4 Current maintainer: Bruce Momjian (bruce@momjian.us)
5 Last updated: Thu Oct 19 15:58:38 EDT 2006
7 The most recent version of this document can be viewed at
8 http://www.postgresql.org/docs/faqs.TODO.html.
10 #A hyphen, "-", marks changes that will appear in the upcoming 8.3 release.#
11 #A percent sign, "%", marks items that are easier to implement.#
13 Bracketed items, "[]", have more detail.
15 This list contains all known PostgreSQL bugs and feature requests. If
16 you would like to work on an item, please read the Developer's FAQ
23 * Allow major upgrades without dump/reload, perhaps using pg_upgrade
25 * Check for unreferenced table files created by transactions that were
26 in-progress when the server terminated abruptly
28 http://archives.postgresql.org/pgsql-patches/2006-06/msg00096.php
30 * Allow administrators to safely terminate individual sessions either
31 via an SQL function or SIGTERM
33 Lock table corruption following SIGTERM of an individual backend
34 has been reported in 8.0. A possible cause was fixed in 8.1, but
35 it is unknown whether other problems exist. This item mostly
36 requires additional testing rather than of writing any new code.
37 http://archives.postgresql.org/pgsql-hackers/2006-08/msg00174.php
39 * %Set proper permissions on non-system schemas during db creation
41 Currently all schemas are owned by the super-user because they are
42 copied from the template1 database.
44 * Support table partitioning that allows a single table to be stored
45 in subtables that are partitioned based on the primary key or a WHERE
47 * Add function to report the time of the most recent server reload
48 * Allow statistics collector information to be pulled from the collector
49 process directly, rather than requiring the collector to write a
50 filesystem file twice a second?
51 * Allow log_min_messages to be specified on a per-module basis
53 This would allow administrators to see more detailed information from
54 specific sections of the backend, e.g. checkpoints, autovacuum, etc.
55 Another idea is to allow separate configuration files for each module,
56 or allow arbitrary SET commands to be passed to them.
58 * Simplify ability to create partitioned tables
60 This would allow creation of partitioned tables without requiring
61 creation of rules for INSERT/UPDATE/DELETE, and constraints for
62 rapid partition selection. Options could include range and hash
65 * Allow auto-selection of partitioned tables for min/max() operations
66 * Allow more complex user/database default GUC settings
68 Currently, ALTER USER and ALTER DATABASE support per-user and
69 per-database defaults. Consider adding per-user-and-database
70 defaults so things like search_path can be defaulted for a
71 specific user connecting to a specific database.
73 * Improve replication solutions
77 You can use any of the master/slave replication servers to use a
78 standby server for data warehousing. To allow read/write queries to
79 multiple servers, you need multi-master replication like pgcluster.
81 o Allow replication over unreliable or non-persistent links
86 o Allow commenting of variables in postgresql.conf to restore them
89 Currently, if a variable is commented out, it keeps the
90 previous uncommented value until a server restarted.
91 http://archives.postgresql.org/pgsql-hackers/2006-09/msg01481.php
93 o Allow pg_hba.conf to specify host names along with IP addresses
95 Host name lookup could occur when the postmaster reads the
96 pg_hba.conf file, or when the backend starts. Another
97 solution would be to reverse lookup the connection IP and
98 check that hostname against the host names in pg_hba.conf.
99 We could also then check that the host name maps to the IP
102 o %Allow postgresql.conf file values to be changed via an SQL
103 API, perhaps using SET GLOBAL
104 o Allow the server to be stopped/restarted via an SQL API
105 o Issue a warning if a change-on-restart-only postgresql.conf value
106 is modified and the server config files are reloaded
107 o Mark change-on-restart-only values in postgresql.conf
112 o Allow a database in tablespace t1 with tables created in
113 tablespace t2 to be used as a template for a new database created
114 with default tablespace t2
116 All objects in the default database tablespace must have default
117 tablespace specifications. This is because new databases are
118 created by copying directories. If you mix default tablespace
119 tables and tablespace-specified tables in the same directory,
120 creating a new database from such a mixed directory would create a
121 new database with tables that had incorrect explicit tablespaces.
122 To fix this would require modifying pg_class in the newly copied
123 database, which we don't currently do.
125 o Allow reporting of which objects are in which tablespaces
127 This item is difficult because a tablespace can contain objects
128 from multiple databases. There is a server-side function that
129 returns the databases which use a specific tablespace, so this
130 requires a tool that will call that function and connect to each
131 database to find the objects in each database for that tablespace.
133 o %Add a GUC variable to control the tablespace for temporary objects
136 It could start with a random tablespace from a supplied list and
137 cycle through the list.
139 o Allow WAL replay of CREATE TABLESPACE to work when the directory
140 structure on the recovery computer is different from the original
142 o Allow per-tablespace quotas
145 * Point-In-Time Recovery (PITR)
147 o Allow a warm standby system to also allow read-only statements
150 This is useful for checking PITR recovery.
152 o %Create dump tool for write-ahead logs for use in determining
153 transaction id for point-in-time recovery
154 o Allow the PITR process to be debugged and data examined
160 * Allow server log information to be output as INSERT statements
162 This would allow server log information to be easily loaded into
163 a database for analysis.
165 * %Add ability to monitor the use of temporary sort files
171 * Improve the MONEY data type
173 Change the MONEY data type to use DECIMAL internally, with special
174 locale-aware output formatting.
175 http://archives.postgresql.org/pgsql-general/2005-08/msg01432.php
176 http://archives.postgresql.org/pgsql-hackers/2006-09/msg01107.php
178 * Change NUMERIC to enforce the maximum precision
179 * Add NUMERIC division operator that doesn't round?
181 Currently NUMERIC _rounds_ the result to the specified precision.
182 This means division can return a result that multiplied by the
183 divisor is greater than the dividend, e.g. this returns a value > 10:
185 SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
187 The positive modulus result returned by NUMERICs might be considered
188 inaccurate, in one sense.
