]> granicus.if.org Git - postgresql/log
postgresql
22 years agoSpacing improvement.
Bruce Momjian [Wed, 14 Aug 2002 03:01:43 +0000 (03:01 +0000)]
Spacing improvement.

22 years agoFix spacing.
Bruce Momjian [Wed, 14 Aug 2002 02:51:13 +0000 (02:51 +0000)]
Fix spacing.

22 years agoAdded:
Bruce Momjian [Wed, 14 Aug 2002 02:49:42 +0000 (02:49 +0000)]
Added:

<
>  o Improve PL/PgSQL exception handling
>  o Allow PL/PgSQL parameters to be specified by name and type during
>    definition
>  o Allow PL/PgSQL function parameters to be passed by name,
>    get_employee_salary(emp_id => 12345, tax_year => 2001)
>  o Add PL/PgSQL packages
>  o Allow array declarations and other data types in PL/PgSQl DECLARE
>  o Add PL/PgSQL PROCEDURES that can return multiple values

22 years agoAdd Cyrillic and other encodings for encoding conversion.
Tatsuo Ishii [Wed, 14 Aug 2002 02:45:10 +0000 (02:45 +0000)]
Add Cyrillic and other encodings for encoding conversion.
Patches submitted by Kaori Inaba (i-kaori@sra.co.jp).

22 years agoRemove namedatalen from TODO.detail. No longer needed.
Bruce Momjian [Wed, 14 Aug 2002 02:28:30 +0000 (02:28 +0000)]
Remove namedatalen from TODO.detail.  No longer needed.

22 years agoDone:
Bruce Momjian [Wed, 14 Aug 2002 02:24:47 +0000 (02:24 +0000)]
Done:

> * -Increase identifier length (NAMEDATALEN) if small performance hit,
> * -Increase maximum number of function parameters if little wasted space

22 years agoPut back old code.
Bruce Momjian [Tue, 13 Aug 2002 21:04:36 +0000 (21:04 +0000)]
Put back old code.

22 years agoUpdate for longer NAMEDATALEN.
Bruce Momjian [Tue, 13 Aug 2002 20:41:13 +0000 (20:41 +0000)]
Update for longer NAMEDATALEN.

22 years agoChange NAMEDATALEN to 64, INDEX_MAX_KEYS/MAX_FUNC_ARGS to 32, per discussion on...
Bruce Momjian [Tue, 13 Aug 2002 20:40:44 +0000 (20:40 +0000)]
Change NAMEDATALEN to 64,  INDEX_MAX_KEYS/MAX_FUNC_ARGS to 32, per discussion on hackers.

22 years agoFix tuple-chain-moving tests to handle marked-for-update tuples correctly
Tom Lane [Tue, 13 Aug 2002 20:14:24 +0000 (20:14 +0000)]
Fix tuple-chain-moving tests to handle marked-for-update tuples correctly
(they are not part of a chain).  When failing to find a parent tuple in
an update chain, emit a warning and abandon repair_frag, but do not give
an error as before.  This should eliminate the infamous 'No one parent tuple
was found' failure, which we now realize is not a can't-happen condition
but a perfectly valid database state.  Per recent pghackers discussion.

22 years agoMake sure that t_ctid is reset to equal t_self in heap_delete and
Tom Lane [Tue, 13 Aug 2002 20:11:03 +0000 (20:11 +0000)]
Make sure that t_ctid is reset to equal t_self in heap_delete and
heap_mark4update; this avoids situations where a deleted tuple might
look like it is chained to something else.  Also, cause all the WAL
redo routines to set t_ctid to equal t_self, rather than leaving it
undefined as before.  Make heap_xlog_clean set the page's LSN and SUI
correctly.  All per past discussions in pghackers, ranging back to
last December.

22 years agoRemove obsolete uses of lanispl. Only used in pg_dump now, but can be
Peter Eisentraut [Tue, 13 Aug 2002 17:22:08 +0000 (17:22 +0000)]
Remove obsolete uses of lanispl.  Only used in pg_dump now, but can be
removed altogether if pg_dump doesn't need it anymore.

22 years agoAdd bitmap index mention.
Bruce Momjian [Tue, 13 Aug 2002 05:08:35 +0000 (05:08 +0000)]
Add bitmap index mention.

22 years agoUpdate:
Bruce Momjian [Tue, 13 Aug 2002 04:32:28 +0000 (04:32 +0000)]
Update:

> * Use bitmaps to fetch heap pages in sequential order [performance]
> * Use bitmaps to combine existing indexes [performance]

22 years agoCorrect description of sequence privileges.
Peter Eisentraut [Mon, 12 Aug 2002 20:02:09 +0000 (20:02 +0000)]
Correct description of sequence privileges.

