1 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
2 "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
5 <title>PostgreSQL TODO List</title>
6 <meta name="generator" content="HTML::TextToHTML v2.25"/>
8 <body bgcolor="#FFFFFF" text="#000000" link="#FF0000" vlink="#A00000" alink="#0000FF">
9 <h1><a name="section_1">PostgreSQL TODO List</a></h1>
10 <p>Current maintainer: Bruce Momjian (<a href="mailto:pgman@candle.pha.pa.us">pgman@candle.pha.pa.us</a>)<br/>
11 Last updated: Fri Sep 2 17:10:58 EDT 2005
13 <p>The most recent version of this document can be viewed at<br/>
14 <a href="http://www.postgresql.org/docs/faqs.TODO.html">http://www.postgresql.org/docs/faqs.TODO.html</a>.
16 <p><strong>A hyphen, "-", marks changes that will appear in the upcoming 8.1 release.</strong><br/>
17 <strong>A percent sign, "%", marks items that are easier to implement.</strong>
19 <p>Bracketed items, "[<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?"></a>]", have more detail.
21 <p>This list contains all known PostgreSQL bugs and feature requests. If<br/>
22 you would like to work on an item, please read the Developer's FAQ<br/>
25 <h1><a name="section_2">Administration</a></h1>
28 <li>%Remove behavior of postmaster -o after making postmaster/postgres
30 </li><li>%Allow pooled connections to list all prepared queries
31 <p> This would allow an application inheriting a pooled connection to know
32 the queries prepared in the current session.
34 </li><li>Allow major upgrades without dump/reload, perhaps using pg_upgrade
35 [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?pg_upgrade">pg_upgrade</a>]
36 </li><li>Check for unreferenced table files created by transactions that were
37 in-progress when the server terminated abruptly
38 </li><li>Allow administrators to safely terminate individual sessions either
39 via an SQL function or SIGTERM
40 <p> Lock table corruption following SIGTERM of an individual backend
41 has been reported in 8.0. A possible cause was fixed in 8.1, but
42 it is unknown whether other problems exist. This item mostly
43 requires additional testing rather than of writing any new code.
45 </li><li>%Set proper permissions on non-system schemas during db creation
46 <p> Currently all schemas are owned by the super-user because they are
47 copied from the template1 database.
49 </li><li>Support table partitioning that allows a single table to be stored
50 in subtables that are partitioned based on the primary key or a WHERE
52 </li><li>Improve replication solutions
55 <p> You can use any of the master/slave replication servers to use a
56 standby server for data warehousing. To allow read/write queries to
57 multiple servers, you need multi-master replication like pgcluster.
59 </li><li>Allow replication over unreliable or non-persistent links
61 </li><li>Configuration files
63 <li>%Add "include file" functionality in postgresql.conf
64 </li><li>%Allow commenting of variables in postgresql.conf to restore them
66 <p> Currently, if a variable is commented out, it keeps the
67 previous uncommented value until a server restarted.
69 </li><li>%Allow pg_hba.conf settings to be controlled via SQL
70 <p> This would add a function to load the SQL table from
71 pg_hba.conf, and one to writes its contents to the flat file.
72 The table should have a line number that is a float so rows
73 can be inserted between existing rows, e.g. row 2.5 goes
74 between row 2 and row 3.
76 </li><li>%Allow postgresql.conf file values to be changed via an SQL
77 API, perhaps using SET GLOBAL
78 </li><li>Allow the server to be stopped/restarted via an SQL API
79 </li><li>Issue a warning if a change-on-restart-only postgresql.conf value
80 is modified and the server config files are reloaded
81 </li><li>Mark change-on-restart-only values in postgresql.conf
85 <li>Allow a database in tablespace t1 with tables created in
86 tablespace t2 to be used as a template for a new database created
87 with default tablespace t2
88 <p> All objects in the default database tablespace must have default
89 tablespace specifications. This is because new databases are
90 created by copying directories. If you mix default tablespace
91 tables and tablespace-specified tables in the same directory,
92 creating a new database from such a mixed directory would create a
93 new database with tables that had incorrect explicit tablespaces.
94 To fix this would require modifying pg_class in the newly copied
95 database, which we don't currently do.
97 </li><li>Allow reporting of which objects are in which tablespaces
98 <p> This item is difficult because a tablespace can contain objects
99 from multiple databases. There is a server-side function that
100 returns the databases which use a specific tablespace, so this
101 requires a tool that will call that function and connect to each
102 database to find the objects in each database for that tablespace.
105 <li>%Add a GUC variable to control the tablespace for temporary objects
107 <p> It could start with a random tablespace from a supplied list and
108 cycle through the list.
110 </li><li>Allow WAL replay of CREATE TABLESPACE to work when the directory
111 structure on the recovery computer is different from the original
112 </li><li>Allow per-tablespace quotas
115 </li><li>Point-In-Time Recovery (PITR)
117 <li>Allow point-in-time recovery to archive partially filled
118 write-ahead logs [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?pitr">pitr</a>]
119 <p> Currently only full WAL files are archived. This means that the
120 most recent transactions aren't available for recovery in case
121 of a disk failure. This could be triggered by a user command or
124 </li><li>Automatically force archiving of partially-filled WAL files when
125 pg_stop_backup() is called or the server is stopped
126 <p> Doing this will allow administrators to know more easily when
127 the archive contins all the files needed for point-in-time
130 </li><li>%Create dump tool for write-ahead logs for use in determining
131 transaction id for point-in-time recovery
132 </li><li>Allow a warm standby system to also allow read-only queries
133 [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?pitr">pitr</a>]
134 <p> This is useful for checking PITR recovery.