190 * Fix data types where equality comparison isn't intuitive, e.g. box
191 * Allow user-defined types to specify a type modifier at table creation
193 * Allow user-defined types to accept 'typmod' parameters
195 http://archives.postgresql.org/pgsql-hackers/2005-08/msg01142.php
196 http://archives.postgresql.org/pgsql-hackers/2005-09/msg00012.php
197 http://archives.postgresql.org/pgsql-hackers/2006-08/msg00149.php
199 * Add support for public SYNONYMs
201 http://archives.postgresql.org/pgsql-hackers/2006-03/msg00519.php
203 * Fix CREATE CAST on DOMAINs
205 http://archives.postgresql.org/pgsql-hackers/2006-05/msg00072.php
206 http://archives.postgresql.org/pgsql-hackers/2006-09/msg01681.php
208 * Add Globally/Universally Unique Identifier (GUID/UUID)
210 http://archives.postgresql.org/pgsql-patches/2006-09/msg00209.php
212 * Add support for SQL-standard GENERATED/IDENTITY columns
214 http://archives.postgresql.org/pgsql-hackers/2006-07/msg00543.php
216 * Support a data type with specific enumerated values (ENUM)
218 http://archives.postgresql.org/pgsql-hackers/2006-08/msg00979.php
220 * Improve XML support
222 http://developer.postgresql.org/index.php/XML_Support
226 o Allow infinite dates and intervals just like infinite timestamps
227 o Merge hardwired timezone names with the TZ database; allow either
228 kind everywhere a TZ name is currently taken
229 o Allow TIMESTAMP WITH TIME ZONE to store the original timezone
230 information, either zone name or offset from UTC [timezone]
232 If the TIMESTAMP value is stored with a time zone name, interval
233 computations should adjust based on the time zone rules.
235 o Fix SELECT '0.01 years'::interval, '0.01 months'::interval
236 o Add a GUC variable to allow output of interval values in ISO8601
238 o Improve timestamptz subtraction to be DST-aware
240 Currently, subtracting one date from another that crosses a
241 daylight savings time adjustment can return '1 day 1 hour', but
242 adding that back to the first date returns a time one hour in
243 the future. This is caused by the adjustment of '25 hours' to
244 '1 day 1 hour', and '1 day' is the same time the next day, even
245 if daylight savings adjustments are involved.
247 o Fix interval display to support values exceeding 2^31 hours
248 o Add overflow checking to timestamp and interval arithmetic
249 o Add ISO INTERVAL handling
251 http://archives.postgresql.org/pgsql-hackers/2006-01/msg00250.php
252 http://archives.postgresql.org/pgsql-bugs/2006-04/msg00248.php
254 o Support ISO INTERVAL syntax if units cannot be determined from
255 the string, and are supplied after the string
257 The SQL standard states that the units after the string
258 specify the units of the string, e.g. INTERVAL '2' MINUTE
259 should return '00:02:00'. The current behavior has the units
260 restrict the interval value to the specified unit or unit
261 range, INTERVAL '70' SECOND returns '00:00:10'.
263 For syntax that isn't uniquely ISO or PG syntax, like '1' or
264 '1:30', treat as ISO if there is a range specification clause,
265 and as PG if there no clause is present, e.g. interpret '1:30'
266 MINUTE TO SECOND as '1 minute 30 seconds', and interpret
267 '1:30' as '1 hour, 30 minutes'.
269 This makes common cases like SELECT INTERVAL '1' MONTH
270 SQL-standard results. The SQL standard supports a limited
271 number of unit combinations and doesn't support unit names in
272 the string. The PostgreSQL syntax is more flexible in the
273 range of units supported, e.g. PostgreSQL supports '1 year 1
274 hour', while the SQL standard does not.
276 o Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
277 o Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
278 INTERVAL MONTH), and this should return '12 months'
279 o Round or truncate values to the requested precision, e.g.
280 INTERVAL '11 months' AS YEAR should return one or zero
281 o Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
286 o Delay resolution of array expression's data type so assignment
287 coercion can be performed on empty array expressions
288 o Add support for arrays of domains
289 o Add support for arrays of complex types
294 o Improve vacuum of large objects, like /contrib/vacuumlo?
295 o Add security checking for large objects
296 o Auto-delete large objects when referencing row is deleted
298 /contrib/lo offers this functionality.
300 o Allow read/write into TOAST values like large objects
302 This requires the TOAST column to be stored EXTERNAL.
304 o Add API for 64-bit large object access
306 http://archives.postgresql.org/pgsql-hackers/2005-09/msg00781.php
312 * Allow INET subnet tests using non-constants to be indexed
313 * %Add pg_get_acldef(), pg_get_typedefault(), pg_get_attrdef(),
314 pg_get_tabledef(), pg_get_domaindef(), pg_get_functiondef()
316 These would be for application use, not for use by pg_dump.
318 * Allow to_date() and to_timestamp() accept localized month names
319 * Add missing parameter handling in to_char()
321 http://archives.postgresql.org/pgsql-hackers/2005-12/msg00948.php
323 * Allow functions to have a schema search path specified at creation time
324 * Allow substring/replace() to get/set bit values
325 * Allow to_char() on interval values to accumulate the highest unit
328 Some special format flag would be required to request such
329 accumulation. Such functionality could also be added to EXTRACT.
330 Prevent accumulation that crosses the month/day boundary because of
331 the uneven number of days in a month.
333 o to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
334 o to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
335 o to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
336 o to_char(INTERVAL '3 years 5 months','MM') => 41
338 * Add SPI_gettypmod() to return the typemod for a TupleDesc
339 * Allow inlining of set-returning functions
342 Multi-Language Support
343 ======================
345 * Add NCHAR (as distinguished from ordinary varchar),
346 * Allow locale to be set at database creation
348 Currently locale can only be set during initdb. No global tables have
349 locale-aware columns. However, the database template used during
350 database creation might have locale-aware indexes. The indexes would
351 need to be reindexed to match the new locale.
353 * Allow encoding on a per-column basis optionally using the ICU library:
355 Right now only one encoding is allowed per database. [locale]
356 http://archives.postgresql.org/pgsql-hackers/2005-03/msg00932.php
357 http://archives.postgresql.org/pgsql-patches/2005-08/msg00309.php
358 http://archives.postgresql.org/pgsql-patches/2006-03/msg00233.php
359 http://archives.postgresql.org/pgsql-hackers/2006-09/msg00662.php
361 * Add CREATE COLLATE? [locale]
362 * Support multiple simultaneous character sets, per SQL92
363 * Improve UTF8 combined character handling?