22 years agoRemove use of __P so that <sys/cdefs.h> is not needed. Per suggestion
Tom Lane [Mon, 12 Aug 2002 15:24:07 +0000 (15:24 +0000)]
Remove use of __P so that <sys/cdefs.h> is not needed.  Per suggestion
from Martin Renters.

22 years agoConvert possibly-unsafe sprintf's to snprintf's.
Tom Lane [Mon, 12 Aug 2002 14:47:46 +0000 (14:47 +0000)]
Convert possibly-unsafe sprintf's to snprintf's.

22 years agoBuffer-overrun paranoia.
Tom Lane [Mon, 12 Aug 2002 14:35:26 +0000 (14:35 +0000)]
Buffer-overrun paranoia.

22 years agoRestructure rowtype-parameter handling to eliminate need for possibly-
Tom Lane [Mon, 12 Aug 2002 14:25:07 +0000 (14:25 +0000)]
Restructure rowtype-parameter handling to eliminate need for possibly-
overflowable buffer for 'name%rowtype'; not to mention avoid problems
with mixed-case type names and other special cases.

22 years agoChange memory-space accounting mechanism in tuplesort.c and tuplestore.c
Tom Lane [Mon, 12 Aug 2002 00:36:12 +0000 (00:36 +0000)]
Change memory-space accounting mechanism in tuplesort.c and tuplestore.c
to make a reasonable attempt at accounting for palloc overhead, not just
the requested size of each memory chunk.  Since in many scenarios this
will make for a significant reduction in the amount of space acquired,
partially compensate by doubling the default value of SORT_MEM to 1Mb.
Per discussion in pgsql-general around 9-Jun-2002..

22 years agoCode review of CLUSTER patch. Clean up problems with relcache getting
Tom Lane [Sun, 11 Aug 2002 21:17:35 +0000 (21:17 +0000)]
Code review of CLUSTER patch.  Clean up problems with relcache getting
confused, toasted data getting lost, etc.

22 years agoCreate/drop cast now requires ownership of at least one of the types.
Peter Eisentraut [Sun, 11 Aug 2002 17:44:12 +0000 (17:44 +0000)]
Create/drop cast now requires ownership of at least one of the types.

22 years agoEditorial improvements.
Tom Lane [Sun, 11 Aug 2002 02:43:57 +0000 (02:43 +0000)]
Editorial improvements.

22 years agoMake cluster regress test functional.
Tom Lane [Sun, 11 Aug 2002 02:06:32 +0000 (02:06 +0000)]
Make cluster regress test functional.

22 years agoNeil's patch was supposed to remove this, too.
Tom Lane [Sun, 11 Aug 2002 00:18:59 +0000 (00:18 +0000)]
Neil's patch was supposed to remove this, too.

22 years agoNow that we allow ANALYZE to run inside a transaction block, the locks
Tom Lane [Sun, 11 Aug 2002 00:08:48 +0000 (00:08 +0000)]
Now that we allow ANALYZE to run inside a transaction block, the locks
it takes could be held for quite awhile after the analyze step completes.
Rethink locking of pg_statistic in light of this fact.  The original
scheme took an exclusive lock on pg_statistic, which was okay when the
lock could be expected to be released shortly, but that doesn't hold
anymore.  Back off to a normal writer's lock (RowExclusiveLock).  This
allows concurrent ANALYZE of nonoverlapping sets of tables, at the price
that concurrent ANALYZEs of the same table may fail with 'tuple
concurrently updated'.

22 years agoMention ANALYZE after CLUSTER.
Bruce Momjian [Sat, 10 Aug 2002 21:03:33 +0000 (21:03 +0000)]
Mention ANALYZE after CLUSTER.

22 years agoUpdate with newer version of CLUSTER patch posted August 8.
Bruce Momjian [Sat, 10 Aug 2002 21:00:34 +0000 (21:00 +0000)]
Update with newer version of CLUSTER patch posted August 8.

> Looks like Alvaro got sideswiped by the system catalog indexing changes
> I made over the weekend.  It's a simple change, just reduce the whole
> mess to a "CatalogUpdateIndexes()" call.

I update two tuples, so I manually CatalogOpenIndexes() and
CatalogIndexInsert() two times, as per comments in
CatalogUpdateIndexes().

I also removed a couple of useless CommandCounterIncrement(), some
useless definitions in src/include/commands/cluster.h and useless
includes in src/backend/commands/cluster.c.  This version passes the
regression test I had made for previous versions.

