OSDN Git Service

pg-rex/syncrep.git
17 years agoCreate hooks to let a loadable plugin monitor (or even replace) the planner
Tom Lane [Fri, 25 May 2007 17:54:25 +0000 (17:54 +0000)]
Create hooks to let a loadable plugin monitor (or even replace) the planner
and/or create plans for hypothetical situations; in particular, investigate
plans that would be generated using hypothetical indexes.  This is a
heavily-rewritten version of the hooks proposed by Gurjeet Singh for his
Index Advisor project.  In this formulation, the index advisor can be
entirely a loadable module instead of requiring a significant part to be
in the core backend, and plans can be generated for hypothetical indexes
without requiring the creation and rolling-back of system catalog entries.

The index advisor patch as-submitted is not compatible with these hooks,
but it needs significant work anyway due to other 8.2-to-8.3 planner
changes.  With these hooks in the core backend, development of the advisor
can proceed as a pgfoundry project.

17 years agoRemove ruleutils.c's use of varnoold/varoattno as a shortcut for determining
Tom Lane [Thu, 24 May 2007 18:58:42 +0000 (18:58 +0000)]
Remove ruleutils.c's use of varnoold/varoattno as a shortcut for determining
what a Var node refers to.  This is no longer necessary because the new
flat-range-table representation of plan trees makes it relatively easy to dig
down through child plan levels to find the original reference; and to keep
doing it that way, we'd have to store joinaliasvars lists in flattened RTEs,
as demonstrated by bug report from Leszek Trenkner.  This change makes
varnoold/varoattno truly just debug aids, which wasn't quite the case before.
Perhaps we should drop them, or only have them in assert-enabled builds?

17 years agoAvoid assuming that the fields of struct timeval have exactly type long.
Tom Lane [Thu, 24 May 2007 18:54:10 +0000 (18:54 +0000)]
Avoid assuming that the fields of struct timeval have exactly type long.
This is probably incorrect on some platforms, and definitely draws a
compiler warning on Darwin.

17 years agoAdd deprecation notice about planned removal in 8.4 and advice to try using
Peter Eisentraut [Thu, 24 May 2007 11:13:08 +0000 (11:13 +0000)]
Add deprecation notice about planned removal in 8.4 and advice to try using
the newer XML stuff in core.

(This should probably also be referred to in the release notes.)

17 years agoRepair planner bug introduced in 8.2 by ability to rearrange outer joins:
Tom Lane [Tue, 22 May 2007 23:23:58 +0000 (23:23 +0000)]
Repair planner bug introduced in 8.2 by ability to rearrange outer joins:
in cases where a sub-SELECT inserts a WHERE clause between two outer joins,
that clause may prevent us from re-ordering the two outer joins.  The code
was considering only the joins' own ON-conditions in determining reordering
safety, which is not good enough.  Add a "delay_upper_joins" flag to
OuterJoinInfo to flag that we have detected such a clause and higher-level
outer joins shouldn't be permitted to commute with this one.  (This might
seem overly coarse, but given the current rules for OJ reordering, it's
sufficient AFAICT.)

The failure case is actually pretty narrow: it needs a WHERE clause within
the RHS of a left join that checks the RHS of a lower left join, but is not
strict for that RHS (else we'd have simplified the lower join to a plain
join).  Even then no failure will be manifest unless the planner chooses to
rearrange the join order.

Per bug report from Adam Terrey.

17 years agoRemove unsupported -u option from pg_standby example usage.
Alvaro Herrera [Tue, 22 May 2007 15:31:44 +0000 (15:31 +0000)]
Remove unsupported -u option from pg_standby example usage.
From Pavel Stehule.

17 years agoFix best_inner_indexscan to return both the cheapest-total-cost and
Tom Lane [Tue, 22 May 2007 01:40:33 +0000 (01:40 +0000)]
Fix best_inner_indexscan to return both the cheapest-total-cost and
cheapest-startup-cost innerjoin indexscans, and make joinpath.c consider
both of these (when different) as the inside of a nestloop join.  The
original design was based on the assumption that indexscan paths always
have negligible startup cost, and so total cost is the only important
figure of merit; an assumption that's obviously broken by bitmap
indexscans.  This oversight could lead to choosing poor plans in cases
where fast-start behavior is more important than total cost, such as
LIMIT and IN queries.  8.1-vintage brain fade exposed by an example from
Chuck D.

17 years agoTeach tuplestore.c to throw away data before the "mark" point when the caller
Tom Lane [Mon, 21 May 2007 17:57:35 +0000 (17:57 +0000)]
Teach tuplestore.c to throw away data before the "mark" point when the caller
is using mark/restore but not rewind or backward-scan capability.  Insert a
materialize plan node between a mergejoin and its inner child if the inner
child is a sort that is expected to spill to disk.  The materialize shields
the sort from the need to do mark/restore and thereby allows it to perform
its final merge pass on-the-fly; while the materialize itself is normally
cheap since it won't spill to disk unless the number of tuples with equal
key values exceeds work_mem.