136 </li><li>Allow the PITR process to be debugged and data examined
139 <h1><a name="section_3">Monitoring</a></h1>
142 <li>Allow server log information to be output as INSERT statements
143 <p> This would allow server log information to be easily loaded into
144 a database for analysis.
146 </li><li>%Add ability to monitor the use of temporary sort files
147 </li><li>Allow server logs to be remotely read and removed using SQL commands
149 <h1><a name="section_4">Data Types</a></h1>
152 <li>Improve the MONEY data type
153 <p> Change the MONEY data type to use DECIMAL internally, with special
154 locale-aware output formatting.
156 </li><li>Change NUMERIC to enforce the maximum precision, and increase it
157 </li><li>Add NUMERIC division operator that doesn't round?
158 <p> Currently NUMERIC _rounds_ the result to the specified precision.
159 This means division can return a result that multiplied by the
160 divisor is greater than the dividend, e.g. this returns a value > 10:
163 <p> SELECT (10::numeric(2,0) / 6::numeric(2,0))::numeric(2,0) * 6;
165 <p> The positive modulus result returned by NUMERICs might be considered<br/>
166 inaccurate, in one sense.
169 <li>Have sequence dependency track use of DEFAULT sequences,
171 </li><li>%Disallow changing default expression of a SERIAL column?
172 </li><li>Fix data types where equality comparison isn't intuitive, e.g. box
173 </li><li>%Prevent INET cast to CIDR if the unmasked bits are not zero, or
175 </li><li>%Prevent INET cast to CIDR from droping netmask, SELECT '<a href="telnet://1.1.1.1">1.1.1.1</a>'::inet::cidr
176 </li><li>Allow INET + INT4 to increment the host part of the address, or
177 throw an error on overflow
178 </li><li>%Add 'tid != tid ' operator for use in corruption recovery
179 </li><li>Allow user-defined types to specify a type modifier at table creation
181 </li><li>Dates and Times
183 <li>Allow infinite dates just like infinite timestamps
184 </li><li>Add a GUC variable to allow output of interval values in ISO8601
186 </li><li>Merge hardwired timezone names with the TZ database; allow either
187 kind everywhere a TZ name is currently taken
188 </li><li>Allow customization of the known set of TZ names (generalize the
189 present australian_timezones hack)
190 </li><li>Allow TIMESTAMP WITH TIME ZONE to store the original timezone
191 information, either zone name or offset from UTC [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?timezone">timezone</a>]
192 <p> If the TIMESTAMP value is stored with a time zone name, interval
193 computations should adjust based on the time zone rules.
195 </li><li>Fix SELECT '0.01 years'::interval, '0.01 months'::interval
196 </li><li>Add ISO INTERVAL handling
198 <li>Add support for day-time syntax, INTERVAL '1 2:03:04' DAY TO
200 </li><li>Add support for year-month syntax, INTERVAL '50-6' YEAR TO MONTH
201 </li><li>For syntax that isn't uniquely ISO or PG syntax, like '1:30' or
202 '1', treat as ISO if there is a range specification clause,
203 and as PG if there no clause is present, e.g. interpret
204 <p> '1:30' MINUTE TO SECOND as '1 minute 30 seconds', and
205 interpret '1:30' as '1 hour, 30 minutes'
206 <li>Interpret INTERVAL '1 year' MONTH as CAST (INTERVAL '1 year' AS
207 INTERVAL MONTH), and this should return '12 months'
208 </li><li>Round or truncate values to the requested precision, e.g.
209 INTERVAL '11 months' AS YEAR should return one or zero
210 </li><li>Support precision, CREATE TABLE foo (a INTERVAL MONTH(3))
215 <li>Allow NULLs in arrays
216 </li><li>Delay resolution of array expression's data type so assignment
217 coercion can be performed on empty array expressions
221 <li>Improve vacuum of large objects, like /contrib/vacuumlo?
222 </li><li>Add security checking for large objects
223 </li><li>Auto-delete large objects when referencing row is deleted
224 <p> /contrib/lo offers this functionality.
226 </li><li>Allow read/write into TOAST values like large objects
227 <p> This requires the TOAST column to be stored EXTERNAL.
231 <h1><a name="section_5">Functions</a></h1>
234 <li>Allow INET subnet tests using non-constants to be indexed
235 </li><li>Add transaction_timestamp(), statement_timestamp(), clock_timestamp()
237 <p> Current CURRENT_TIMESTAMP returns the start time of the current
238 transaction, and gettimeofday() returns the wallclock time. This will
239 make time reporting more consistent and will allow reporting of
240 the statement start time.
242 </li><li>%Add pg_get_acldef(), pg_get_typedefault(), and pg_get_attrdef()
243 </li><li>Allow to_char() to print localized month names
244 </li><li>Allow functions to have a schema search path specified at creation time
245 </li><li>Allow substring/replace() to get/set bit values
246 </li><li>Allow to_char() on interval values to accumulate the highest unit
248 <p> Some special format flag would be required to request such
249 accumulation. Such functionality could also be added to EXTRACT.
250 Prevent accumulation that crosses the month/day boundary because of
251 the uneven number of days in a month.
254 <li>to_char(INTERVAL '1 hour 5 minutes', 'MI') => 65
255 </li><li>to_char(INTERVAL '43 hours 20 minutes', 'MI' ) => 2600
256 </li><li>to_char(INTERVAL '43 hours 20 minutes', 'WK:DD:HR:MI') => 0:1:19:20
257 </li><li>to_char(INTERVAL '3 years 5 months','MM') => 41
259 </li><li>Add sleep() function, remove from regress.c
261 <h1><a name="section_6">Multi-Language Support</a></h1>
264 <li>Add NCHAR (as distinguished from ordinary varchar),
265 </li><li>Allow locale to be set at database creation
266 <p> Currently locale can only be set during initdb. No global tables have
267 locale-aware columns. However, the database template used during
268 database creation might have locale-aware indexes. The indexes would
269 need to be reindexed to match the new locale.