Alvaro Herrera

22 years agoRemove error files.
Bruce Momjian [Sat, 10 Aug 2002 20:49:37 +0000 (20:49 +0000)]
Remove error files.

22 years agoFixed very stupid but important bug: mixing calls of some founctions from
Bruce Momjian [Sat, 10 Aug 2002 20:46:24 +0000 (20:46 +0000)]
Fixed very stupid but important bug: mixing calls of some founctions from
contrib/tsearch and contrib/ltree :)

Teodor Sigaev

22 years agoThe patch solves this problem, I hope...
Bruce Momjian [Sat, 10 Aug 2002 20:45:48 +0000 (20:45 +0000)]
The patch solves this problem, I hope...

Christopher Kings-Lynne wrote:
> I'm still getting ltree failures on 64bit freebsd:
>
> sed 's,MODULE_PATHNAME,$libdir/ltree,g' ltree.sql.in >ltree.sql
> gcc -pipe -O -g -Wall -Wmissing-prototypes -Wmissing-declarations -fpic -DPI
> C -DLOWER_NODE -I. -I../../src/include   -c -o ltree_io.o ltree_io.c -MMD
> ltree_io.c: In function `ltree_in':
> ltree_io.c:57: warning: int format, different type arg (arg 3)
> ltree_io.c:63: warning: int format, different type arg (arg 4)
> ltree_io.c:68: warning: int format, different type arg (arg 3)

Teodor Sigaev

22 years ago > create view pg_locks as select l.relation, l.database, l.backendpid,
Bruce Momjian [Sat, 10 Aug 2002 20:44:48 +0000 (20:44 +0000)]
 > create view pg_locks as select l.relation, l.database, l.backendpid,
 > l.mode, l.isgranted from pg_lock_info() as l(relation oid, database oid,
 > backendpid int4, mode text, isgranted bool);
 > ERROR:  badly formatted planstring "COLUMNDEF "...
 >

Reported by Neil Conway -- I never implemented readfuncs.c support for
ColumnDef or TypeName, which is needed so that views can be created on
functions returning type RECORD. Here's a patch.

Joe Conway

22 years agoMajor improvement in CLUSTER which preserves table characteristics using
Bruce Momjian [Sat, 10 Aug 2002 20:43:46 +0000 (20:43 +0000)]
Major improvement in CLUSTER which preserves table characteristics using
relfilenode.

I sent the CLUSTER patch a few days ago and I think it was missed.  I
append it again, this time including the regression test files.  For the
committer, please note that you have to cvs add the files as they don't
exist.  Maybe add to the parallel and serial schedules also, but I don't
know such stuff.

Alvaro Herrera (<alvherre[a]atentus.com>)

22 years agoAugust 6, 2002
Bruce Momjian [Sat, 10 Aug 2002 20:38:29 +0000 (20:38 +0000)]
August 6, 2002
   1. Reworked patch from Andrey Oktyabrski (ano@spider.ru) with
      functions: icount, sort, sort_asc, uniq, idx, subarray
      operations: #, +, -, |, &

FUNCTIONS:

  int   icount(int[]) - the number of elements in intarray
  int[] sort(int[], 'asc' | 'desc') - sort intarray
  int[] sort(int[]) - sort in ascending order
  int[] sort_asc(int[]),sort_desc(int[]) - shortcuts for sort
  int[] uniq(int[]) - returns unique elements
  int   idx(int[], int item) - returns index of first intarray matching element
                               to item, or '0' if matching failed.
  int[] subarray(int[],int START [, int LEN]) - returns part of intarray
                               starting from element number START (from 1)
                               and length LEN.
OPERATIONS:

  int[] && int[]  - overlap - returns TRUE if arrays has at least one common elements.
  int[] @  int[]  - contains - returns TRUE if left array contains right array
  int[] ~ int[]   - contained - returns TRUE if left array is contained in right array
  # int[]         - return the number of elements in array
  int[] + int     - push element to array ( add to end of array)
  int[] + int[]   - merge of arrays (right array added to the end of left one)
  int[] - int     - remove entries matched by right argument from array
  int[] - int[]   - remove left array from right
  int[] | int     - returns intarray - union of arguments
  int[] | int[]   - returns intarray as a union of two arrays
  int[] & int[]   - returns intersection of arrays

Oleg Bartunov

22 years agoThis patch removes a lot of unused code related to assertions and
Bruce Momjian [Sat, 10 Aug 2002 20:29:18 +0000 (20:29 +0000)]
This patch removes a lot of unused code related to assertions and
error handling, and simplifies the code that remains. Apparently,
the code that left Berkeley had a whole "error handling subsystem",
which exceptions and whatnot. Since we don't use that anymore,
there's no reason to keep it around.