Greg Stark, with some kibitzing from Tom Lane.

17 years agoXPath fixes:
Peter Eisentraut [Mon, 21 May 2007 17:10:29 +0000 (17:10 +0000)]
XPath fixes:
 - Function renamed to "xpath".
 - Function is now strict, per discussion.
 - Return empty array in case when XPath expression detects nothing
   (previously, NULL was returned in such case), per discussion.
 - (bugfix) Work with fragments with prologue: select xpath('/a',
   '<?xml version="1.0"?><a /><b />'); // now XML datum is always wrapped
   with dummy <x>...</x>, XML prologue simply goes away (if any).
 - Some cleanup.

Nikolay Samokhvalov

Some code cleanup and documentation work by myself.

17 years agoFix spurious German index entry
Peter Eisentraut [Mon, 21 May 2007 15:11:12 +0000 (15:11 +0000)]
Fix spurious German index entry

17 years agoAlso removed the function not just the call, sorry.
Michael Meskes [Mon, 21 May 2007 07:07:48 +0000 (07:07 +0000)]
Also removed the function not just the call, sorry.

17 years agoDo not try to find local timezone in DecodeDateTime() because the
Michael Meskes [Mon, 21 May 2007 07:04:00 +0000 (07:04 +0000)]
Do not try to find local timezone in DecodeDateTime() because the
result is not used anyway. This also fixes Vista's build problems.

17 years agoTo support external compression of archived WAL data, add a flag bit to
Tom Lane [Sun, 20 May 2007 21:08:19 +0000 (21:08 +0000)]
To support external compression of archived WAL data, add a flag bit to
WAL records that shows whether it is safe to remove full-page images
(ie, whether or not an on-line backup was in progress when the WAL entry
was made).  Also make provision for an XLOG_NOOP record type that can be
used to fill in the extra space when decompressing the data for restore.

This is the portion of Koichi Suzuki's "full page writes" patch that
has to go into the core database.  The remainder of that work is two
external compression and decompression programs, which for the time being
will undergo separate development on pgfoundry.  Per discussion.

Also, twiddle the handling of BTREE_SPLIT records to ensure it'll be
possible to compress them (the previous coding caused essential info
to be omitted).  The other commonly-used record types seem OK already,
with the possible exception of GIN and GIST WAL records, which I don't
understand well enough to opine on.

17 years agotackling Vista problem - another try
Michael Meskes [Sun, 20 May 2007 11:30:30 +0000 (11:30 +0000)]
tackling Vista problem - another try

17 years agoFix dumb compile error in the last patch.
Alvaro Herrera [Sat, 19 May 2007 01:02:34 +0000 (01:02 +0000)]
Fix dumb compile error in the last patch.

17 years agoHave CLUSTER advance the table's relfrozenxid. The new frozen point is the
Alvaro Herrera [Fri, 18 May 2007 23:19:42 +0000 (23:19 +0000)]
Have CLUSTER advance the table's relfrozenxid.  The new frozen point is the
FreezeXid introduced in a recent commit, so there isn't any data loss in this
approach.

Doing it causes ALTER TABLE (or rather, the forms of it that cause a full table
rewrite) to be affected as well.  In this case, the frozen point is RecentXmin,
because after the rewrite all the tuples are relabeled with the rewriting
transaction's Xid.

TOAST tables are fixed automatically as well, as fallout of the way they were
already being handled in the respective code paths.

With this patch, there is no longer need to VACUUM tables for Xid wraparound
purposes that have been cleaned up via TRUNCATE or CLUSTER.

17 years agoMark some intermediate targets as .PRECIOUS, to replace the recently removed
Peter Eisentraut [Fri, 18 May 2007 15:55:03 +0000 (15:55 +0000)]
Mark some intermediate targets as .PRECIOUS, to replace the recently removed
.SECONDARY target.  This makes experimentation with the PDF builds easier.

17 years agoDone:
Bruce Momjian [Fri, 18 May 2007 01:22:09 +0000 (01:22 +0000)]
Done:

< * Fix problem with excessive logging during SSL disconnection
> * -Fix problem with excessive logging during SSL disconnection

17 years agoRemove redundant logging of send failures when SSL is in use. While pqcomm.c
Tom Lane [Fri, 18 May 2007 01:20:16 +0000 (01:20 +0000)]
Remove redundant logging of send failures when SSL is in use.  While pqcomm.c
had been taught not to do that ages ago, the SSL code was helpfully bleating
anyway.  Resolves some recent reports such as bug #3266; however the
underlying cause of the related bug #2829 is still unclear.

17 years agodded to TODO:
Bruce Momjian [Thu, 17 May 2007 23:42:33 +0000 (23:42 +0000)]
dded to TODO:

>
> * Support scoped IPv6 addresses
>
>   http://archives.postgresql.org/pgsql-bugs/2007-05/msg00111.php

17 years agoTweaks for SGML markup.
Neil Conway [Thu, 17 May 2007 23:36:04 +0000 (23:36 +0000)]
Tweaks for SGML markup.