364 * Add octet_length_server() and octet_length_client()
365 * Make octet_length_client() the same as octet_length()?
366 * Fix problems with wrong runtime encoding conversion for NLS message files
367 * Add URL to more complete multi-byte regression tests
369 http://archives.postgresql.org/pgsql-hackers/2005-07/msg00272.php
371 * Fix ILIKE and regular expressions to handle case insensitivity
372 properly in multibyte encodings
374 http://archives.postgresql.org/pgsql-bugs/2005-10/msg00001.php
375 http://archives.postgresql.org/pgsql-patches/2005-11/msg00173.php
377 * Set client encoding based on the client operating system encoding
379 Currently client_encoding is set in postgresql.conf, which
380 defaults to the server encoding.
381 http://archives.postgresql.org/pgsql-hackers/2006-08/msg01696.php
387 * Automatically create rules on views so they are updateable, per SQL99
389 We can only auto-create rules for simple views. For more complex
390 cases users will still have to write rules manually.
391 http://archives.postgresql.org/pgsql-hackers/2006-03/msg00586.php
393 * Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
394 * Allow NOTIFY in rules involving conditionals
395 * Allow VIEW/RULE recompilation when the underlying tables change
397 Another issue is whether underlying table changes should be reflected
398 in the view, e.g. should SELECT * show additional columns if they
399 are added after the view is created.
405 * Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
406 * Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
407 * %Allow SET CONSTRAINTS to be qualified by schema/table name
408 * %Add a separate TRUNCATE permission
410 Currently only the owner can TRUNCATE a table because triggers are not
411 called, and the table is locked in exclusive mode.
413 * Allow PREPARE of cursors
414 * Allow finer control over the caching of prepared query plans
416 Currently, queries prepared via the libpq API are planned on first
417 execute using the supplied parameters --- allow SQL PREPARE to do the
418 same. Also, allow control over replanning prepared queries either
419 manually or automatically when statistics for execute parameters
420 differ dramatically from those used during planning.
422 * Invalidate prepared queries, like INSERT, when the table definition
424 * Allow LISTEN/NOTIFY to store info in memory rather than tables?
426 Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
427 such information in memory would improve performance.
429 * Add optional textual message to NOTIFY
431 This would allow an informational message to be added to the notify
432 message, perhaps indicating the row modified or other custom
435 * Add a GUC variable to warn about non-standard SQL usage in queries
436 * Add SQL-standard MERGE command, typically used to merge two tables
439 This is similar to UPDATE, then for unmatched rows, INSERT.
440 Whether concurrent access allows modifications which could cause
441 row loss is implementation independent.
443 * Add REPLACE or UPSERT command that does UPDATE, or on failure, INSERT
446 To implement this cleanly requires that the table have a unique index
447 so duplicate checking can be easily performed. It is possible to
448 do it without a unique index if we require the user to LOCK the table
451 * Add NOVICE output level for helpful messages like automatic sequence/index
453 * Add RESET CONNECTION command to reset all session state
455 This would include resetting of all variables (RESET ALL), dropping of
456 temporary tables, removing any NOTIFYs, cursors, open transactions,
457 prepared queries, currval()s, etc. This could be used for connection
458 pooling. We could also change RESET ALL to have this functionality.
459 The difficult of this features is allowing RESET ALL to not affect
460 changes made by the interface driver for its internal use. One idea
461 is for this to be a protocol-only feature. Another approach is to
462 notify the protocol when a RESET CONNECTION command is used.
463 http://archives.postgresql.org/pgsql-patches/2006-04/msg00192.php
465 * Add GUC to issue notice about statements that use unjoined tables
466 * Allow EXPLAIN to identify tables that were skipped because of
468 * Allow EXPLAIN output to be more easily processed by scripts
469 * Enable standard_conforming_strings
470 * Make standard_conforming_strings the default in 8.3?
472 When this is done, backslash-quote should be prohibited in non-E''
473 strings because of possible confusion over how such strings treat
474 backslashes. Basically, '' is always safe for a literal single
475 quote, while \' might or might not be based on the backslash
478 * Simplify dropping roles that have objects in several databases
479 * Allow COMMENT ON to accept an expression rather than just a string
480 * Allow the count returned by SELECT, etc to be to represent as an int64
481 to allow a higher range of values
482 * Make CLUSTER preserve recently-dead tuples per MVCC requirements
483 * Add SQL99 WITH clause to SELECT
484 * Add SQL:2003 WITH RECURSIVE (hierarchical) queries to SELECT
485 * Add DEFAULT .. AS OWNER so permission checks are done as the table
488 This would be useful for SERIAL nextval() calls and CHECK constraints.
490 * Add a GUC to control whether BEGIN inside a transcation should abort
492 * Allow DISTINCT to work in multiple-argument aggregate calls
493 * Add column to pg_stat_activity that shows the progress of long-running
494 commands like CREATE INDEX and VACUUM
495 * Implement SQL:2003 window functions
500 o Allow CREATE TABLE AS to determine column lengths for complex
501 expressions like SELECT col1 || col2
502 o Use more reliable method for CREATE DATABASE to get a consistent
507 o Allow UPDATE tab SET ROW (col, ...) = (SELECT...)
509 http://archives.postgresql.org/pgsql-hackers/2006-07/msg01306.php
514 o %Have ALTER TABLE RENAME rename SERIAL sequence names
515 o Add ALTER DOMAIN to modify the underlying data type
516 o %Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
518 http://archives.postgresql.org/pgsql-patches/2006-02/msg00168.php
520 o %Allow ALTER TABLE to change constraint deferrability and actions
521 o Add missing object types for ALTER ... SET SCHEMA
522 o Allow ALTER TABLESPACE to move to different directories
523 o Allow databases to be moved to different tablespaces
524 o Allow moving system tables to other tablespaces, where possible
526 Currently non-global system tables must be in the default database
527 tablespace. Global system tables can never be moved.