271 </li><li>Allow encoding on a per-column basis
272 <p> Right now only one encoding is allowed per database.
274 </li><li>Support multiple simultaneous character sets, per SQL92
275 </li><li>Improve UTF8 combined character handling?
276 </li><li>Add octet_length_server() and octet_length_client()
277 </li><li>Make octet_length_client() the same as octet_length()?
278 </li><li>Fix problems with wrong runtime encoding conversion for NLS message files
280 <h1><a name="section_7">Views / Rules</a></h1>
283 <li>%Automatically create rules on views so they are updateable, per SQL99
284 <p> We can only auto-create rules for simple views. For more complex
285 cases users will still have to write rules.
287 </li><li>Add the functionality for WITH CHECK OPTION clause of CREATE VIEW
288 </li><li>Allow NOTIFY in rules involving conditionals
289 </li><li>Allow VIEW/RULE recompilation when the underlying tables change
290 <p> Another issue is whether underlying table changes should be reflected
291 in the view, e.g. should SELECT * show additional columns if they
292 are added after the view is created.
295 <h1><a name="section_8">SQL Commands</a></h1>
298 <li>Change LIMIT/OFFSET and FETCH/MOVE to use int8
299 </li><li>Add CORRESPONDING BY to UNION/INTERSECT/EXCEPT
300 </li><li>Add ROLLUP, CUBE, GROUPING SETS options to GROUP BY
301 </li><li>%Allow SET CONSTRAINTS to be qualified by schema/table name
302 </li><li>%Allow TRUNCATE ... CASCADE/RESTRICT
303 <p> This is like DELETE CASCADE, but truncates.
305 </li><li>%Add a separate TRUNCATE permission
306 <p> Currently only the owner can TRUNCATE a table because triggers are not
307 called, and the table is locked in exclusive mode.
309 </li><li>Allow PREPARE of cursors
310 </li><li>Allow PREPARE to automatically determine parameter types based on the SQL
312 </li><li>Allow finer control over the caching of prepared query plans
313 <p> Currently, queries prepared via the libpq API are planned on first
314 execute using the supplied parameters --- allow SQL PREPARE to do the
315 same. Also, allow control over replanning prepared queries either
316 manually or automatically when statistics for execute parameters
317 differ dramatically from those used during planning.
319 </li><li>Allow LISTEN/NOTIFY to store info in memory rather than tables?
320 <p> Currently LISTEN/NOTIFY information is stored in pg_listener. Storing
321 such information in memory would improve performance.
323 </li><li>Add optional textual message to NOTIFY
324 <p> This would allow an informational message to be added to the notify
325 message, perhaps indicating the row modified or other custom
328 </li><li>Add a GUC variable to warn about non-standard SQL usage in queries
329 </li><li>Add MERGE command that does UPDATE/DELETE, or on failure, INSERT (rules,
331 </li><li>Add NOVICE output level for helpful messages like automatic sequence/index
333 </li><li>%Add COMMENT ON for all cluster global objects (roles, databases
335 </li><li>%Make row-wise comparisons work per SQL spec
336 </li><li>Add RESET CONNECTION command to reset all session state
337 <p> This would include resetting of all variables (RESET ALL), dropping of
338 temporary tables, removing any NOTIFYs, cursors, open transactions,
339 prepared queries, currval()s, etc. This could be used for connection
340 pooling. We could also change RESET ALL to have this functionality.
341 The difficult of this features is allowing RESET ALL to not affect
342 changes made by the interface driver for its internal use. One idea
343 is for this to be a protocol-only feature. Another approach is to
344 notify the protocol when a RESET CONNECTION command is used.
346 </li><li>Add GUC to issue notice about queries that use unjoined tables
347 </li><li>Allow EXPLAIN to identify tables that were skipped because of
349 </li><li>Allow EXPLAIN output to be more easily processed by scripts
350 </li><li>Eventually enable escape_string<u>warning and standard</u>conforming_strings
353 <li>Allow CREATE TABLE AS to determine column lengths for complex
354 expressions like SELECT col1 || col2
355 </li><li>Use more reliable method for CREATE DATABASE to get a consistent
357 </li><li>Add ON COMMIT capability to CREATE TABLE AS ... SELECT
361 <li>Allow UPDATE to handle complex aggregates [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?update">update</a>]?
362 </li><li>Allow an alias to be provided for the target table in
364 <p> This is not SQL-spec but many DBMSs allow it.
366 </li><li>Allow UPDATE tab SET ROW (col, ...) = (...) for updating multiple
371 <li>%Have ALTER TABLE RENAME rename SERIAL sequence names
372 </li><li>Add ALTER DOMAIN to modify the underlying data type
373 </li><li>%Allow ALTER TABLE ... ALTER CONSTRAINT ... RENAME
374 </li><li>%Allow ALTER TABLE to change constraint deferrability and actions
375 </li><li>Add missing object types for ALTER ... SET SCHEMA
376 </li><li>Allow ALTER TABLESPACE to move to different directories
377 </li><li>Allow databases to be moved to different tablespaces
378 </li><li>Allow moving system tables to other tablespaces, where possible
379 <p> Currently non-global system tables must be in the default database
380 tablespace. Global system tables can never be moved.