The regression tests pass with the patch applied. Unless anyone
sees a problem, please apply.

Neil Conway

22 years agopsql prints its version number in its startup message, per recent
Tom Lane [Sat, 10 Aug 2002 19:35:01 +0000 (19:35 +0000)]
psql prints its version number in its startup message, per recent
discussion.  Also, cause the \timing command to display time in a
format consistent with the backend's EXPLAIN ANALYZE output.

22 years agoAllow schema-qualified operator names to be used in the optional
Tom Lane [Sat, 10 Aug 2002 19:01:53 +0000 (19:01 +0000)]
Allow schema-qualified operator names to be used in the optional
arguments of CREATE OPERATOR.

22 years agoFix failure to relink postmaster executable in the first make run if only a
Peter Eisentraut [Sat, 10 Aug 2002 17:59:28 +0000 (17:59 +0000)]
Fix failure to relink postmaster executable in the first make run if only a
single source file a few directories deep in the backend tree has changed.

22 years agoMake sure all clients have the same understanding of default user name
Peter Eisentraut [Sat, 10 Aug 2002 16:57:32 +0000 (16:57 +0000)]
Make sure all clients have the same understanding of default user name
and database.  In particular, make script wrappers understand the
PGDATABASE environment variable.

22 years agoGet rid of double PQclear(res) in error path.
Tom Lane [Sat, 10 Aug 2002 16:01:16 +0000 (16:01 +0000)]
Get rid of double PQclear(res) in error path.

22 years agoAdd a sanity check to make sure that all system catalogs that have OIDs
Tom Lane [Sat, 10 Aug 2002 15:54:04 +0000 (15:54 +0000)]
Add a sanity check to make sure that all system catalogs that have OIDs
also have a unique index on OID.

22 years agopsql backslash commands are schema-aware. Pattern matching behavior
Tom Lane [Sat, 10 Aug 2002 03:56:24 +0000 (03:56 +0000)]
psql backslash commands are schema-aware.  Pattern matching behavior
follows recent pghackers discussion.  This commit includes all the
relevant fixes from Greg Mullane's patch of 24-June.

22 years agoMake sure monetary, numeric, and time locale categories are set to C and
Peter Eisentraut [Fri, 9 Aug 2002 22:52:04 +0000 (22:52 +0000)]
Make sure monetary, numeric, and time locale categories are set to C and
are only activated temporarily to read out formatting information.

22 years agoColumn ordering in \d output should agree with sort ordering;
Tom Lane [Fri, 9 Aug 2002 18:06:57 +0000 (18:06 +0000)]
Column ordering in \d output should agree with sort ordering;
per Joe Conway's patch of 20-July.

22 years agohas_table_privilege spawns scions has_database_privilege, has_function_privilege,
Tom Lane [Fri, 9 Aug 2002 16:45:16 +0000 (16:45 +0000)]
has_table_privilege spawns scions has_database_privilege, has_function_privilege,
has_language_privilege, has_schema_privilege to let SQL queries test
all the new privilege types in 7.3.  Also, add functions pg_table_is_visible,
pg_type_is_visible, pg_function_is_visible, pg_operator_is_visible,
pg_opclass_is_visible to test whether objects contained in schemas are
visible in the current search path.  Do some minor cleanup to centralize
accesses to pg_database, as well.

22 years agoChange LOCK statements to start new transaction
Hiroshi Inoue [Fri, 9 Aug 2002 10:04:21 +0000 (10:04 +0000)]
Change LOCK statements to start new transaction
in auto-commit off mode.

22 years agoDone:
Bruce Momjian [Thu, 8 Aug 2002 23:50:38 +0000 (23:50 +0000)]
Done:

>  o -Fix PL/PgSQL to handle quoted mixed-case identifiers

22 years agoAdd:
Bruce Momjian [Thu, 8 Aug 2002 23:20:11 +0000 (23:20 +0000)]
Add:

> * Allow INET subnet tests using non-constants

22 years agoMake the standard system views schema-aware.
Tom Lane [Thu, 8 Aug 2002 19:39:05 +0000 (19:39 +0000)]
Make the standard system views schema-aware.
I did not force an initdb via catversion ... but the rules
regression test will fail until you do an initdb.

22 years agoArrange to collect statistics on a few more system catalog columns.
Tom Lane [Thu, 8 Aug 2002 19:37:11 +0000 (19:37 +0000)]
Arrange to collect statistics on a few more system catalog columns.