17 years agoTemporary fix for the problem that pg_stat_activity, inet_client_addr(),
Tom Lane [Thu, 17 May 2007 23:31:49 +0000 (23:31 +0000)]
Temporary fix for the problem that pg_stat_activity, inet_client_addr(),
and inet_server_addr() fail if the client connected over a "scoped" IPv6
address.  In this case getnameinfo() will return a string ending with
a poorly-standardized "%something" zone specifier, which these functions
try to feed to network_in(), which won't take it.  So that we don't lose
functionality altogether, suppress the zone specifier before giving the
string to network_in().  Per report from Brian Hirt.

TODO: probably someday the inet type should support scoped IPv6 addresses,
and then this patch should be reverted.

Backpatch to 8.2 ... is it worth going further?

17 years agoAdd URL for:
Bruce Momjian [Thu, 17 May 2007 22:53:23 +0000 (22:53 +0000)]
Add URL for:

* Implement the SQL standard mechanism whereby REVOKE ROLE revokes only
  the privilege granted by the invoking role, and not those granted
  by other roles
>
>   http://archives.postgresql.org/pgsql-bugs/2007-05/msg00010.php

17 years agoAdd, per Alvaro:
Bruce Momjian [Thu, 17 May 2007 22:44:11 +0000 (22:44 +0000)]
Add, per Alvaro:

>
> * Implement the SQL standard mechanism whereby REVOKE ROLE revokes only
>   the privilege granted by the invoking role, and not those granted
>   by other roles

17 years agoAdd:
Bruce Momjian [Thu, 17 May 2007 22:42:54 +0000 (22:42 +0000)]
Add:

>
> * Fix problem with excessive logging during SSL disconnection
>
>   http://archives.postgresql.org/pgsql-bugs/2006-12/msg00122.php
>   http://archives.postgresql.org/pgsql-bugs/2007-05/msg00065.php

17 years agoAdd CVS comment for recent commit:
Bruce Momjian [Thu, 17 May 2007 21:43:19 +0000 (21:43 +0000)]
Add CVS comment for recent commit:

        Moved page-level functions from pgstattuple to
contrib/pageinspect.

17 years agoRemoved mktime() call to hopefully get rid of some compatibility problems.
Michael Meskes [Thu, 17 May 2007 20:07:35 +0000 (20:07 +0000)]
Removed mktime() call to hopefully get rid of some compatibility problems.

17 years agoFix parameter recalculation for Limit nodes: during a ReScan call we must
Tom Lane [Thu, 17 May 2007 19:35:08 +0000 (19:35 +0000)]
Fix parameter recalculation for Limit nodes: during a ReScan call we must
recompute the limit/offset immediately, so that the updated values are
available when the child's ReScan function is invoked.  Add a regression
test for this, too.  Bug is new in HEAD (due to the bounded-sorting patch)
so no need for back-patch.

I did not do anything about merging this signaling with chgParam processing,
but if we were to do that we'd still need to compute the updated values
at this point rather than during the first ProcNode call.

Per observation and test case from Greg Stark, though I didn't use his patch.

17 years agoAdd database page inspection /contrib module.
Bruce Momjian [Thu, 17 May 2007 19:11:25 +0000 (19:11 +0000)]
Add database page inspection /contrib module.

Simon and Heikki

17 years agoMove the tuple freezing point in CLUSTER to a point further back in the past,
Alvaro Herrera [Thu, 17 May 2007 15:28:29 +0000 (15:28 +0000)]
Move the tuple freezing point in CLUSTER to a point further back in the past,
to avoid losing useful Xid information in not-so-old tuples.  This makes
CLUSTER behave the same as VACUUM as far a tuple-freezing behavior goes
(though CLUSTER does not yet advance the table's relfrozenxid).

While at it, move the actual freezing operation in rewriteheap.c to a more
appropriate place, and document it thoroughly.  This part of the patch from
Tom Lane.

17 years agoHave TRUNCATE advance the affected table's relfrozenxid to RecentXmin, to
Alvaro Herrera [Wed, 16 May 2007 17:28:20 +0000 (17:28 +0000)]
Have TRUNCATE advance the affected table's relfrozenxid to RecentXmin, to
avoid a later needless VACUUM for Xid-wraparound purposes.  We can do this
since the table is known to be left empty, so no Xid remains on it.

Per discussion.

17 years agoHave the rewriteheap code freeze old tuples. This is safe because it is only
Alvaro Herrera [Wed, 16 May 2007 16:36:56 +0000 (16:36 +0000)]
Have the rewriteheap code freeze old tuples.  This is safe because it is only
applied to live tuples older than a recent Xmin, not to tuples that may be part
of an update chain.  Those still keep their original markings.

This patch makes it possible for CLUSTER to advance relfrozenxid, thus avoiding
the need of vacuuming the table for Xid wraparound purposes.  That will be
patched separately.

Patch from Heikki Linnakangas.

