom Lane)

E.96.3.4.4. Object Permissions

  • Add a security_barrier option for views (KaiGai Kohei, Robert Haas)

    This option prevents optimizations that might allow view-protected data to be exposed to users, for example pushing a clause involving an insecure function into the WHERE clause of the view. Such views can be expected to perform more poorly than ordinary views.

  • Add a new LEAKPROOF function attribute to mark functions that can safely be pushed down into security_barrier views (KaiGai Kohei)

  • Add support for privileges on data types (Peter Eisentraut)

    This adds support for the SQL -conforming USAGE privilege on types and domains. The intent is to be able to restrict which users can create dependencies on types, since such dependencies limit the owner's ability to alter the type.

  • Check for INSERT privileges in SELECT INTO / CREATE TABLE AS (KaiGai Kohei)

    Because the object is being created by SELECT INTO or CREATE TABLE AS , the creator would ordinarily have insert permissions; but there are corner cases where this is not true, such as when ALTER DEFAULT PRIVILEGES has removed such permissions.

E.96.3.5. Utility Operations

  • Allow VACUUM to more easily skip pages that cannot be locked (Simon Riggs, Robert Haas)

    This change should greatly reduce the incidence of VACUUM getting "stuck" waiting for other sessions.

  • Make EXPLAIN (BUFFERS) count blocks dirtied and written (Robert Haas)

  • Make EXPLAIN ANALYZE report the number of rows rejected by filter steps (Marko Tiikkaja)

  • Allow EXPLAIN ANALYZE to avoid timing overhead when time values are not wanted (Tomas Vondra)

    This is accomplished by setting the new TIMING option to FALSE .

E.96.3.6. Data Types

  • Add support for range data types (Jeff Davis, Tom Lane, Alexander Korotkov)

    A range data type stores a lower and upper bound belonging to its base data type. It supports operations like contains, overlaps, and intersection.

  • Add a JSON data type (Robert Haas)

    This type stores JSON (JavaScript Object Notation) data with proper validation.

  • Add array_to_json() and row_to_json() (Andrew Dunstan)

  • Add a SMALLSERIAL data type (Mike Pultz)

    This is like SERIAL , except it stores the sequence in a two-byte integer column ( int2 ).

  • Allow domains to be declared NOT VALID (Álvaro Herrera)

    This option can be set at domain creation time, or via ALTER DOMAIN ... ADD CONSTRAINT ... NOT VALID . ALTER DOMAIN ... VALIDATE CONSTRAINT fully validates the constraint.

  • Support more locale-specific formatting options for the money data type (Tom Lane)

    Specifically, honor all the POSIX options for ordering of the value, sign, and currency symbol in monetary output. Also, make sure that the thousands separator is only inserted to the left of the decimal point, as required by POSIX.

  • Add bitwise "and" , "or" , and "not" operators for the macaddr data type (Brendan Jurd)

  • Allow xpath() to return a single-element XML array when supplied a scalar value (Florian Pflug)

    Previously, it returned an empty array. This change will also cause xpath_exists() to return true, not false, for such expressions.

  • Improve XML error handling to be more robust (Florian Pflug)

E.96.3.7. Functions

  • Allow non-superusers to use pg_cancel_backend() and pg_terminate_backend() on other sessions belonging to the same user (Magnus Hagander, Josh Kupershmidt, Dan Farina)

    Previously only superusers were allowed to use these functions.

  • Allow importing and exporting of transaction snapshots (Joachim Wieland, Tom Lane)

    This allows multiple transactions to share identical views of the database state. Snapshots are exported via pg_export_snapshot() and imported via SET TRANSACTION SNAPSHOT . Only snapshots from currently-running transactions can be imported.

  • Support COLLATION FOR on expressions (Peter Eisentraut)

    This returns a string representing the collation of the expression.

  • Add pg_opfamily_is_visible() (Josh Kupershmidt)

  • Add a numeric variant of pg_size_pretty() for use with pg_xlog_location_diff() (Fujii Masao)

  • Add a pg_trigger_depth() function (Kevin Grittner)

    This reports the current trigger call depth.

  • Allow string_agg() to process bytea values (Pavel Stehule)

  • Fix regular expressions in which a back-reference occurs within a larger quantified subexpression (Tom Lane)

    For example, ^(\w+)( \1)+$ . Previous releases did not check that the back-reference actually matched the first occurrence.