22 years agoCause view/rule display to work as expected after rename of an underlying
Tom Lane [Thu, 8 Aug 2002 17:00:19 +0000 (17:00 +0000)]
Cause view/rule display to work as expected after rename of an underlying
table or column, or of an output column of the view itself.

22 years agoSome improvements in geometric-operators documentation.
Tom Lane [Thu, 8 Aug 2002 14:29:07 +0000 (14:29 +0000)]
Some improvements in geometric-operators documentation.

22 years agoFix bug in encoding conversion table
Tatsuo Ishii [Thu, 8 Aug 2002 08:21:52 +0000 (08:21 +0000)]
Fix bug in encoding conversion table

22 years agoAddress build problems on cygwin and (hopefully) AIX.
Tatsuo Ishii [Thu, 8 Aug 2002 07:47:43 +0000 (07:47 +0000)]
Address build problems on cygwin and (hopefully) AIX.

22 years agoLoad and keep conversion function info when SET CLIENT_ENCODING TO is
Tatsuo Ishii [Thu, 8 Aug 2002 06:35:26 +0000 (06:35 +0000)]
Load and keep conversion function info when SET CLIENT_ENCODING TO is
executed to prevent database access while performing encoding
conversion.

22 years agoRemove #ifdef MULTIBYTE
Tatsuo Ishii [Thu, 8 Aug 2002 06:32:26 +0000 (06:32 +0000)]
Remove #ifdef MULTIBYTE

22 years agoCause schema-qualified FROM items and schema-qualified variable references
Tom Lane [Thu, 8 Aug 2002 01:44:31 +0000 (01:44 +0000)]
Cause schema-qualified FROM items and schema-qualified variable references
to behave according to SQL92 (or according to my current understanding
of same, anyway).  Per pghackers discussion way back in March 2002:
thread 'Do FROM items of different schemas conflict?'

22 years agoClean up plpgsql identifier handling: process quoted identifiers
Tom Lane [Thu, 8 Aug 2002 01:36:05 +0000 (01:36 +0000)]
Clean up plpgsql identifier handling: process quoted identifiers
correctly, truncate to NAMEDATALEN where needed, allow whitespace
around dots in qualified identifiers.  Get rid of T_RECFIELD and
T_TGARGV token categories, which weren't accomplishing anything
except to create room for sins of omission in the grammar, ie,
places that should have allowed them and didn't.  Fix a few other
bugs en passant.

22 years agoparseTypeString wasn't quite paranoid enough.
Tom Lane [Thu, 8 Aug 2002 01:22:35 +0000 (01:22 +0000)]
parseTypeString wasn't quite paranoid enough.

22 years agoFix permission checking for temp-table namespace.
Tom Lane [Wed, 7 Aug 2002 21:45:02 +0000 (21:45 +0000)]
Fix permission checking for temp-table namespace.

22 years agoImprove 'failed to initialize' messages.
Tom Lane [Wed, 7 Aug 2002 17:26:24 +0000 (17:26 +0000)]
Improve 'failed to initialize' messages.

22 years agoUpdated translation from Dennis Bjorklund <db@zigo.dhs.org>
Peter Eisentraut [Tue, 6 Aug 2002 21:13:44 +0000 (21:13 +0000)]
Updated translation from Dennis Bjorklund <db@zigo.dhs.org>

22 years agoStill more paranoia in PageAddItem: disallow specification of an item
Tom Lane [Tue, 6 Aug 2002 19:41:23 +0000 (19:41 +0000)]
Still more paranoia in PageAddItem: disallow specification of an item
offset past the last-used-item-plus-one, since that would result in
leaving uninitialized holes in the item pointer array.  AFAICT the only
place that was depending on this was btree index build, which was being
cavalier about when to fill in the P_HIKEY pointer; easily fixed.
Also a small performance improvement: shuffle itemid's by means of
memmove, not a one-at-a-time loop.

22 years agoRemove no-longer-used PageManagerMode enum.
Tom Lane [Tue, 6 Aug 2002 19:37:10 +0000 (19:37 +0000)]
Remove no-longer-used PageManagerMode enum.

22 years agoMove pg_convert3 declaration to suppress compile warning. Really
Tom Lane [Tue, 6 Aug 2002 14:11:06 +0000 (14:11 +0000)]
Move pg_convert3 declaration to suppress compile warning.  Really
pg_convert3 does not belong in a backend/catalog file at all, IMHO;
it should be in utils/adt.

22 years agoAdd SQL99 CONVERT() function.
Tatsuo Ishii [Tue, 6 Aug 2002 05:40:47 +0000 (05:40 +0000)]
Add SQL99 CONVERT() function.