17 years agoAvoid emitting empty role names in the GRANTED BY clause of GRANT ROLE
Alvaro Herrera [Tue, 15 May 2007 20:20:21 +0000 (20:20 +0000)]
Avoid emitting empty role names in the GRANTED BY clause of GRANT ROLE
when the grantor has been dropped.  This is a workaround for the fact
that we don't track the grantor as a shared dependency.

17 years agoRemove directory qualification in <ossp/uuid.h> because it's not always installed...
Andrew Dunstan [Tue, 15 May 2007 19:47:51 +0000 (19:47 +0000)]
Remove directory qualification in <ossp/uuid.h> because it's not always installed in ossp.
Workaround for when it is: include the ossp directory using --with-includes.

17 years agoFix a bunch of bad grammar in the docs: "<link>, which see for more
Neil Conway [Tue, 15 May 2007 19:43:51 +0000 (19:43 +0000)]
Fix a bunch of bad grammar in the docs: "<link>, which see for more
information" is un-good English.

17 years agoVarious fixes for the SGML docs. Consistently use spaces before/after
Neil Conway [Tue, 15 May 2007 19:13:55 +0000 (19:13 +0000)]
Various fixes for the SGML docs. Consistently use spaces before/after
parentheses in syntax descriptions. Consistently use the present tense
when describing the basic purpose of each "DROP" command. Add a few
more hyperlinks.

17 years agoUpdate comments for PG_DETOAST_PACKED and VARDATA_ANY on a structures
Bruce Momjian [Tue, 15 May 2007 17:39:54 +0000 (17:39 +0000)]
Update comments for PG_DETOAST_PACKED and VARDATA_ANY on a structures
that require alignment.

Add a paragraph to the "User-Defined Types" chapter on using these
macros since it seems like they're a hit.

Gregory Stark

17 years agoMake a few marginal improvements to the documentation for the AV
Neil Conway [Tue, 15 May 2007 15:52:40 +0000 (15:52 +0000)]
Make a few marginal improvements to the documentation for the AV
launcher daemon.

17 years agoAdd a note to the documentation to clarify that even when
Neil Conway [Tue, 15 May 2007 15:35:46 +0000 (15:35 +0000)]
Add a note to the documentation to clarify that even when
"autovacuum = off", the system may still periodically start autovacuum
processes to prevent XID wraparound. Patch from David Fetter, with
editorializing.

17 years agoAdd URLs for:
Bruce Momjian [Tue, 15 May 2007 02:36:43 +0000 (02:36 +0000)]
Add URLs for:

* Add support for SQL-standard GENERATED/IDENTITY columns

>   http://archives.postgresql.org/pgsql-hackers/2007-05/msg00344.php
>   http://archives.postgresql.org/pgsql-patches/2007-05/msg00076.php

17 years agofix perl thinko
Andrew Dunstan [Tue, 15 May 2007 01:57:57 +0000 (01:57 +0000)]
fix perl thinko

17 years agoUse configured pgport in MSVC config file.
Andrew Dunstan [Tue, 15 May 2007 00:15:01 +0000 (00:15 +0000)]
Use configured pgport in MSVC config file.

17 years agoGet rid of the pg_shdepend entry for a TOAST table; it's unnecessary since
Tom Lane [Mon, 14 May 2007 20:24:41 +0000 (20:24 +0000)]
Get rid of the pg_shdepend entry for a TOAST table; it's unnecessary since
there's an indirect dependency on the owner via the parent table.  We were
already handling indexes that way, but not toast tables for some reason.
Saves a little catalog space and cuts down the verbosity of checkSharedDependencies
reports.

17 years agoFix up grammar and translatability of recent checkSharedDependencies
Tom Lane [Mon, 14 May 2007 20:07:01 +0000 (20:07 +0000)]
Fix up grammar and translatability of recent checkSharedDependencies
patch; also make the code logic a bit more self-consistent.

17 years agoPrevent RevalidateCachedPlan from making any permanent change in
Tom Lane [Mon, 14 May 2007 18:13:21 +0000 (18:13 +0000)]
Prevent RevalidateCachedPlan from making any permanent change in
ActiveSnapshot.  Having it affect ActiveSnapshot only in the unusual
case of needing to replan seems a bad idea, and there's also the problem
that the created snap might be in a relatively short-lived context, as
noted by Jan Wieck.  Also, there's no need to force a new snap at all
unless we are called with no snap currently set, which is an unusual
case in itself.

17 years agoReport all dependent objects to the server log when a shared object is dropped,
Alvaro Herrera [Mon, 14 May 2007 16:50:36 +0000 (16:50 +0000)]
Report all dependent objects to the server log when a shared object is dropped,
and only a truncated log of the objects in the current database to the client.
Also, instead of reporting object counts for all databases on which the user
might own objects, report only as many as fit in the predefined line count.

This is to avoid flooding the client when the user owns too many objects,
which could cause problems.

Per report from Ed L. on April 4th and subsequent discussion.

17 years agoUpdate instructions on creating TODO entry.
Bruce Momjian [Mon, 14 May 2007 16:38:42 +0000 (16:38 +0000)]
Update instructions on creating TODO entry.