382 </li><li>%Disallow dropping of an inherited constraint
383 </li><li>%Prevent child tables from altering or dropping constraints
384 like CHECK that were inherited from the parent table
388 <li>Automatically maintain clustering on a table
389 <p> This might require some background daemon to maintain clustering
390 during periods of low usage. It might also require tables to be only
391 paritally filled for easier reorganization. Another idea would
392 be to create a merged heap/index data file so an index lookup would
393 automatically access the heap data too. A third idea would be to
394 store heap rows in hashed groups, perhaps using a user-supplied
397 </li><li>%Add default clustering to system tables
398 <p> To do this, determine the ideal cluster index for each system
399 table and set the cluster setting during initdb.
404 <li>Allow COPY to report error lines and continue
405 <p> This requires the use of a savepoint before each COPY line is
406 processed, with ROLLBACK on COPY failure.
408 </li><li>%Have COPY return the number of rows loaded/unloaded?
409 </li><li>Allow COPY on a newly-created table to skip WAL logging
410 <p> On crash recovery, the table involved in the COPY would
411 have its heap and index files truncated. One issue is
412 that no other backend should be able to add to the table
413 at the same time, which is something that is currently
417 </li><li>GRANT/REVOKE
419 <li>Allow column-level privileges
420 </li><li>%Allow GRANT/REVOKE permissions to be applied to all schema objects
422 <p> The proposed syntax is:
423 </p><p> GRANT SELECT ON ALL TABLES IN public TO phpuser;
424 GRANT SELECT ON NEW TABLES IN public TO phpuser;
427 <li>Allow GRANT/REVOKE permissions to be inherited by objects based on
433 <li>Allow UPDATE/DELETE WHERE CURRENT OF cursor
434 <p> This requires using the row ctid to map cursor rows back to the
435 original heap row. This become more complicated if WITH HOLD cursors
436 are to be supported because WITH HOLD cursors have a copy of the row
437 and no FOR UPDATE lock.
439 </li><li>Prevent DROP TABLE from dropping a row referenced by its own open
441 </li><li>%Allow pooled connections to list all open WITH HOLD cursors
442 <p> Because WITH HOLD cursors exist outside transactions, this allows
443 them to be listed so they can be closed.
448 <li>Allow INSERT/UPDATE of the system-generated oid value for a row
449 </li><li>Allow INSERT INTO tab (col1, ..) VALUES (val1, ..), (val2, ..)
450 </li><li>Allow INSERT/UPDATE ... RETURNING new.col or old.col
451 <p> This is useful for returning the auto-generated key for an INSERT.
452 One complication is how to handle rules that run as part of
458 <li>Add SET PERFORMANCE_TIPS option to suggest INDEX, VACUUM, VACUUM
460 </li><li>Add SET PATH for schemas?
461 <p> This is basically the same as SET search_path.
464 </li><li>Server-Side Languages
466 <li>Fix PL/pgSQL RENAME to work on variables other than OLD/NEW
467 </li><li>Allow function parameters to be passed by name,
468 get_employee_salary(emp_id => 12345, tax_year => 2001)
469 </li><li>Add Oracle-style packages
470 </li><li>Add table function support to pltcl, plpython
471 </li><li>Add capability to create and call PROCEDURES
472 </li><li>Allow PL/pgSQL to handle %TYPE arrays, e.g. tab.col%TYPE[<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?"></a>]
473 </li><li>Allow function argument names to be queries from PL/PgSQL
474 </li><li>Add MOVE to PL/pgSQL
475 </li><li>Add support for polymorphic arguments and return types to
476 languages other than PL/PgSQL
477 </li><li>Add support for OUT and INOUT parameters to languages other
481 <h1><a name="section_9">Clients</a></h1>
484 <li>Add a libpq function to support Parse/DescribeStatement capability
485 </li><li>Prevent libpq's PQfnumber() from lowercasing the column name?
486 </li><li>Add PQescapeIdentifier() to libpq
487 </li><li>Have initdb set the input DateStyle (MDY or DMY) based on locale?
488 </li><li>Have pg_ctl look at PGHOST in case it is a socket directory?
489 </li><li>Allow pg_ctl to work properly with configuration files located outside
491 <p> pg_ctl can not read the pid file because it isn't located in the
492 config directory but in the PGDATA directory. The solution is to
493 allow pg_ctl to read and understand postgresql.conf to find the
494 data_directory value.
498 <li>Have psql show current values for a sequence
499 </li><li>Move psql backslash database information into the backend, use
500 mnemonic commands? [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?psql">psql</a>]
501 <p> This would allow non-psql clients to pull the same information out
502 of the database as psql.
504 </li><li>Fix psql's display of schema information (Neil)
505 </li><li>Allow psql \pset boolean variables to set to fixed values, rather
507 </li><li>Consistently display privilege information for all objects in psql
508 </li><li>Improve psql's handling of multi-line queries
509 <p> Currently, while \e saves a single query as one entry, interactive
510 queries are saved one line at a time. Ideally all queries
511 whould be saved like \e does.
513 </li><li>Allow multi-line column values to align in the proper columns
514 <p> If the second output column value is 'a\nb', the 'b' should appear
515 in the second display column, rather than the first column as it
521 <li>%Have pg_dump use multi-statement transactions for INSERT dumps
522 </li><li>%Allow pg_dump to use multiple -t and -n switches [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?pg_dump">pg_dump</a>]
523 </li><li>%Add dumping of comments on composite type columns
524 </li><li>%Add dumping of comments on index columns
525 </li><li>%Replace crude DELETE FROM method of pg_dumpall --clean for
526 cleaning of roles with separate DROP commands
527 </li><li>Stop dumping CASCADE on DROP TYPE commands in clean mode
528 </li><li>%Add full object name to the tag field. eg. for operators we need
529 '=(integer, integer)', instead of just '='.