E.96.3.8. Information Schema

  • Add information schema views role_udt_grants , udt_privileges , and user_defined_types (Peter Eisentraut)

  • Add composite-type attributes to the information schema element_types view (Peter Eisentraut)

  • Implement interval_type columns in the information schema (Peter Eisentraut)

    Formerly these columns read as nulls.

  • Implement collation-related columns in the information schema attributes , columns , domains , and element_types views (Peter Eisentraut)

  • Implement the with_hierarchy column in the information schema table_privileges view (Peter Eisentraut)

  • Add display of sequence USAGE privileges to information schema (Peter Eisentraut)

  • Make the information schema show default privileges (Peter Eisentraut)

    Previously, non-empty default permissions were not represented in the views.

E.96.3.9. Server-Side Languages

E.96.3.9.1. PL/pgSQL Server-Side Language

  • Allow the PL/pgSQL OPEN cursor command to supply parameters by name (Yeb Havinga)

  • Add a GET STACKED DIAGNOSTICS PL/pgSQL command to retrieve exception info (Pavel Stehule)

  • Speed up PL/pgSQL array assignment by caching type information (Pavel Stehule)

  • Improve performance and memory consumption for long chains of ELSIF clauses (Tom Lane)

  • Output the function signature, not just the name, in PL/pgSQL error messages (Pavel Stehule)

E.96.3.9.2. PL/Python Server-Side Language

  • Add PL/Python SPI cursor support (Jan Urbanski)

    This allows PL/Python to read partial result sets.

  • Add result metadata functions to PL/Python (Peter Eisentraut)

    Specifically, this adds result object functions .colnames , .coltypes , and .coltypmods .

  • Remove support for Python 2.2 (Peter Eisentraut)

E.96.3.9.3. SQL Server-Side Language

  • Allow SQL -language functions to reference parameters by name (Matthew Draper)

    To use this, simply name the function arguments and then reference the argument names in the SQL function body.

E.96.3.10. Client Applications

  • Add initdb options --auth-local and --auth-host (Peter Eisentraut)

    This allows separate control of local and host pg_hba.conf authentication settings. --auth still controls both.

  • Add --replication / --no-replication flags to createuser to control replication permission (Fujii Masao)

  • Add the --if-exists option to dropdb and dropuser (Josh Kupershmidt)

  • Give command-line tools the ability to specify the name of the database to connect to, and fall back to template1 if a postgres database connection fails (Robert Haas)

E.96.3.10.1. psql

  • Add a display mode to auto-expand output based on the display width (Peter Eisentraut)

    This adds the auto option to the \x command, which switches to the expanded mode when the normal output would be wider than the screen.

  • Allow inclusion of a script file that is named relative to the directory of the file from which it was invoked (Gurjeet Singh)

    This is done with a new command \ir .

  • Add support for non- ASCII characters in psql variable names (Tom Lane)

  • Add support for major-version-specific .psqlrc files (Bruce Momjian)

    psql already supported minor-version-specific .psqlrc files.

  • Provide environment variable overrides for psql history and startup file locations (Andrew Dunstan)

    PSQL_HISTORY and PSQLRC now determine these file names if set.

  • Add a \setenv command to modify the environment variables passed to child processes (Andrew Dunstan)

  • Name psql 's temporary editor files with a .sql extension (Peter Eisentraut)

    This allows extension-sensitive editors to select the right mode.

  • Allow psql to use zero-byte field and record separators (Peter Eisentraut)

    Various shell tools use zero-byte (NUL) separators, e.g. find .

  • Make the \timing option report times for failed queries (Magnus Hagander)

    Previously times were reported only for successful queries.

  • Unify and tighten psql 's treatment of \copy and SQL COPY (Noah Misch)

    This fix makes failure behavior more predictable and honors \set ON_ERROR_ROLLBACK .

E.96.3.10.2. Informational Commands

  • Make \d on a sequence show the table/column name owning it (Magnus Hagander)

  • Show statistics target for columns in \d+ (Magnus Hagander)

  • Show role password expiration dates in \du (Fabrízio de Royes Mello)

  • Display comments for casts, conversions, domains, and languages (Josh Kupershmidt)

    These are included in the output of \dC+ , \dc+ , \dD+ , and \dL respectively.