17 years agoLog directory we're installing into.
Magnus Hagander [Sun, 13 May 2007 15:33:07 +0000 (15:33 +0000)]
Log directory we're installing into.

Per request from Andrew Dunstan.

17 years agoMark as done, add URL for other item:
Bruce Momjian [Sun, 13 May 2007 11:22:04 +0000 (11:22 +0000)]
Mark as done, add URL for other item:

<  o Add support for arrays of complex types
>
>    http://archives.postgresql.org/pgsql-patches/2007-05/msg00114.php
>
>  o -Add support for arrays of complex types

17 years agoAdd:
Bruce Momjian [Sat, 12 May 2007 22:10:40 +0000 (22:10 +0000)]
Add:

> * Have configure choose integer datetimes by default
>
>   http://archives.postgresql.org/pgsql-patches/2007-05/msg00046.php

17 years agoAdd:
Bruce Momjian [Sat, 12 May 2007 22:06:42 +0000 (22:06 +0000)]
Add:

>  o Allow data to be passed in native language formats, rather
>    than only text
>    http://archives.postgresql.org/pgsql-hackers/2007-05/msg00289.php

17 years agoImprove predicate_refuted_by_simple_clause() to handle IS NULL and IS NOT NULL
Tom Lane [Sat, 12 May 2007 19:22:35 +0000 (19:22 +0000)]
Improve predicate_refuted_by_simple_clause() to handle IS NULL and IS NOT NULL
more completely.  The motivation for having it understand IS NULL at all was
to allow use of "foo IS NULL" as one of the subsets of a partitioning on
"foo", but as reported by Aleksander Kmetec, it wasn't really getting the job
done.  Backpatch to 8.2 since this is arguably a performance bug.

17 years agoFix the problem that creating a user-defined type named _foo, followed by one
Tom Lane [Sat, 12 May 2007 00:55:00 +0000 (00:55 +0000)]
Fix the problem that creating a user-defined type named _foo, followed by one
named foo, would work but the other ordering would not.  If a user-specified
type or table name collides with an existing auto-generated array name, just
rename the array type out of the way by prepending more underscores.  This
should not create any backward-compatibility issues, since the cases in which
this will happen would have failed outright in prior releases.

Also fix an oversight in the arrays-of-composites patch: ALTER TABLE RENAME
renamed the table's rowtype but not its array type.

17 years agoFix my oversight in enabling domains-of-domains: ALTER DOMAIN ADD CONSTRAINT
Tom Lane [Fri, 11 May 2007 20:17:15 +0000 (20:17 +0000)]
Fix my oversight in enabling domains-of-domains: ALTER DOMAIN ADD CONSTRAINT
needs to check the new constraint against columns of derived domains too.

Also, make it error out if the domain to be modified is used within any
composite-type columns.  Eventually we should support that case, but it seems
a bit painful, and not suitable for a back-patch.  For the moment just let the
user know we can't do it.

Backpatch to 8.2, which is the only released version that allows nested
domains.  Possibly the other part should be back-patched further.

17 years agoImprovements to the SGML docs for TRUNCATE and CLUSTER.
Neil Conway [Fri, 11 May 2007 19:40:08 +0000 (19:40 +0000)]
Improvements to the SGML docs for TRUNCATE and CLUSTER.

17 years agoSupport arrays of composite types, including the rowtypes of regular tables
Tom Lane [Fri, 11 May 2007 17:57:14 +0000 (17:57 +0000)]
Support arrays of composite types, including the rowtypes of regular tables
and views (but not system catalogs, nor sequences or toast tables).  Get rid
of the hardwired convention that a type's array type is named exactly "_type",
instead using a new column pg_type.typarray to provide the linkage.  (It still
will be named "_type", though, except in odd corner cases such as
maximum-length type names.)

Along the way, make tracking of owner and schema dependencies for types more
uniform: a type directly created by the user has these dependencies, while a
table rowtype or auto-generated array type does not have them, but depends on
its parent object instead.

David Fetter, Andrew Dunstan, Tom Lane

17 years agoAdded some more error logging.
Michael Meskes [Thu, 10 May 2007 14:29:21 +0000 (14:29 +0000)]
Added some more error logging.

17 years ago- Synced parser and keyword list
Michael Meskes [Thu, 10 May 2007 09:53:17 +0000 (09:53 +0000)]
- Synced parser and keyword list
- Renamed update test so it hopefully runs on Vista

17 years agoReserve some pg_statistic "kind" codes for use by the ESRI ST_Geometry
Tom Lane [Tue, 8 May 2007 19:13:52 +0000 (19:13 +0000)]
Reserve some pg_statistic "kind" codes for use by the ESRI ST_Geometry
datatype project.  Per request from Ale Raza (araza at esri.com).

17 years agoAdd a hash function for "numeric". Mark the equality operator for
Neil Conway [Tue, 8 May 2007 18:56:48 +0000 (18:56 +0000)]
Add a hash function for "numeric". Mark the equality operator for
numerics as "oprcanhash", and make the corresponding system catalog
updates. As a result, hash indexes, hashed aggregation, and hash
joins can now be used with the numeric type. Bump the catversion.