530 </li><li>Add pg_dumpall custom format dumps?
531 </li><li>%Add CSV output format
532 </li><li>Update pg_dump and psql to use the new COPY libpq API (Christopher)
533 </li><li>Remove unnecessary function pointer abstractions in pg_dump source
539 <p> Document differences between ecpg and the SQL standard and
540 information about the Informix-compatibility module.
542 </li><li>Solve cardinality > 1 for input descriptors / variables?
543 </li><li>Add a semantic check level, e.g. check if a table really exists
544 </li><li>fix handling of DB attributes that are arrays
545 </li><li>Use backend PREPARE/EXECUTE facility for ecpg where possible
546 </li><li>Implement SQLDA
547 </li><li>Fix nested C comments
548 </li><li>%sqlwarn[<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?6">6</a>] should be 'W' if the PRECISION or SCALE value specified
549 </li><li>Make SET CONNECTION thread-aware, non-standard?
550 </li><li>Allow multidimensional arrays
551 </li><li>Add internationalized message strings
554 <h1><a name="section_10">Referential Integrity</a></h1>
557 <li>Add MATCH PARTIAL referential integrity
558 </li><li>Add deferred trigger queue file
559 <p> Right now all deferred trigger information is stored in backend
560 memory. This could exhaust memory for very large trigger queues.
561 This item involves dumping large queues into files.
563 </li><li>Change foreign key constraint for array -> element to mean element
565 </li><li>Allow DEFERRABLE UNIQUE constraints?
566 </li><li>Allow triggers to be disabled in only the current session.
567 <p> This is currently possible by starting a multi-statement transaction,
568 modifying the system tables, performing the desired SQL, restoring the
569 system tables, and committing the transaction. ALTER TABLE ...
570 TRIGGER requires a table lock so it is not ideal for this usage.
572 </li><li>With disabled triggers, allow pg_dump to use ALTER TABLE ADD FOREIGN KEY
573 <p> If the dump is known to be valid, allow foreign keys to be added
574 without revalidating the data.
576 </li><li>Allow statement-level triggers to access modified rows
577 </li><li>Support triggers on columns (Greg Sabino Mullane)
578 </li><li>Enforce referential integrity for system tables
579 </li><li>Allow AFTER triggers on system tables
580 <p> System tables are modified in many places in the backend without going
581 through the executor and therefore not causing triggers to fire. To
582 complete this item, the functions that modify system tables will have
586 <h1><a name="section_11">Dependency Checking</a></h1>
589 <li>Flush cached query plans when the dependent objects change
590 </li><li>Track dependencies in function bodies and recompile/invalidate
591 <p> This is particularly important for references to temporary tables
592 in PL/PgSQL because PL/PgSQL caches query plans. The only workaround
593 in PL/PgSQL is to use EXECUTE. One complexity is that a function
594 might itself drop and recreate dependent tables, causing it to
595 invalidate its own query plan.
598 <h1><a name="section_12">Exotic Features</a></h1>
601 <li>Add SQL99 WITH clause to SELECT
602 </li><li>Add SQL99 WITH RECURSIVE to SELECT
603 </li><li>Add pre-parsing phase that converts non-ISO syntax to supported
605 <p> This could allow SQL written for other databases to run without
608 </li><li>Allow plug-in modules to emulate features from other databases
609 </li><li>SQL*Net listener that makes PostgreSQL appear as an Oracle database
611 </li><li>Allow queries across databases or servers with transaction
613 <p> This can be done using dblink and two-phase commit.
615 </li><li>Add the features of packages
617 <li>Make private objects accessable only to objects in the same schema
618 </li><li>Allow current_schema.objname to access current schema objects
619 </li><li>Add session variables
620 </li><li>Allow nested schemas
623 <h1><a name="section_13">Indexes</a></h1>
626 <li>Allow inherited tables to inherit index, UNIQUE constraint, and primary
628 </li><li>UNIQUE INDEX on base column not honored on INSERTs/UPDATEs from
629 inherited table: INSERT INTO inherit_table (unique_index_col) VALUES
631 <p> The main difficulty with this item is the problem of creating an index
632 that can span more than one table.
634 </li><li>Allow SELECT ... FOR UPDATE on inherited tables
635 </li><li>Add UNIQUE capability to non-btree indexes
636 </li><li>Prevent index uniqueness checks when UPDATE does not modify the column
637 <p> Uniqueness (index) checks are done when updating a column even if the
638 column is not modified by the UPDATE.
640 </li><li>Allow the creation of on-disk bitmap indexes which can be quickly
641 combined with other bitmap indexes
642 <p> Such indexes could be more compact if there are only a few distinct values.
643 Such indexes can also be compressed. Keeping such indexes updated can be
646 </li><li>Allow use of indexes to search for NULLs
647 <p> One solution is to create a partial index on an IS NULL expression.
649 </li><li>Allow accurate statistics to be collected on indexes with more than
650 one column or expression indexes, perhaps using per-index statistics
651 </li><li>Add fillfactor to control reserved free space during index creation
652 </li><li>Allow the creation of indexes with mixed ascending/descending specifiers
653 </li><li>Allow constraint_exclusion to work for UNIONs like it does for
654 inheritance, allow it to work for UPDATE and DELETE queries, and allow
655 it to be used for all queries with little performance impact
658 <li>Add more GIST index support for geometric data types
659 </li><li>Allow GIST indexes to create certain complex index types, like
660 digital trees (see Aoki)
664 <li>Pack hash index buckets onto disk pages more efficiently
665 <p> Currently only one hash bucket can be stored on a page. Ideally
666 several hash buckets could be stored on a single page and greater
667 granularity used for the hash algorithm.