529 o Prevent parent tables from altering or dropping constraints
530 like CHECK that are inherited by child tables unless CASCADE
532 o %Prevent child tables from altering or dropping constraints
533 like CHECK that were inherited from the parent table
534 o Have ALTER INDEX update the name of a constraint using that index
535 o Add ALTER TABLE RENAME CONSTRAINT, update index name also
540 o Automatically maintain clustering on a table
542 This might require some background daemon to maintain clustering
543 during periods of low usage. It might also require tables to be only
544 partially filled for easier reorganization. Another idea would
545 be to create a merged heap/index data file so an index lookup would
546 automatically access the heap data too. A third idea would be to
547 store heap rows in hashed groups, perhaps using a user-supplied
549 http://archives.postgresql.org/pgsql-performance/2004-08/msg00349.php
551 o %Add default clustering to system tables
553 To do this, determine the ideal cluster index for each system
554 table and set the cluster setting during initdb.
559 o Allow COPY to report error lines and continue
561 This requires the use of a savepoint before each COPY line is
562 processed, with ROLLBACK on COPY failure.
564 o Allow COPY on a newly-created table to skip WAL logging
566 On crash recovery, the table involved in the COPY would
567 be removed or have its heap and index files truncated. One
568 issue is that no other backend should be able to add to
569 the table at the same time, which is something that is
575 o Allow column-level privileges
576 o %Allow GRANT/REVOKE permissions to be applied to all schema objects
579 The proposed syntax is:
580 GRANT SELECT ON ALL TABLES IN public TO phpuser;
581 GRANT SELECT ON NEW TABLES IN public TO phpuser;
583 o Allow GRANT/REVOKE permissions to be inherited by objects based on
586 o Allow SERIAL sequences to inherit permissions from the base table?
591 o Allow UPDATE/DELETE WHERE CURRENT OF cursor
593 This requires using the row ctid to map cursor rows back to the
594 original heap row. This become more complicated if WITH HOLD cursors
595 are to be supported because WITH HOLD cursors have a copy of the row
596 and no FOR UPDATE lock.
598 o Prevent DROP TABLE from dropping a row referenced by its own open
604 o Allow INSERT/UPDATE of the system-generated oid value for a row
605 o In rules, allow VALUES() to contain a mixture of 'old' and 'new'
611 o Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
613 o Add SET PATH for schemas?
615 This is basically the same as SET search_path.
618 * Referential Integrity
620 o Add MATCH PARTIAL referential integrity
621 o Change foreign key constraint for array -> element to mean element
623 o Enforce referential integrity for system tables
624 o Fix problem when cascading referential triggers make changes on
625 cascaded tables, seeing the tables in an intermediate state
627 http://archives.postgresql.org/pgsql-hackers/2005-09/msg00174.php
628 http://archives.postgresql.org/pgsql-hackers/2005-09/msg00174.php
630 o Allow DEFERRABLE and end-of-statement UNIQUE constraints?
632 This would allow UPDATE tab SET col = col + 1 to work if col has
633 a unique index. Currently, uniqueness checks are done while the
634 command is being executed, rather than at the end of the statement
636 http://people.planetpostgresql.org/greg/index.php?/archives/2006/06/10.html
637 http://archives.postgresql.org/pgsql-hackers/2006-09/msg01458.php
640 * Server-Side Languages
643 o Fix RENAME to work on variables other than OLD/NEW
644 o Allow function parameters to be passed by name,
645 get_employee_salary(12345 AS emp_id, 2001 AS tax_year)
646 o Add Oracle-style packages (Pavel)
648 A package would be a schema with session-local variables,
649 public/private functions, and initialization functions. It
650 is also possible to implement these capabilities
651 in all schemas and not use a separate "packages"
653 http://archives.postgresql.org/pgsql-hackers/2006-08/msg00384.php
655 o Allow handling of %TYPE arrays, e.g. tab.col%TYPE[]
656 o Allow listing of record column names, and access to
657 record columns via variables, e.g. columns := r.(*),
660 http://archives.postgresql.org/pgsql-patches/2005-07/msg00458.php
661 http://archives.postgresql.org/pgsql-patches/2006-05/msg00302.php
662 http://archives.postgresql.org/pgsql-patches/2006-06/msg00031.php
665 o Add single-step debugging of functions
666 o Add support for WITH HOLD and SCROLL cursors
668 PL/pgSQL cursors should support the same syntax as
671 o Allow PL/RETURN to return row or record functions
673 http://archives.postgresql.org/pgsql-patches/2005-11/msg00045.php
675 o Fix memory leak from exceptions
677 http://archives.postgresql.org/pgsql-performance/2006-06/msg00305.php
679 o Fix problems with RETURN NEXT on tables with
680 dropped/added columns after function creation
682 http://archives.postgresql.org/pgsql-patches/2006-02/msg00165.php
685 o Add table function support to pltcl, plpython
686 o Add support for polymorphic arguments and return types to
687 languages other than PL/PgSQL
688 o Add capability to create and call PROCEDURES
689 o Add support for OUT and INOUT parameters to languages other
691 o Add PL/Python tracebacks
693 http://archives.postgresql.org/pgsql-patches/2006-02/msg00288.php
699 * Have pg_ctl look at PGHOST in case it is a socket directory?
700 * Allow pg_ctl to work properly with configuration files located outside
703 pg_ctl can not read the pid file because it isn't located in the
704 config directory but in the PGDATA directory. The solution is to
705 allow pg_ctl to read and understand postgresql.conf to find the
706 data_directory value.
711 o Have psql show current values for a sequence
712 o Move psql backslash database information into the backend, use
713 mnemonic commands? [psql]
715 This would allow non-psql clients to pull the same information out
716 of the database as psql.
718 o Fix psql's \d commands more consistent
720 http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php
721 http://archives.postgresql.org/pgsql-hackers/2004-11/msg00014.php
723 o Allow psql \pset boolean variables to set to fixed values, rather
725 o Consistently display privilege information for all objects in psql
726 o Add auto-expanded mode so expanded output is used if the row
727 length is wider than the screen width.
729 Consider using auto-expanded mode for backslash commands like \df+.
731 o Prevent tab completion of SET TRANSACTION from querying the
732 database and therefore preventing the transaction isolation
733 level from being set.
735 Currently, SET <tab> causes a database lookup to check all
736 supported session variables. This query causes problems
737 because setting the transaction isolation level must be the
738 first statement of a transaction.