The only tricky aspect to doing this is writing a correct hash
function: it's possible for two Numerics to be equal according to
their equality operator, but have different in-memory bit patterns.
To cope with this, the hash function doesn't consider the Numeric's
"scale" or "sign", and explictly skips any leading or trailing
zeros in the Numeric's digit buffer (the current implementation
should suppress any such zeros, but it seems unwise to rely upon
this). See discussion on pgsql-patches for more details.

17 years agoAdd an explicit comment about POSIX time zone names having the reverse
Tom Lane [Tue, 8 May 2007 17:02:59 +0000 (17:02 +0000)]
Add an explicit comment about POSIX time zone names having the reverse
sign convention from everyplace else in Postgres.  I don't suppose that
this will stop people from being confused, but at least we can say that
it's documented.

17 years agoThe appended patch addresses the outstanding issues of the recent guc patch.
Peter Eisentraut [Tue, 8 May 2007 16:33:51 +0000 (16:33 +0000)]
The appended patch addresses the outstanding issues of the recent guc patch.
It makes PGCLIENTENCODING work again and uses bsearch() instead of
iterating over the array of guc variables in guc_get_index().

Joachim Wieland

17 years agoMake sure we don't skip databases that are supposed to be vacuumed "exactly
Alvaro Herrera [Mon, 7 May 2007 20:41:24 +0000 (20:41 +0000)]
Make sure we don't skip databases that are supposed to be vacuumed "exactly
now".  This can happen if the time granularity is not very high.

Per ITAGAKI Takahiro.

17 years agoGrammar correction
Peter Eisentraut [Mon, 7 May 2007 07:53:26 +0000 (07:53 +0000)]
Grammar correction

17 years agoDoc tweak: add an xref.
Neil Conway [Mon, 7 May 2007 02:03:58 +0000 (02:03 +0000)]
Doc tweak: add an xref.

17 years agoCheck return code from strxfrm on Windows since it has a
Magnus Hagander [Sat, 5 May 2007 17:05:48 +0000 (17:05 +0000)]
Check return code from strxfrm on Windows since it has a
non-standard way of indicating errors, so we don't try to
allocate INT_MAX bytes to store a result in.

17 years agoDone:
Bruce Momjian [Sat, 5 May 2007 15:40:01 +0000 (15:40 +0000)]
Done:

< Last updated: Sat May  5 10:47:39 EDT 2007
> Last updated: Sat May  5 11:39:57 EDT 2007
< * Flush cached query plans when the dependent objects change,
<   when the cardinality of parameters changes dramatically, or
> * -Flush cached query plans when the dependent objects change or
<
<   A more complex solution would be to save multiple plans for different
<   cardinality and use the appropriate plan based on the EXECUTE values.
<
< * Track dependencies in function bodies and recompile/invalidate
<
<   This is particularly important for references to temporary tables
<   in PL/PgSQL because PL/PgSQL caches query plans.  The only workaround
<   in PL/PgSQL is to use EXECUTE.  One complexity is that a function
<   might itself drop and recreate dependent tables, causing it to
<   invalidate its own query plan.
<
< * Invalidate prepared queries, like INSERT, when the table definition
> * -Track dependencies in function bodies and recompile/invalidate
> * -Invalidate prepared queries, like INSERT, when the table definition

17 years agoMove item:
Bruce Momjian [Sat, 5 May 2007 14:47:45 +0000 (14:47 +0000)]
Move item:

< * Invalidate prepared queries, like INSERT, when the table definition
<   is altered
>
> * Invalidate prepared queries, like INSERT, when the table definition
>   is altered

17 years agoDone:
Bruce Momjian [Sat, 5 May 2007 14:44:15 +0000 (14:44 +0000)]
Done:

>  o -Allow multiple vacuums so large tables do not starve small

17 years agoWording update to FAQ_DEV..
Bruce Momjian [Sat, 5 May 2007 14:33:55 +0000 (14:33 +0000)]
Wording update to FAQ_DEV..

17 years agoAdd note to FAQ_DEV that regression tests might need to be added.
Bruce Momjian [Sat, 5 May 2007 14:31:16 +0000 (14:31 +0000)]
Add note to FAQ_DEV that regression tests might need to be added.

17 years agoIn developer's FAQ, update list API, from Tom Lane.
Bruce Momjian [Sat, 5 May 2007 10:21:13 +0000 (10:21 +0000)]
In developer's FAQ, update list API, from Tom Lane.

17 years agoAdd FAQ item about how patches are reviewed.
Bruce Momjian [Sat, 5 May 2007 04:09:25 +0000 (04:09 +0000)]
Add FAQ item about how patches are reviewed.