669 </li><li>Consider sorting hash buckets so entries can be found using a
670 binary search, rather than a linear scan
671 </li><li>In hash indexes, consider storing the hash value with or instead
673 </li><li>Add WAL logging for crash recovery
674 </li><li>Allow multi-column hash indexes
677 <h1><a name="section_14">Fsync</a></h1>
680 <li>Improve commit_delay handling to reduce fsync()
681 </li><li>Determine optimal fdatasync/fsync, O_SYNC/O_DSYNC options
682 <p> Ideally this requires a separate test program that can be run
683 at initdb time or optionally later.
685 </li><li>%Add an option to sync() before fsync()'ing checkpoint files
686 </li><li>Add program to test if fsync has a delay compared to non-fsync
688 <h1><a name="section_15">Cache Usage</a></h1>
691 <li>Allow free-behind capability for large sequential scans, perhaps using
693 <p> Posix_fadvise() can control both sequential/random file caching and
694 free-behind behavior, but it is unclear how the setting affects other
695 backends that also have the file open, and the feature is not supported
696 on all operating systems.
698 </li><li>Speed up COUNT(*)
699 <p> We could use a fixed row count and a +/- count to follow MVCC
700 visibility rules, or a single cached value could be used and
701 invalidated if anyone modifies the table. Another idea is to
702 get a count directly from a unique index, but for this to be
703 faster than a sequential scan it must avoid access to the heap
704 to obtain tuple visibility information.
706 </li><li>Allow data to be pulled directly from indexes
707 <p> Currently indexes do not have enough tuple visibility information
708 to allow data to be pulled from the index without also accessing
709 the heap. One way to allow this is to set a bit to index tuples
710 to indicate if a tuple is currently visible to all transactions
711 when the first valid heap lookup happens. This bit would have to
712 be cleared when a heap tuple is expired.
714 </li><li>Consider automatic caching of queries at various levels:
716 <li>Parsed query tree
717 </li><li>Query execute plan
718 </li><li>Query results
720 </li><li>Allow sequential scans to take advantage of other concurrent
721 sequentiqal scans, also called "Synchronised Scanning"
722 <p> One possible implementation is to start sequential scans from the lowest
723 numbered buffer in the shared cache, and when reaching the end wrap
724 around to the beginning, rather than always starting sequential scans
725 at the start of the table.
728 <h1><a name="section_16">Vacuum</a></h1>
731 <li>Improve speed with indexes
732 <p> For large table adjustements during VACUUM FULL, it is faster to
733 reindex rather than update the index.
735 </li><li>Reduce lock time during VACUUM FULL by moving tuples with read lock,
736 then write lock and truncate table
737 <p> Moved tuples are invisible to other backends so they don't require a
738 write lock. However, the read lock promotion to write lock could lead
739 to deadlock situations.
741 </li><li>Auto-fill the free space map by scanning the buffer cache or by
742 checking pages written by the background writer
743 </li><li>Create a bitmap of pages that need vacuuming
744 <p> Instead of sequentially scanning the entire table, have the background
745 writer or some other process record pages that have expired rows, then
746 VACUUM can look at just those pages rather than the entire table. In
747 the event of a system crash, the bitmap would probably be invalidated.
748 One complexity is that index entries still have to be vacuumed, and
749 doing this without an index scan (by using the heap values to find the
750 index entry) might be slow and unreliable, especially for user-defined
753 </li><li>%Add system view to show free space map contents
756 <li>Use free-space map information to guide refilling
757 </li><li>%Issue log message to suggest VACUUM FULL if a table is nearly
759 </li><li>Improve xid wraparound detection by recording per-table rather
763 <h1><a name="section_17">Locking</a></h1>
766 <li>Add code to detect an SMP machine and handle spinlocks accordingly
767 from distributted.net, <a href="http://www1.distributed.net/source">http://www1.distributed.net/source</a>,
768 in client/common/cpucheck.cpp
769 <p> On SMP machines, it is possible that locks might be released shortly,
770 while on non-SMP machines, the backend should sleep so the process
771 holding the lock can complete and release it.
773 </li><li>Research use of sched_yield() for spinlock acquisition failure
774 </li><li>Fix priority ordering of read and write light-weight locks (Neil)
776 <h1><a name="section_18">Startup Time Improvements</a></h1>
779 <li>Experiment with multi-threaded backend [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?thread">thread</a>]
780 <p> This would prevent the overhead associated with process creation. Most
781 operating systems have trivial process creation time compared to
782 database startup overhead, but a few operating systems (WIn32,
783 Solaris) might benefit from threading. Also explore the idea of
784 a single session using multiple threads to execute a query faster.
786 </li><li>Add connection pooling
787 <p> It is unclear if this should be done inside the backend code or done
788 by something external like pgpool. The passing of file descriptors to
789 existing backends is one of the difficulties with a backend approach.
792 <h1><a name="section_19">Write-Ahead Log</a></h1>
795 <li>Eliminate need to write full pages to WAL before page modification [<a href="http://momjian.postgresql.org/cgi-bin/pgtodo?wal">wal</a>]
796 <p> Currently, to protect against partial disk page writes, we write
797 full page images to WAL before they are modified so we can correct any
798 partial page writes during recovery. These pages can also be
799 eliminated from point-in-time archive files.
802 <li>When off, write CRC to WAL and check file system blocks
804 <p> If CRC check fails during recovery, remember the page in case
805 a later CRC for that page properly matches.