  • Display comments for SQL / MED objects (Josh Kupershmidt)

    These are included in the output of \des+ , \det+ , and \dew+ for foreign servers, foreign tables, and foreign data wrappers respectively.

  • Change \dd to display comments only for object types without their own backslash command (Josh Kupershmidt)

E.96.3.10.3. Tab Completion

  • In psql tab completion, complete SQL keywords in either upper or lower case according to the new COMP_KEYWORD_CASE setting (Peter Eisentraut)

  • Add tab completion support for EXECUTE (Andreas Karlsson)

  • Allow tab completion of role references in GRANT / REVOKE (Peter Eisentraut)

  • Allow tab completion of file names to supply quotes, when necessary (Noah Misch)

  • Change tab completion support for TABLE to also include views (Magnus Hagander)

E.96.3.10.4. pg_dump

  • Add an --exclude-table-data option to pg_dump (Andrew Dunstan)

    This allows dumping of a table's definition but not its data, on a per-table basis.

  • Add a --section option to pg_dump and pg_restore (Andrew Dunstan)

    Valid values are pre-data , data , and post-data . The option can be given more than once to select two or more sections.

  • Make pg_dumpall dump all roles first, then all configuration settings on roles (Phil Sorber)

    This allows a role's configuration settings to mention other roles without generating an error.

  • Allow pg_dumpall to avoid errors if the postgres database is missing in the new cluster (Robert Haas)

  • Dump foreign server user mappings in user name order (Peter Eisentraut)

    This helps produce deterministic dump files.

  • Dump operators in a predictable order (Peter Eisentraut)

  • Tighten rules for when extension configuration tables are dumped by pg_dump (Tom Lane)

  • Make pg_dump emit more useful dependency information (Tom Lane)

    The dependency links included in archive-format dumps were formerly of very limited use, because they frequently referenced objects that appeared nowhere in the dump. Now they represent actual dependencies (possibly indirect) among the dumped objects.

  • Improve pg_dump 's performance when dumping many database objects (Tom Lane)

E.96.3.11. libpq

  • Allow libpq connection strings to have the format of a URI (Alexander Shulgin)

    The syntax begins with postgres:// . This can allow applications to avoid implementing their own parser for URIs representing database connections.

  • Add a connection option to disable SSL compression (Laurenz Albe)

    This can be used to remove the overhead of SSL compression on fast networks.

  • Add a single-row processing mode for better handling of large result sets (Kyotaro Horiguchi, Marko Kreen)

    Previously, libpq always collected the entire query result in memory before passing it back to the application.

  • Add const qualifiers to the declarations of the functions PQconnectdbParams , PQconnectStartParams , and PQpingParams (Lionel Elie Mamane)

  • Allow the .pgpass file to include escaped characters in the password field (Robert Haas)

  • Make library functions use abort() instead of exit() when it is necessary to terminate the process (Peter Eisentraut)

    This choice does not interfere with the normal exit codes used by the program, and generates a signal that can be caught by the caller.

E.96.3.12. Source Code

  • Remove dead ports (Peter Eisentraut)

    The following platforms are no longer supported: dgux, nextstep, sunos4, svr4, ultrix4, univel, bsdi.

  • Add support for building with MS Visual Studio 2010 (Brar Piening)

  • Enable compiling with the MinGW-w64 32-bit compiler (Lars Kanis)

  • Install plpgsql.h into include/server during installation (Heikki Linnakangas)

  • Improve the latch facility to include detection of postmaster death (Peter Geoghegan, Heikki Linnakangas, Tom Lane)

    This eliminates one of the main reasons that background processes formerly had to wake up to poll for events.

  • Use C flexible array members, where supported (Peter Eisentraut)

  • Improve the concurrent transaction regression tests ( isolationtester ) (Noah Misch)

  • Modify thread_test to create its test files in the current directory, rather than /tmp (Bruce Momjian)

  • Improve flex and bison warning and error reporting (Tom Lane)

  • Add memory barrier support (Robert Haas)

    This is currently unused.