17 years agoDone:
Bruce Momjian [Sat, 5 May 2007 03:14:40 +0000 (03:14 +0000)]
Done:

> * -Allow ORDER BY ... LIMIT # to select high/low value without sort or
<
<   Right now, if no index exists, ORDER BY ... LIMIT # requires we sort
<   all values to return the high/low value.  Instead The idea is to do a
<   sequential scan to find the high/low value, thus avoiding the sort.
<   MIN/MAX already does this, but not for LIMIT > 1.
<

17 years agoAdd a line to the EXPLAIN ANALYZE output for a Sort node, showing the
Tom Lane [Fri, 4 May 2007 21:29:53 +0000 (21:29 +0000)]
Add a line to the EXPLAIN ANALYZE output for a Sort node, showing the
actual sort strategy and amount of space used.  By popular demand.

17 years agotas() support for Renesas' M32R processor. Kazuhiro Inaoka
Tom Lane [Fri, 4 May 2007 15:20:52 +0000 (15:20 +0000)]
tas() support for Renesas' M32R processor.  Kazuhiro Inaoka

17 years agoMake clearer how arguments and return values in pl/perl are escaped. This is to clari...
Andrew Dunstan [Fri, 4 May 2007 14:55:32 +0000 (14:55 +0000)]
Make clearer how arguments and return values in pl/perl are escaped. This is to clarify the situation that Theo Schlossnagle recently reported on -bugs.

17 years agoSuppress a recently-introduced 'variable might be clobbered by longjmp' warning.
Tom Lane [Fri, 4 May 2007 02:06:13 +0000 (02:06 +0000)]
Suppress a recently-introduced 'variable might be clobbered by longjmp' warning.

17 years agoA few fixups in error handling: mark pg_re_throw() as noreturn for gcc,
Tom Lane [Fri, 4 May 2007 02:01:02 +0000 (02:01 +0000)]
A few fixups in error handling: mark pg_re_throw() as noreturn for gcc,
and for other compilers, insert a dummy exit() call so that they understand
PG_RE_THROW() doesn't return.  Insert fflush(stderr) in ExceptionalCondition,
per recent buildfarm evidence that that might not happen automatically on some
platforms.  And const-ify ExceptionalCondition's declaration while at it.

17 years agoTeach tuplesort.c about "top N" sorting, in which only the first N tuples
Tom Lane [Fri, 4 May 2007 01:13:45 +0000 (01:13 +0000)]
Teach tuplesort.c about "top N" sorting, in which only the first N tuples
need be returned.  We keep a heap of the current best N tuples and sift-up
new tuples into it as we scan the input.  For M input tuples this means
only about M*log(N) comparisons instead of M*log(M), not to mention a lot
less workspace when N is small --- avoiding spill-to-disk for large M
is actually the most attractive thing about it.  Patch includes planner
and executor support for invoking this facility in ORDER BY ... LIMIT
queries.  Greg Stark, with some editorialization by moi.

17 years agoTweak hash index AM to use the new ReadOrZeroBuffer bufmgr API when fetching
Tom Lane [Thu, 3 May 2007 16:45:58 +0000 (16:45 +0000)]
Tweak hash index AM to use the new ReadOrZeroBuffer bufmgr API when fetching
pages it intends to zero immediately.  Just to show there is some use for that
function besides WAL recovery :-).
Along the way, fold _hash_checkpage and _hash_pageinit calls into _hash_getbuf
and friends, instead of expecting callers to do that separately.

17 years agoFix a few more double words in docs.
Alvaro Herrera [Thu, 3 May 2007 15:47:48 +0000 (15:47 +0000)]
Fix a few more double words in docs.

17 years agoFix some typos in the documentation. Patch from Brian Gough. Backport
Neil Conway [Thu, 3 May 2007 15:05:56 +0000 (15:05 +0000)]
Fix some typos in the documentation. Patch from Brian Gough. Backport
the relevant fixes to 8.2 as well.

17 years agoRelease builds generate different strangely formatted export names
Magnus Hagander [Thu, 3 May 2007 14:04:03 +0000 (14:04 +0000)]
Release builds generate different strangely formatted export names
for local symbols, that shouldn't be exported. This patch excludes them,
cutting down about 10,000 exported symbols and decreasing the binary size
by 20%.

17 years agoDept. of second thoughts: add comments cautioning against using
Tom Lane [Wed, 2 May 2007 23:34:48 +0000 (23:34 +0000)]
Dept. of second thoughts: add comments cautioning against using
ReadOrZeroBuffer to fetch pages from beyond physical EOF.  This would
usually work, but would cause problems for md.c if writes occurred
beyond a segment boundary when the previous segment file hadn't been
fully extended.

17 years agoDuring WAL recovery, when reading a page that we intend to overwrite completely
Tom Lane [Wed, 2 May 2007 23:18:03 +0000 (23:18 +0000)]
During WAL recovery, when reading a page that we intend to overwrite completely
from the WAL data, don't bother to physically read it; just have bufmgr.c
return a zeroed-out buffer instead.  This speeds recovery significantly,
and also avoids unnecessary failures when a page-to-be-overwritten has corrupt
page headers on disk.  This replaces a former kluge that accomplished the
latter by pretending zero_damaged_pages was always ON during WAL recovery;
which was OK when the kluge was put in, but is unsafe when restoring a WAL
log that was written with full_page_writes off.