22 years agoPatch for current CVS. It add test of lca() to ltree test suite.
Bruce Momjian [Tue, 6 Aug 2002 05:35:29 +0000 (05:35 +0000)]
Patch for current CVS. It add test of lca() to ltree test suite.

Teodor Sigaev

22 years agoOf course, the simple change has ripple effects! Here's a patch for the
Bruce Momjian [Tue, 6 Aug 2002 05:34:46 +0000 (05:34 +0000)]
Of course, the simple change has ripple effects! Here's a patch for the
rangefunc regression test for the new behavior.

Joe Conway

22 years ago>> Hm. I'd sort of expect the "z" to become both the table and column
Bruce Momjian [Tue, 6 Aug 2002 05:34:10 +0000 (05:34 +0000)]
>> Hm.  I'd sort of expect the "z" to become both the table and column
>> alias in this case.  What do you think?
>
> I guess that would make sense. I'll make a separate patch just for
that
> change if that's OK.
>

Simple change -- patch attached.

test=# select * from myfoo1() as z;
  z
----
   1
   2
   3
(3 rows)

Joe Conway

22 years agoThe attached patch disallows the use of coldeflists for functions that
Bruce Momjian [Tue, 6 Aug 2002 05:33:29 +0000 (05:33 +0000)]
The attached patch disallows the use of coldeflists for functions that
don't return type RECORD. It also catches a core dump condition when a
function returning RECORD had an alias list instead of a coldeflist.

Now both conditions throw an ERROR.

Joe Conway

22 years agoThis patch fixes two typos in the documentation for the newly added
Bruce Momjian [Tue, 6 Aug 2002 05:32:16 +0000 (05:32 +0000)]
This patch fixes two typos in the documentation for the newly added
START TRANSACTION command.

Neil Conway

22 years agoThis patch changes the behavior of PostgreSQL so that if any queries are
Bruce Momjian [Tue, 6 Aug 2002 05:24:04 +0000 (05:24 +0000)]
This patch changes the behavior of PostgreSQL so that if any queries are
executed in an implicitely aborted transaction (e.g. after an occur
occurs), we return an error (and not just a warning). For example:

nconway=# begin;
BEGIN
nconway=# insert; -- syntax error
ERROR:  parser: parse error at or near ";"
nconway=# select * from a;
ERROR:  current transaction is aborted, queries ignored until end of
transaction block

The old behavior was:

nconway=# begin;
BEGIN
nconway=# insert;
ERROR:  parser: parse error at or near ";"
nconway=# select * from a;
WARNING:  current transaction is aborted, queries ignored until end
of transaction block
*ABORT STATE*

Which can be confusing: if the client isn't paying careful attention,
they will conclude that the query has executed (because no error is
returned).

Neil Conway

22 years agoRestructure local-buffer handling per recent pghackers discussion.
Tom Lane [Tue, 6 Aug 2002 02:36:35 +0000 (02:36 +0000)]
Restructure local-buffer handling per recent pghackers discussion.
The local buffer manager is no longer used for newly-created relations
(unless they are TEMP); a new non-TEMP relation goes through the shared
bufmgr and thus will participate normally in checkpoints.  But TEMP relations
use the local buffer manager throughout their lifespan.  Also, operations
in TEMP relations are not logged in WAL, thus improving performance.
Since it's no longer necessary to fsync relations as they move out of the
local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code
is no longer needed and has been removed: there's no concept of a dirty
relation anymore in md.c/fd.c, and we never fsync anything but WAL.
Still TODO: improve local buffer management algorithms so that it would
be reasonable to increase NLocBuffer.

22 years agoForgot to add/remove files.
Peter Eisentraut [Mon, 5 Aug 2002 19:44:58 +0000 (19:44 +0000)]
Forgot to add/remove files.

22 years agoAdd User's Guide chapters on Data Definition and Data Manipulation.
Peter Eisentraut [Mon, 5 Aug 2002 19:43:31 +0000 (19:43 +0000)]
Add User's Guide chapters on Data Definition and Data Manipulation.
Still needs to be filled with more information, but it gives us a
framework to have a User's Guide with complete coverage of the basic
SQL operations.  Move arrays into data type chapter, inheritance into
DDL chapter (for now).

Make <comment>s show up in the output while the version number ends in
"devel".

Allow cross-book references with entities &cite-user; etc.