807 </li><li>Write full pages during file system write and not when
808 the page is modified in the buffer cache
809 <p> This allows most full page writes to happen in the background
810 writer. It might cause problems for applying WAL on recovery
811 into a partially-written page, but later the full page will be
815 </li><li>Reduce WAL traffic so only modified values are written rather than
817 </li><li>Allow the pg_xlog directory location to be specified during initdb
818 with a symlink back to the /data location
819 </li><li>Allow WAL information to recover corrupted pg_controldata
820 </li><li>Find a way to reduce rotational delay when repeatedly writing
822 <p> Currently fsync of WAL requires the disk platter to perform a full
823 rotation to fsync again. One idea is to write the WAL to different
824 offsets that might reduce the rotational delay.
826 </li><li>Allow buffered WAL writes and fsync
827 <p> Instead of guaranteeing recovery of all committed transactions, this
828 would provide improved performance by delaying WAL writes and fsync
829 so an abrupt operating system restart might lose a few seconds of
830 committed transactions but still be consistent. We could perhaps
831 remove the 'fsync' parameter (which results in an an inconsistent
832 database) in favor of this capability.
835 <h1><a name="section_20">Optimizer / Executor</a></h1>
838 <li>Add missing optimizer selectivities for date, r-tree, etc
839 </li><li>Allow ORDER BY ... LIMIT # to select high/low value without sort or
840 index using a sequential scan for highest/lowest values
841 <p> Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
842 all values to return the high/low value. Instead The idea is to do a
843 sequential scan to find the high/low value, thus avoiding the sort.
844 MIN/MAX already does this, but not for LIMIT > 1.
846 </li><li>Precompile SQL functions to avoid overhead
847 </li><li>Create utility to compute accurate random_page_cost value
848 </li><li>Improve ability to display optimizer analysis using OPTIMIZER_DEBUG
849 </li><li>Have EXPLAIN ANALYZE highlight poor optimizer estimates
850 </li><li>Consider using hash buckets to do DISTINCT, rather than sorting
851 <p> This would be beneficial when there are few distinct values.
853 </li><li>Log queries where the optimizer row estimates were dramatically
854 different from the number of rows actually found?
856 <h1><a name="section_21">Miscellaneous Performance</a></h1>
859 <li>Do async I/O for faster random read-ahead of data
860 <p> Async I/O allows multiple I/O requests to be sent to the disk with
861 results coming back asynchronously.
863 </li><li>Use mmap() rather than SYSV shared memory or to write WAL files?
864 <p> This would remove the requirement for SYSV SHM but would introduce
865 portability issues. Anonymous mmap (or mmap to /dev/zero) is required
866 to prevent I/O overhead.
868 </li><li>Consider mmap()'ing files into a backend?
869 <p> Doing I/O to large tables would consume a lot of address space or
870 require frequent mapping/unmapping. Extending the file also causes
871 mapping problems that might require mapping only individual pages,
872 leading to thousands of mappings. Another problem is that there is no
873 way to _prevent_ I/O to disk from the dirty shared buffers so changes
874 could hit disk before WAL is written.
876 </li><li>Add a script to ask system configuration questions and tune postgresql.conf
877 </li><li>Merge xmin/xmax/cmin/cmax back into three header fields
878 <p> Before subtransactions, there used to be only three fields needed to
879 store these four values. This was possible because only the current
880 transaction looks at the cmin/cmax values. If the current transaction
881 created and expired the row the fields stored where xmin (same as
882 xmax), cmin, cmax, and if the transaction was expiring a row from a
883 another transaction, the fields stored were xmin (cmin was not
884 needed), xmax, and cmax. Such a system worked because a transaction
885 could only see rows from another completed transaction. However,
886 subtransactions can see rows from outer transactions, and once the
887 subtransaction completes, the outer transaction continues, requiring
888 the storage of all four fields. With subtransactions, an outer
889 transaction can create a row, a subtransaction expire it, and when the
890 subtransaction completes, the outer transaction still has to have
891 proper visibility of the row's cmin, for example, for cursors.
893 <p> One possible solution is to create a phantom cid which represents a
894 cmin/cmax pair and is stored in local memory.
896 </li><li>Use a phantom command counter for nested subtransactions to reduce
898 </li><li>Research storing disk pages with no alignment/padding
900 <h1><a name="section_22">Source Code</a></h1>
903 <li>Add use of 'const' for variables in source tree
904 </li><li>Rename some /contrib modules from pg* to pg_*
905 </li><li>Move some things from /contrib into main tree
906 </li><li>Move some /contrib modules out to their own project sites
907 </li><li>%Remove warnings created by -Wcast-align
908 </li><li>Move platform-specific ps status display info from ps_status.c to ports
909 </li><li>Add optional CRC checksum to heap and index pages
910 </li><li>Improve documentation to build only interfaces (Marc)
911 </li><li>Remove or relicense modules that are not under the BSD license, if possible
912 </li><li>%Remove memory/file descriptor freeing before ereport(ERROR)
913 </li><li>Acquire lock on a relation before building a relcache entry for it
914 </li><li>%Promote debug_query_string into a server-side function current_query()
915 </li><li>%Allow the identifier length to be increased via a configure option
916 </li><li>Remove Win32 rename/unlink looping if unnecessary
917 </li><li>Allow cross-compiling by generating the zic database on the target system
918 </li><li>Improve NLS maintenace of libpgport messages linked onto applications
919 </li><li>Allow ecpg to work with MSVC and BCC
920 </li><li>Add xpath_array() to /contrib/xml2 to return results as an array
921 </li><li>Allow building in directories containing spaces
922 <p> This is probably not possible because 'gmake' and other compiler tools
923 do not fully support quoting of paths with spaces.