743 o %Add dumping of comments on index columns and composite type columns
744 o %Add full object name to the tag field. eg. for operators we need
745 '=(integer, integer)', instead of just '='.
746 o Add pg_dumpall custom format dumps?
747 o Remove unnecessary function pointer abstractions in pg_dump source
749 o Allow selection of individual object(s) of all types, not just
751 o In a selective dump, allow dumping of an object and all its
753 o Add options like pg_restore -l and -L to pg_dump
754 o Stop dumping CASCADE on DROP TYPE commands in clean mode
755 o Allow pg_dump --clean to drop roles that own objects or have
757 o Add -f to pg_dumpall
764 Document differences between ecpg and the SQL standard and
765 information about the Informix-compatibility module.
767 o Solve cardinality > 1 for input descriptors / variables?
768 o Add a semantic check level, e.g. check if a table really exists
769 o fix handling of DB attributes that are arrays
770 o Use backend PREPARE/EXECUTE facility for ecpg where possible
772 o Fix nested C comments
773 o %sqlwarn[6] should be 'W' if the PRECISION or SCALE value specified
774 o Make SET CONNECTION thread-aware, non-standard?
775 o Allow multidimensional arrays
776 o Add internationalized message strings
777 o Implement COPY FROM STDIN
782 o Add PQescapeIdentifierConn()
783 o Prevent PQfnumber() from lowercasing unquoted the column name
785 PQfnumber() should never have been doing lowercasing, but
786 historically it has so we need a way to prevent it
788 o Allow statement results to be automatically batched to the client
790 Currently, all statement results are transferred to the libpq
791 client before libpq makes the results available to the
792 application. This feature would allow the application to make
793 use of the first result rows while the rest are transferred, or
794 held on the server waiting for them to be requested by libpq.
795 One complexity is that a statement like SELECT 1/col could error
796 out mid-way through the result set.
797 * Fix SSL retry to avoid useless repeated connection attempts and
798 ensuing misleading error messages
804 * Add deferred trigger queue file
806 Right now all deferred trigger information is stored in backend
807 memory. This could exhaust memory for very large trigger queues.
808 This item involves dumping large queues into files.
810 * Allow triggers to be disabled in only the current session.
812 This is currently possible by starting a multi-statement transaction,
813 modifying the system tables, performing the desired SQL, restoring the
814 system tables, and committing the transaction. ALTER TABLE ...
815 TRIGGER requires a table lock so it is not ideal for this usage.
817 * With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
819 If the dump is known to be valid, allow foreign keys to be added
820 without revalidating the data.
822 * Allow statement-level triggers to access modified rows
823 * Support triggers on columns
825 http://archives.postgresql.org/pgsql-patches/2005-07/msg00107.php
827 * Allow AFTER triggers on system tables
829 System tables are modified in many places in the backend without going
830 through the executor and therefore not causing triggers to fire. To
831 complete this item, the functions that modify system tables will have
838 * Flush cached query plans when the dependent objects change,
839 when the cardinality of parameters changes dramatically, or
840 when new ANALYZE statistics are available
842 A more complex solution would be to save multiple plans for different
843 cardinality and use the appropriate plan based on the EXECUTE values.
845 * Track dependencies in function bodies and recompile/invalidate
847 This is particularly important for references to temporary tables
848 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
849 in PL/PgSQL is to use EXECUTE. One complexity is that a function
850 might itself drop and recreate dependent tables, causing it to
851 invalidate its own query plan.
857 * Allow inherited tables to inherit index, UNIQUE constraint, and primary
859 * UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
860 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
863 The main difficulty with this item is the problem of creating an index
864 that can span more than one table.
866 * Allow SELECT ... FOR UPDATE on inherited tables
867 * Add UNIQUE capability to non-btree indexes
868 * Prevent index uniqueness checks when UPDATE does not modify the column
870 Uniqueness (index) checks are done when updating a column even if the
871 column is not modified by the UPDATE.
873 * Allow the creation of on-disk bitmap indexes which can be quickly
874 combined with other bitmap indexes
876 Such indexes could be more compact if there are only a few distinct values.
877 Such indexes can also be compressed. Keeping such indexes updated can be
879 http://archives.postgresql.org/pgsql-patches/2005-07/msg00512.php
881 * Allow use of indexes to search for NULLs
883 One solution is to create a partial index on an IS NULL expression.
885 * Allow accurate statistics to be collected on indexes with more than
886 one column or expression indexes, perhaps using per-index statistics
887 * Allow the creation of indexes with mixed ascending/descending
890 This is possible now by creating an operator class with reversed sort
891 operators. One complexity is that NULLs would then appear at the start
892 of the result set, and this might affect certain sort types, like
895 * Allow constraint_exclusion to work for UNIONs like it does for
896 inheritance, allow it to work for UPDATE and DELETE statements, and allow
897 it to be used for all statements with little performance impact
898 * Allow CREATE INDEX to take an additional parameter for use with
900 * Consider compressing indexes by storing key values duplicated in
901 several rows as a single index entry
903 This is difficult because it requires datatype-specific knowledge.
908 o Add more GIST index support for geometric data types
909 o Allow GIST indexes to create certain complex index types, like
910 digital trees (see Aoki)
914 o Pack hash index buckets onto disk pages more efficiently
916 Currently only one hash bucket can be stored on a page. Ideally
917 several hash buckets could be stored on a single page and greater
918 granularity used for the hash algorithm.
920 o Consider sorting hash buckets so entries can be found using a
921 binary search, rather than a linear scan
923 o In hash indexes, consider storing the hash value with or instead
926 o Add WAL logging for crash recovery
927 o Allow multi-column hash indexes
933 * Improve commit_delay handling to reduce fsync()
934 * Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
936 Ideally this requires a separate test program that can be run
937 at initdb time or optionally later. Consider O_SYNC when
940 * %Add an option to sync() before fsync()'ing checkpoint files
941 * Add program to test if fsync has a delay compared to non-fsync
947 * Allow free-behind capability for large sequential scans, perhaps using
950 Posix_fadvise() can control both sequential/random file caching and
951 free-behind behavior, but it is unclear how the setting affects other
952 backends that also have the file open, and the feature is not supported
953 on all operating systems.