22 years agoRestructure system-catalog index updating logic. Instead of having
Tom Lane [Mon, 5 Aug 2002 03:29:17 +0000 (03:29 +0000)]
Restructure system-catalog index updating logic.  Instead of having
hardwired lists of index names for each catalog, use the relcache's
mechanism for caching lists of OIDs of indexes of any table.  This
reduces the common case of updating system catalog indexes to a single
line, makes it much easier to add a new system index (in fact, you
can now do so on-the-fly if you want to), and as a nice side benefit
improves performance a little.  Per recent pghackers discussion.

22 years agoPreliminary code review for anonymous-composite-types patch: fix breakage
Tom Lane [Mon, 5 Aug 2002 02:30:50 +0000 (02:30 +0000)]
Preliminary code review for anonymous-composite-types patch: fix breakage
of functions returning domain types, update documentation for typtype,
move get_typtype to lsyscache.c (actually, resurrect the old version),
add defense against creating pseudo-typed table columns, fix some
bogus list-parsing in grammar.  Issues remain with respect to alias
handling and type checking; Joe is on those.

22 years agoFix compilation problem with assert checking enabled for recent xlog
Thomas G. Lockhart [Mon, 5 Aug 2002 01:24:16 +0000 (01:24 +0000)]
Fix compilation problem with assert checking enabled for recent xlog
 location feature.

22 years agoCode review for anonymous-functions patch --- clean up some confusion
Tom Lane [Mon, 5 Aug 2002 00:21:27 +0000 (00:21 +0000)]
Code review for anonymous-functions patch --- clean up some confusion
in checkretval about which paths are for base or complex return type.

22 years agoTemporary solution for XLogDir breakage.
Tom Lane [Sun, 4 Aug 2002 23:56:01 +0000 (23:56 +0000)]
Temporary solution for XLogDir breakage.

22 years agoFix merge failures for anonymous-type patch. From Joe Conway.
Tom Lane [Sun, 4 Aug 2002 23:49:59 +0000 (23:49 +0000)]
Fix merge failures for anonymous-type patch.  From Joe Conway.

22 years agoFix broken pg_backend_pid reference, per Joe Conway.
Tom Lane [Sun, 4 Aug 2002 23:46:38 +0000 (23:46 +0000)]
Fix broken pg_backend_pid reference, per Joe Conway.

22 years agoAnother backend_pid rename.
Bruce Momjian [Sun, 4 Aug 2002 20:01:33 +0000 (20:01 +0000)]
Another backend_pid rename.

22 years agoFix compile failures for FRS composite tyhpe patch until Joe can fix it.
Bruce Momjian [Sun, 4 Aug 2002 20:00:15 +0000 (20:00 +0000)]
Fix compile failures for FRS composite tyhpe patch until Joe can fix it.

22 years agoRename backend_pid to pg_backend_pid, move docs to monitoring section.
Bruce Momjian [Sun, 4 Aug 2002 19:51:30 +0000 (19:51 +0000)]
Rename backend_pid to pg_backend_pid, move docs to monitoring section.

22 years agoAttached are two patches to implement and document anonymous composite
Bruce Momjian [Sun, 4 Aug 2002 19:48:11 +0000 (19:48 +0000)]
Attached are two patches to implement and document anonymous composite
types for Table Functions, as previously proposed on HACKERS. Here is a
brief explanation:

1. Creates a new pg_type typtype: 'p' for pseudo type (currently either
     'b' for base or 'c' for catalog, i.e. a class).

2. Creates new builtin type of typtype='p' named RECORD. This is the
     first of potentially several pseudo types.

3. Modify FROM clause grammer to accept:
     SELECT * FROM my_func() AS m(colname1 type1, colname2 type1, ...)
     where m is the table alias, colname1, etc are the column names, and
     type1, etc are the column types.

4. When typtype == 'p' and the function return type is RECORD, a list
     of column defs is required, and when typtype != 'p', it is
disallowed.

5. A check was added to ensure that the tupdesc provide via the parser
     and the actual return tupdesc match in number and type of
attributes.

When creating a function you can do:
     CREATE FUNCTION foo(text) RETURNS setof RECORD ...

When using it you can do:
     SELECT * from foo(sqlstmt) AS (f1 int, f2 text, f3 timestamp)
       or
     SELECT * from foo(sqlstmt) AS f(f1 int, f2 text, f3 timestamp)
       or
     SELECT * from foo(sqlstmt) f(f1 int, f2 text, f3 timestamp)

Included in the patches are adjustments to the regression test sql and
expected files, and documentation.

p.s.
     This potentially solves (or at least improves) the issue of builtin
     Table Functions. They can be bootstrapped as returning RECORD, and
     we can wrap system views around them with properly specified column
     defs. For example:

     CREATE VIEW pg_settings AS
       SELECT s.name, s.setting
       FROM show_all_settings()AS s(name text, setting text);

     Then we can also add the UPDATE RULE that I previously posted to
     pg_settings, and have pg_settings act like a virtual table, allowing
     settings to be queried and set.