  • Modify pgindent to use a typedef file (Bruce Momjian)

  • Add a hook for processing messages due to be sent to the server log (Martin Pihlak)

  • Add object access hooks for DROP commands (KaiGai Kohei)

  • Centralize DROP handling for some object types (KaiGai Kohei)

  • Add a pg_upgrade test suite (Peter Eisentraut)

  • Sync regular expression code with TCL 8.5.11 and improve internal processing (Tom Lane)

  • Move CRC tables to libpgport, and provide them in a separate include file (Daniel Farina)

  • Add options to git_changelog for use in major release note creation (Bruce Momjian)

  • Support Linux's /proc/self/oom_score_adj API (Tom Lane)

E.96.3.13. Additional Modules

  • Improve efficiency of dblink by using libpq's new single-row processing mode (Kyotaro Horiguchi, Marko Kreen)

    This improvement does not apply to dblink_send_query() / dblink_get_result() .

  • Support force_not_null option in file_fdw (Shigeru Hanada)

  • Implement dry-run mode for pg_archivecleanup (Gabriele Bartolini)

    This only outputs the names of files to be deleted.

  • Add new pgbench switches --unlogged-tables , --tablespace , and --index-tablespace (Robert Haas)

  • Change pg_test_fsync to test for a fixed amount of time, rather than a fixed number of cycles (Bruce Momjian)

    The -o /cycles option was removed, and -s /seconds added.

  • Add a pg_test_timing utility to measure clock monotonicity and timing overhead (Ants Aasma, Greg Smith)

  • Add a tcn (triggered change notification) module to generate NOTIFY events on table changes (Kevin Grittner)

E.96.3.13.1. pg_upgrade

  • Adjust pg_upgrade environment variables (Bruce Momjian)

    Rename data, bin, and port environment variables to begin with PG , and support PGPORTOLD / PGPORTNEW , to replace PGPORT .

  • Overhaul pg_upgrade logging and failure reporting (Bruce Momjian)

    Create four append-only log files, and delete them on success. Add -r / --retain option to unconditionally retain these files. Also remove pg_upgrade options -g / -G / -l options as unnecessary, and tighten log file permissions.

  • Make pg_upgrade create a script to incrementally generate more accurate optimizer statistics (Bruce Momjian)

    This reduces the time needed to generate minimal cluster statistics after an upgrade.

  • Allow pg_upgrade to upgrade an old cluster that does not have a postgres database (Bruce Momjian)

  • Allow pg_upgrade to handle cases where some old or new databases are missing, as long as they are empty (Bruce Momjian)

  • Allow pg_upgrade to handle configuration-only directory installations (Bruce Momjian)

  • In pg_upgrade , add -o / -O options to pass parameters to the servers (Bruce Momjian)

    This is useful for configuration-only directory installs.

  • Change pg_upgrade to use port 50432 by default (Bruce Momjian)

    This helps avoid unintended client connections during the upgrade.

  • Reduce cluster locking in pg_upgrade (Bruce Momjian)

    Specifically, only lock the old cluster if link mode is used, and do it right after the schema is restored.

E.96.3.13.2. pg_stat_statements

  • Allow pg_stat_statements to aggregate similar queries via SQL text normalization (Peter Geoghegan, Tom Lane)

    Users with applications that use non-parameterized SQL will now be able to monitor query performance without detailed log analysis.

  • Add dirtied and written block counts and read/write times to pg_stat_statements (Robert Haas, Ants Aasma)

  • Prevent pg_stat_statements from double-counting PREPARE and EXECUTE commands (Tom Lane)

E.96.3.13.3. sepgsql

  • Support SECURITY LABEL on global objects (KaiGai Kohei, Robert Haas)

    Specifically, add security labels to databases, tablespaces, and roles.

  • Allow sepgsql to honor database labels (KaiGai Kohei)

  • Perform sepgsql permission checks during the creation of various objects (KaiGai Kohei)

  • Add sepgsql_setcon() and related functions to control the sepgsql security domain (KaiGai Kohei)

  • Add a user space access cache to sepgsql to improve performance (KaiGai Kohei)

E.96.3.14. Documentation

  • Add a rule to optionally build HTML documentation using the stylesheet from the website (Magnus Hagander)

    Use gmake STYLE=website draft .

  • Improve EXPLAIN documentation (Tom Lane)

  • Document that user/database names are preserved with double-quoting by command-line tools like vacuumdb (Bruce Momjian)

  • Document the actual string returned by the client for MD5 authentication (Cyan Ogilvie)

  • Deprecate use of GLOBAL and LOCAL in CREATE TEMP TABLE (Noah Misch)

    PostgreSQL has long treated these keyword as no-ops, and continues to do so; but in future they might mean what the SQL standard says they mean, so applications should avoid using them.