957 We could use a fixed row count and a +/- count to follow MVCC
958 visibility rules, or a single cached value could be used and
959 invalidated if anyone modifies the table. Another idea is to
960 get a count directly from a unique index, but for this to be
961 faster than a sequential scan it must avoid access to the heap
962 to obtain tuple visibility information.
964 * Add estimated_count(*) to return an estimate of COUNT(*)
966 This would use the planner ANALYZE statistics to return an estimated
969 * Allow data to be pulled directly from indexes
971 Currently indexes do not have enough tuple visibility information
972 to allow data to be pulled from the index without also accessing
973 the heap. One way to allow this is to set a bit on index tuples
974 to indicate if a tuple is currently visible to all transactions
975 when the first valid heap lookup happens. This bit would have to
976 be cleared when a heap tuple is expired.
978 Another idea is to maintain a bitmap of heap pages where all rows
979 are visible to all backends, and allow index lookups to reference
980 that bitmap to avoid heap lookups, perhaps the same bitmap we might
981 add someday to determine which heap pages need vacuuming. Frequently
982 accessed bitmaps would have to be stored in shared memory. One 8k
983 page of bitmaps could track 512MB of heap pages.
985 * Consider automatic caching of statements at various levels:
991 * Allow sequential scans to take advantage of other concurrent
992 sequential scans, also called "Synchronised Scanning"
994 One possible implementation is to start sequential scans from the lowest
995 numbered buffer in the shared cache, and when reaching the end wrap
996 around to the beginning, rather than always starting sequential scans
997 at the start of the table.
999 * Consider increasing internal areas when shared buffers is increased
1001 http://archives.postgresql.org/pgsql-hackers/2005-10/msg01419.php
1007 * Improve speed with indexes
1009 For large table adjustments during VACUUM FULL, it is faster to
1010 reindex rather than update the index.
1012 * Reduce lock time during VACUUM FULL by moving tuples with read lock,
1013 then write lock and truncate table
1015 Moved tuples are invisible to other backends so they don't require a
1016 write lock. However, the read lock promotion to write lock could lead
1017 to deadlock situations.
1019 * Auto-fill the free space map by scanning the buffer cache or by
1020 checking pages written by the background writer
1022 http://archives.postgresql.org/pgsql-hackers/2006-02/msg01125.php
1023 http://archives.postgresql.org/pgsql-hackers/2006-03/msg00011.php
1025 * Create a bitmap of pages that need vacuuming
1027 Instead of sequentially scanning the entire table, have the background
1028 writer or some other process record pages that have expired rows, then
1029 VACUUM can look at just those pages rather than the entire table. In
1030 the event of a system crash, the bitmap would probably be invalidated.
1031 One complexity is that index entries still have to be vacuumed, and
1032 doing this without an index scan (by using the heap values to find the
1033 index entry) might be slow and unreliable, especially for user-defined
1036 * Allow FSM to return free space toward the beginning of the heap file,
1037 in hopes that empty pages at the end can be truncated by VACUUM
1038 * Allow FSM page return free space based on table clustering, to assist
1039 in maintaining clustering?
1040 * Consider shrinking expired tuples to just their headers
1042 http://archives.postgresql.org/pgsql-patches/2006-03/msg00142.php
1044 * Allow heap reuse of UPDATEd rows if no indexed columns are changed,
1045 and old and new versions are on the same heap page?
1047 While vacuum handles DELETEs fine, updating of non-indexed columns, like
1048 counters, are difficult for VACUUM to handle efficiently. This method
1049 is possible for same-page updates because a single index row can be
1050 used to point to both old and new values.
1051 http://archives.postgresql.org/pgsql-hackers/2006-06/msg01305.php
1052 http://archives.postgresql.org/pgsql-hackers/2006-06/msg01534.php
1054 * Reuse index tuples that point to heap tuples that are not visible to
1058 o Use free-space map information to guide refilling
1059 o %Issue log message to suggest VACUUM FULL if a table is nearly
1061 o Improve xid wraparound detection by recording per-table rather
1063 o Consider logging activity either to the logs or a system view
1064 o Turn on by default
1066 http://archives.postgresql.org/pgsql-hackers/2006-08/msg01852.php
1072 * Fix priority ordering of read and write light-weight locks (Neil)
1075 Startup Time Improvements
1076 =========================
1078 * Experiment with multi-threaded backend for backend creation [thread]
1080 This would prevent the overhead associated with process creation. Most
1081 operating systems have trivial process creation time compared to
1082 database startup overhead, but a few operating systems (Win32,
1083 Solaris) might benefit from threading. Also explore the idea of
1084 a single session using multiple threads to execute a statement faster.
1086 * Experiment with multi-threaded backend better resource utilization
1088 This would allow a single query to make use of multiple CPU's or
1089 multiple I/O channels simultaneously. One idea is to create a
1090 background reader that can pre-fetch sequential and index scan
1091 pages needed by other backends. This could be expanded to allow
1092 concurrent reads from multiple devices in a partitioned table.
1094 * Add connection pooling
1096 It is unclear if this should be done inside the backend code or done
1097 by something external like pgpool. The passing of file descriptors to
1098 existing backends is one of the difficulties with a backend approach.
1104 * Eliminate need to write full pages to WAL before page modification [wal]
1106 Currently, to protect against partial disk page writes, we write
1107 full page images to WAL before they are modified so we can correct any
1108 partial page writes during recovery. These pages can also be
1109 eliminated from point-in-time archive files.
1111 o When off, write CRC to WAL and check file system blocks
1114 If CRC check fails during recovery, remember the page in case
1115 a later CRC for that page properly matches.
1117 o Write full pages during file system write and not when
1118 the page is modified in the buffer cache
1120 This allows most full page writes to happen in the background
1121 writer. It might cause problems for applying WAL on recovery
1122 into a partially-written page, but later the full page will be
1125 * Allow WAL traffic to be streamed to another server for stand-by
1127 * Reduce WAL traffic so only modified values are written rather than
1129 * Allow the pg_xlog directory location to be specified during initdb
1130 with a symlink back to the /data location
1131 * Allow WAL information to recover corrupted pg_controldata
1133 http://archives.postgresql.org/pgsql-patches/2006-06/msg00025.php
1135 * Find a way to reduce rotational delay when repeatedly writing
1138 Currently fsync of WAL requires the disk platter to perform a full
1139 rotation to fsync again. One idea is to write the WAL to different
1140 offsets that might reduce the rotational delay.