Joe Conway

22 years agoFix obsolete comment.
Tom Lane [Sun, 4 Aug 2002 18:12:15 +0000 (18:12 +0000)]
Fix obsolete comment.

22 years agoMinor clarification or fix of typos.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:54:10 +0000 (06:54 +0000)]
Minor clarification or fix of typos.

22 years agoImplement WAL log location control using "-X" or PGXLOG.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:53:10 +0000 (06:53 +0000)]
Implement WAL log location control using "-X" or PGXLOG.

22 years agoImplement IS OF and IS NOT OF type predicate.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:51:23 +0000 (06:51 +0000)]
Implement IS OF and IS NOT OF type predicate.
 Can now do queries of the form: SELECT value IS OF (integer, float8);
Define macros for handling typmod manipulation for date/time types.
 Should be more robust than all of that brute-force inline code.
Rename macros for masking and typmod manipulation to put TIMESTAMP_
 or INTERVAL_ in front of the macro name, to reduce the possibility
 of name space collisions.
Allow bit string constants without fully-specified length declaration.
Try implementing CREATE TABLE/OF as a mapping to inheritance.
 May be appropriate, or may be replace later with something more exactly
 like one might expect from databases without the feature.

22 years agoImplement IS OF type predicate. Can now do queries of the form:
Thomas G. Lockhart [Sun, 4 Aug 2002 06:46:12 +0000 (06:46 +0000)]
Implement IS OF type predicate. Can now do queries of the form:
 select value IS OF (integer, float8);

22 years agoAdd guard code to protect from buffer overruns on long date/time input
Thomas G. Lockhart [Sun, 4 Aug 2002 06:44:47 +0000 (06:44 +0000)]
Add guard code to protect from buffer overruns on long date/time input
 strings. Should go back in and look at doing this a bit more elegantly
 and (hopefully) cheaper. Probably not too bad anyway, but it seems a
 shame to scan the strings twice: once for length for this buffer overrun
 protection, and once to parse the line.
Remove use of pow() in date/time handling; was already gone from everything
 *but* the time data types.
Define macros for handling typmod manipulation for date/time types.
 Should be more robust than all of that brute-force inline code.
Rename macros for masking and typmod manipulation to put TIMESTAMP_
 or INTERVAL_ in front of the macro name, to reduce the possibility
 of name space collisions.

22 years agoDefine macros for handling typmod manipulation for date/time types.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:42:18 +0000 (06:42 +0000)]
Define macros for handling typmod manipulation for date/time types.
 Should be more robust than all of that brute-force inline code.
Rename macros for masking and typmod manipulation to put TIMESTAMP_
 or INTERVAL_ in front of the macro name, to reduce the possibility
 of name space collisions.

22 years agoImplement hex literal conversion to bit string literal.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:36:18 +0000 (06:36 +0000)]
Implement hex literal conversion to bit string literal.
 May not be the long-term solution (some continuing discussion with
 Peter E.) but better than the current mapping of a conversion to integer
 which I'd put in years ago before we had any bit string types at all.
This is already supported in the bit string implementation elsewhere.

22 years agoAllow bit string constants without fully-specified length declaration.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:33:59 +0000 (06:33 +0000)]
Allow bit string constants without fully-specified length declaration.
Implement conversion between 8-byte integers and bit strings.
 Similar to what is done for 4-byte integers.

22 years agoImplement WAL log location control using "-X" or PGXLOG.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:26:38 +0000 (06:26 +0000)]
Implement WAL log location control using "-X" or PGXLOG.

22 years agoAdd IS OF type predicate.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:17:29 +0000 (06:17 +0000)]
Add IS OF type predicate.

22 years agoAdd Myannar Time, Iran Time variant name, and Marquesas Time.
Thomas G. Lockhart [Sun, 4 Aug 2002 06:15:45 +0000 (06:15 +0000)]
Add Myannar Time, Iran Time variant name, and Marquesas Time.

22 years agoFix broken markup.
Tom Lane [Sun, 4 Aug 2002 05:46:02 +0000 (05:46 +0000)]
Fix broken markup.

22 years agoNeil's patch claimed a column list didn't work for COPY BINARY.
Tom Lane [Sun, 4 Aug 2002 05:22:02 +0000 (05:22 +0000)]
Neil's patch claimed a column list didn't work for COPY BINARY.
Which was true when he submitted it, but is so no longer.