925 </li><li>Allow installing to directories containing spaces
926 <p> This is possible if proper quoting is added to the makefiles for the
927 install targets. Because PostgreSQL supports relocatable installs, it
928 is already possible to install into a directory that doesn't contain
929 spaces and then copy the install to a directory with spaces.
931 </li><li>Fix sgmltools so PDFs can be generated with bookmarks
932 </li><li>%Clean up compiler warnings (especially with gcc version 4)
933 </li><li>Add function to return the thread safety status of libpq and ecpg
934 </li><li>Use UTF8 encoding for NLS messages so all server encodings can
938 <li>Remove configure.in check for link failure when cause is found
939 </li><li>Remove readdir() errno patch when runtime/mingwex/dirent.c rev
941 </li><li>Remove psql newline patch when we find out why mingw outputs an
943 </li><li>Allow psql to use readline once non-US code pages work with
945 </li><li>Re-enable timezone output on log_line_prefix '%t' when a
946 shorter timezone string is available
947 </li><li>Fix problem with shared memory on the Win32 Terminal Server
949 </li><li>Wire Protocol Changes
951 <li>Allow dynamic character set handling
952 </li><li>Add decoded type, length, precision
953 </li><li>Use compression?
954 </li><li>Update clients to use data types, typmod, <a href="http://schema.table.column">schema.table.column</a>/ names
955 of result sets using new query protocol
960 <h2><a name="section_22_1">Developers who have claimed items are:</a></h2>
962 <li>Alvaro is Alvaro Herrera <<a href="mailto:alvherre@dcc.uchile.cl">alvherre@dcc.uchile.cl</a>>
963 </li><li>Andrew is Andrew Dunstan <<a href="mailto:andrew@dunslane.net">andrew@dunslane.net</a>>
964 </li><li>Bruce is Bruce Momjian <<a href="mailto:pgman@candle.pha.pa.us">pgman@candle.pha.pa.us</a>> of Software Research Assoc.
965 </li><li>Christopher is Christopher Kings-Lynne <<a href="mailto:chriskl@familyhealth.com.au">chriskl@familyhealth.com.au</a>> of
966 Family Health Network
967 </li><li>Claudio is Claudio Natoli <<a href="mailto:claudio.natoli@memetrics.com">claudio.natoli@memetrics.com</a>>
968 </li><li>D'Arcy is D'Arcy J.M. Cain <<a href="mailto:darcy@druid.net">darcy@druid.net</a>> of The Cain Gang Ltd.
969 </li><li>Fabien is Fabien Coelho <<a href="mailto:coelho@cri.ensmp.fr">coelho@cri.ensmp.fr</a>>
970 </li><li>Gavin is Gavin Sherry <<a href="mailto:swm@linuxworld.com.au">swm@linuxworld.com.au</a>> of Alcove Systems Engineering
971 </li><li>Greg is Greg Sabino Mullane <<a href="mailto:greg@turnstep.com">greg@turnstep.com</a>>
972 </li><li>Hiroshi is Hiroshi Inoue <<a href="mailto:Inoue@tpf.co.jp">Inoue@tpf.co.jp</a>>
973 </li><li>Jan is Jan Wieck <<a href="mailto:JanWieck@Yahoo.com">JanWieck@Yahoo.com</a>> of Afilias, Inc.
974 </li><li>Joe is Joe Conway <<a href="mailto:mail@joeconway.com">mail@joeconway.com</a>>
975 </li><li>Karel is Karel Zak <<a href="mailto:zakkr@zf.jcu.cz">zakkr@zf.jcu.cz</a>>
976 </li><li>Magnus is Magnus Hagander <<a href="mailto:mha@sollentuna.net">mha@sollentuna.net</a>>
977 </li><li>Marc is Marc Fournier <<a href="mailto:scrappy@hub.org">scrappy@hub.org</a>> of PostgreSQL, Inc.
978 </li><li>Matthew T. O'Connor <<a href="mailto:matthew@zeut.net">matthew@zeut.net</a>>
979 </li><li>Michael is Michael Meskes <<a href="mailto:meskes@postgresql.org">meskes@postgresql.org</a>> of Credativ
980 </li><li>Neil is Neil Conway <<a href="mailto:neilc@samurai.com">neilc@samurai.com</a>>
981 </li><li>Oleg is Oleg Bartunov <<a href="mailto:oleg@sai.msu.su">oleg@sai.msu.su</a>>
982 </li><li>Peter is Peter Eisentraut <<a href="mailto:peter_e@gmx.net">peter_e@gmx.net</a>>
983 </li><li>Philip is Philip Warner <<a href="mailto:pjw@rhyme.com.au">pjw@rhyme.com.au</a>> of Albatross Consulting Pty. Ltd.
984 </li><li>Rod is Rod Taylor <<a href="mailto:pg@rbt.ca">pg@rbt.ca</a>>
985 </li><li>Simon is Simon Riggs <<a href="mailto:simon@2ndquadrant.com">simon@2ndquadrant.com</a>>
986 </li><li>Stephan is Stephan Szabo <<a href="mailto:sszabo@megazone23.bigpanda.com">sszabo@megazone23.bigpanda.com</a>>
987 </li><li>Tatsuo is Tatsuo Ishii <<a href="mailto:t-ishii@sra.co.jp">t-ishii@sra.co.jp</a>> of Software Research Assoc.
988 </li><li>Tom is Tom Lane <<a href="mailto:tgl@sss.pgh.pa.us">tgl@sss.pgh.pa.us</a>> of Red Hat