1142 * Allow buffered WAL writes and fsync
1144 Instead of guaranteeing recovery of all committed transactions, this
1145 would provide improved performance by delaying WAL writes and fsync
1146 so an abrupt operating system restart might lose a few seconds of
1147 committed transactions but still be consistent. We could perhaps
1148 remove the 'fsync' parameter (which results in an an inconsistent
1149 database) in favor of this capability.
1151 * Allow WAL logging to be turned off for a table, but the table
1152 might be dropped or truncated during crash recovery [walcontrol]
1154 Allow tables to bypass WAL writes and just fsync() dirty pages on
1155 commit. This should be implemented using ALTER TABLE, e.g. ALTER
1156 TABLE PERSISTENCE [ DROP | TRUNCATE | DEFAULT ]. Tables using
1157 non-default logging should not use referential integrity with
1158 default-logging tables. A table without dirty buffers during a
1159 crash could perhaps avoid the drop/truncate.
1161 * Allow WAL logging to be turned off for a table, but the table would
1162 avoid being truncated/dropped [walcontrol]
1164 To do this, only a single writer can modify the table, and writes
1165 must happen only on new pages so the new pages can be removed during
1166 crash recovery. Readers can continue accessing the table. Such
1167 tables probably cannot have indexes. One complexity is the handling
1168 of indexes on TOAST tables.
1171 Optimizer / Executor
1172 ====================
1174 * Improve selectivity functions for geometric operators
1175 * Allow ORDER BY ... LIMIT # to select high/low value without sort or
1176 index using a sequential scan for highest/lowest values
1178 Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
1179 all values to return the high/low value. Instead The idea is to do a
1180 sequential scan to find the high/low value, thus avoiding the sort.
1181 MIN/MAX already does this, but not for LIMIT > 1.
1183 * Precompile SQL functions to avoid overhead
1184 * Create utility to compute accurate random_page_cost value
1185 * Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
1186 * Have EXPLAIN ANALYZE highlight poor optimizer estimates
1187 * Consider using hash buckets to do DISTINCT, rather than sorting
1189 This would be beneficial when there are few distinct values. This is
1190 already used by GROUP BY.
1192 * Log statements where the optimizer row estimates were dramatically
1193 different from the number of rows actually found?
1194 * Consider compressed annealing to search for query plans
1196 This might replace GEQO, http://sixdemonbag.org/Djinni.
1199 Miscellaneous Performance
1200 =========================
1202 * Do async I/O for faster random read-ahead of data
1204 Async I/O allows multiple I/O requests to be sent to the disk with
1205 results coming back asynchronously.
1207 * Use mmap() rather than SYSV shared memory or to write WAL files?
1209 This would remove the requirement for SYSV SHM but would introduce
1210 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
1211 to prevent I/O overhead.
1213 * Consider mmap()'ing files into a backend?
1215 Doing I/O to large tables would consume a lot of address space or
1216 require frequent mapping/unmapping. Extending the file also causes
1217 mapping problems that might require mapping only individual pages,
1218 leading to thousands of mappings. Another problem is that there is no
1219 way to _prevent_ I/O to disk from the dirty shared buffers so changes
1220 could hit disk before WAL is written.
1222 * Add a script to ask system configuration questions and tune postgresql.conf
1223 * Merge xmin/xmax/cmin/cmax back into three header fields
1225 Before subtransactions, there used to be only three fields needed to
1226 store these four values. This was possible because only the current
1227 transaction looks at the cmin/cmax values. If the current transaction
1228 created and expired the row the fields stored where xmin (same as
1229 xmax), cmin, cmax, and if the transaction was expiring a row from a
1230 another transaction, the fields stored were xmin (cmin was not
1231 needed), xmax, and cmax. Such a system worked because a transaction
1232 could only see rows from another completed transaction. However,
1233 subtransactions can see rows from outer transactions, and once the
1234 subtransaction completes, the outer transaction continues, requiring
1235 the storage of all four fields. With subtransactions, an outer
1236 transaction can create a row, a subtransaction expire it, and when the
1237 subtransaction completes, the outer transaction still has to have
1238 proper visibility of the row's cmin, for example, for cursors.
1240 One possible solution is to create a phantom cid which represents a
1241 cmin/cmax pair and is stored in local memory. Another idea is to
1242 store both cmin and cmax only in local memory.
1244 * Consider ways of storing rows more compactly on disk
1246 o Support a smaller header for short variable-length fields?
1248 One idea is to create zero-or-one-byte-header versions
1249 of varlena data types. In involves setting the high-bit and
1250 0-127 length in the single-byte header, or clear the high bit
1251 and store the 7-bit ASCII value in the rest of the byte.
1252 The small-header versions have no alignment requirements.
1253 http://archives.postgresql.org/pgsql-hackers/2006-09/msg01372.php
1255 o Reduce the row header size?
1261 * Add use of 'const' for variables in source tree
1262 * Move some things from /contrib into main tree
1263 * Move some /contrib modules out to their own project sites
1265 Particularly, move GPL-licensed /contrib/userlock and
1266 /contrib/dbmirror/clean_pending.pl.
1268 * %Remove warnings created by -Wcast-align
1269 * Move platform-specific ps status display info from ps_status.c to ports
1270 * Add optional CRC checksum to heap and index pages
1271 * Improve documentation to build only interfaces (Marc)
1272 * Remove or relicense modules that are not under the BSD license, if possible
1273 * %Remove memory/file descriptor freeing before ereport(ERROR)
1274 * Acquire lock on a relation before building a relcache entry for it
1275 * %Promote debug_query_string into a server-side function current_query()
1276 * %Allow the identifier length to be increased via a configure option
1277 * Allow cross-compiling by generating the zic database on the target system
1278 * Improve NLS maintenance of libpgport messages linked onto applications
1279 * Allow ecpg to work with MSVC and BCC
1280 * Add xpath_array() to /contrib/xml2 to return results as an array
1281 * Allow building in directories containing spaces
1283 This is probably not possible because 'gmake' and other compiler tools
1284 do not fully support quoting of paths with spaces.
1286 * Fix sgmltools so PDFs can be generated with bookmarks
1287 * Use UTF8 encoding for NLS messages so all server encodings can
1289 * Update Bonjour to work with newer cross-platform SDK
1290 * Split out libpq pgpass and environment documentation sections to make
1291 it easier for non-developers to find
1292 * Consider detoasting keys before sorting
1293 * Consider GnuTLS if OpenSSL license becomes a problem
1295 http://archives.postgresql.org/pgsql-patches/2006-05/msg00040.php
1297 * Use strlcpy() rather than our StrNCpy() macro
1299 http://archives.postgresql.org/pgsql-hackers/2006-09/msg02108.php
1304 o Remove configure.in check for link failure when cause is found
1305 o Remove readdir() errno patch when runtime/mingwex/dirent.c rev
1307 o Remove psql newline patch when we find out why mingw outputs an
1309 o Allow psql to use readline once non-US code pages work with
1311 o Re-enable timezone output on log_line_prefix '%t' when a
1312 shorter timezone string is available
1313 o Fix problem with shared memory on the Win32 Terminal Server
1314 o Improve signal handling
1316 http://archives.postgresql.org/pgsql-patches/2005-06/msg00027.php
1318 o Add long file support for binary pg_dump output
1320 While Win32 supports 64-bit files, the MinGW API does not,
1321 meaning we have to build an fseeko replacement on top of the
1322 Win32 API, and we have to make sure MinGW handles it. Another
1323 option is to wait for the MinGW project to fix it, or use the
1324 code from the LibGW32C project as a guide.
1326 o Check WSACancelBlockingCall() for interrupts [win32intr]
1329 * Wire Protocol Changes
1331 o Allow dynamic character set handling
1332 o Add decoded type, length, precision
1334 o Update clients to use data types, typmod, schema.table.column names
1335 of result sets using new statement protocol
1341 * Add pre-parsing phase that converts non-ISO syntax to supported
1344 This could allow SQL written for other databases to run without
1347 * Allow plug-in modules to emulate features from other databases
1348 * SQL*Net listener that makes PostgreSQL appear as an Oracle database
1350 * Allow statements across databases or servers with transaction
1353 This can be done using dblink and two-phase commit.
1355 * Add the features of packages
1357 o Make private objects accessible only to objects in the same schema
1358 o Allow current_schema.objname to access current schema objects
1359 o Add session variables
1360 o Allow nested schemas
1362 * Consider allowing control of upper/lower case folding of unquoted
1365 http://archives.postgresql.org/pgsql-hackers/2004-04/msg00818.php
1369 Features We Do _Not_ Want
1370 =========================
1372 * All backends running as threads in a single process (not wanted)
1374 This eliminates the process protection we get from the current setup.
1375 Thread creation is usually the same overhead as process creation on
1376 modern systems, so it seems unwise to use a pure threaded model.
1378 * Optimizer hints (not wanted)
1380 Optimizer hints are used to work around problems in the optimizer. We
1381 would rather have the problems reported and fixed.
1382 http://archives.postgresql.org/pgsql-hackers/2006-08/msg00506.php
1383 http://archives.postgresql.org/pgsql-hackers/2006-10/msg00517.php
1384 http://archives.postgresql.org/pgsql-hackers/2006-10/msg00663.php
1386 * Allow AS in "SELECT col AS label" to be optional (not wanted)
1388 Because we support postfix operators, it isn't possible to make AS
1389 optional and continue to use bison.
1390 http://archives.postgresql.org/pgsql-sql/2006-08/msg00164.php
1393 ---------------------------------------------------------------------------
1396 Developers who have claimed items are:
1397 --------------------------------------
1398 * Alvaro is Alvaro Herrera <alvherre@dcc.uchile.cl>
1399 * Andrew is Andrew Dunstan <andrew@dunslane.net>
1400 * Bruce is Bruce Momjian <bruce@momjian.us> of EnterpriseDB
1401 * Christopher is Christopher Kings-Lynne <chriskl@familyhealth.com.au> of
1402 Family Health Network
1403 * D'Arcy is D'Arcy J.M. Cain <darcy@druid.net> of The Cain Gang Ltd.
1404 * David is David Fetter <david@fetter.org>
1405 * Fabien is Fabien Coelho <coelho@cri.ensmp.fr>
1406 * Gavin is Gavin Sherry <swm@linuxworld.com.au> of Alcove Systems Engineering
1407 * Greg is Greg Sabino Mullane <greg@turnstep.com>
1408 * Jan is Jan Wieck <JanWieck@Yahoo.com> of Afilias, Inc.
1409 * Joe is Joe Conway <mail@joeconway.com>
1410 * Karel is Karel Zak <zakkr@zf.jcu.cz>
1411 * Magnus is Magnus Hagander <mha@sollentuna.net>
1412 * Marc is Marc Fournier <scrappy@hub.org> of PostgreSQL, Inc.
1413 * Matthew T. O'Connor <matthew@zeut.net>
1414 * Michael is Michael Meskes <meskes@postgresql.org> of Credativ
1415 * Neil is Neil Conway <neilc@samurai.com>
1416 * Oleg is Oleg Bartunov <oleg@sai.msu.su>
1417 * Pavel is Pavel Stehule <pavel.stehule@hotmail.com>
1418 * Peter is Peter Eisentraut <peter_e@gmx.net>
1419 * Philip is Philip Warner <pjw@rhyme.com.au> of Albatross Consulting Pty. Ltd.
1420 * Rod is Rod Taylor <pg@rbt.ca>
1421 * Simon is Simon Riggs <simon@2ndquadrant.com>
1422 * Stephan is Stephan Szabo <sszabo@megazone23.bigpanda.com>
1423 * Tatsuo is Tatsuo Ishii <ishii@sraoss.co.jp> of SRA OSS, Inc. Japan
1424 * Teodor is Teodor Sigaev <teodor@sigaev.ru>
1425 * Tom is Tom Lane <tgl@sss.pgh.pa.us> of Red Hat