Heikki Linnakangas

17 years agoFix things so that when CREATE INDEX CONCURRENTLY sets pg_index.indisvalid
Tom Lane [Wed, 2 May 2007 21:08:46 +0000 (21:08 +0000)]
Fix things so that when CREATE INDEX CONCURRENTLY sets pg_index.indisvalid
true at the very end of its processing, the update is broadcast via a
shared-cache-inval message for the index; without this, existing backends that
already have relcache entries for the index might never see it become valid.
Also, force a relcache inval on the index's parent table at the same time,
so that any cached plans for that table are re-planned; this ensures that
the newly valid index will be used if appropriate.  Aside from making
C.I.C. behave more reasonably, this is necessary infrastructure for some
aspects of the HOT patch.  Pavan Deolasee, with a little further stuff from
me.

17 years agoUse the new TimestampDifferenceExceeds API instead of timestamp_cmp_internal
Alvaro Herrera [Wed, 2 May 2007 18:27:57 +0000 (18:27 +0000)]
Use the new TimestampDifferenceExceeds API instead of timestamp_cmp_internal
and TimestampDifference, to make coding clearer.  I think this should also fix
the failure to start workers in platforms with low resolution timers, as
reported by Itagaki Takahiro.

17 years agoFix failure to check for INVALID worker entry in the new autovacuum code, which
Alvaro Herrera [Wed, 2 May 2007 15:47:14 +0000 (15:47 +0000)]
Fix failure to check for INVALID worker entry in the new autovacuum code, which
could happen when a worker took to long to start and was thus "aborted" by the
launcher.  Noticed by lionfish buildfarm member.

17 years agoFix oversight in PG_RE_THROW processing: it's entirely possible that there
Tom Lane [Wed, 2 May 2007 15:32:42 +0000 (15:32 +0000)]
Fix oversight in PG_RE_THROW processing: it's entirely possible that there
isn't any place to throw the error to.  If so, we should treat the error
as FATAL, just as we would have if it'd been thrown outside the PG_TRY
block to begin with.

Although this is clearly a *potential* source of bugs, it is not clear
at the moment whether it is an *actual* source of bugs; there may not
presently be any PG_TRY blocks in code that can be reached with no outer
longjmp catcher.  So for the moment I'm going to be conservative and not
back-patch this.  The change breaks ABI for users of PG_RE_THROW and hence
might create compatibility problems for loadable modules, so we should not
put it into released branches without proof that it's needed.

17 years agoAdd wrapper script around install.pl that calls buildenv.bat before doing
Magnus Hagander [Tue, 1 May 2007 20:11:14 +0000 (20:11 +0000)]
Add wrapper script around install.pl that calls buildenv.bat before doing
the install.

Dave Page

17 years agoFix a thinko in my patch of a couple months ago for bug #3116: it did the
Tom Lane [Tue, 1 May 2007 18:53:52 +0000 (18:53 +0000)]
Fix a thinko in my patch of a couple months ago for bug #3116: it did the
wrong thing when inlining polymorphic SQL functions, because it was using the
function's declared return type where it should have used the actual result
type of the current call.  In 8.1 and 8.2 this causes obvious failures even if
you don't have assertions turned on; in 8.0 and 7.4 it would only be a problem
if the inlined expression were used as an input to a function that did
run-time type determination on its inputs.  Add a regression test, since this
is evidently an under-tested area.

17 years agoChange the timestamps recorded in transaction commit/abort xlog records
Tom Lane [Mon, 30 Apr 2007 21:01:53 +0000 (21:01 +0000)]
Change the timestamps recorded in transaction commit/abort xlog records
from time_t to TimestampTz representation.  This provides full gettimeofday()
resolution of the timestamps, which might be useful when attempting to
do point-in-time recovery --- previously it was not possible to specify
the stop point with sub-second resolution.  But mostly this is to get
rid of TimestampTz-to-time_t conversion overhead during commit.  Per my
proposal of a day or two back.

17 years agoFix oversight in my patch of yesterday: forgot to ensure that stats would
Tom Lane [Mon, 30 Apr 2007 16:37:08 +0000 (16:37 +0000)]
Fix oversight in my patch of yesterday: forgot to ensure that stats would
still be forced out at backend exit.

17 years agoImplement rate-limiting logic on how often backends will attempt to send
Tom Lane [Mon, 30 Apr 2007 03:23:49 +0000 (03:23 +0000)]
Implement rate-limiting logic on how often backends will attempt to send
messages to the stats collector.  This avoids the problem that enabling
stats_row_level for autovacuum has a significant overhead for short
read-only transactions, as noted by Arjen van der Meijden.  We can avoid
an extra gettimeofday call by piggybacking on the one done for WAL-logging
xact commit or abort (although that doesn't help read-only transactions,
since they don't WAL-log anything).

In my proposal for this, I noted that we could change the WAL log entries
for commit/abort to record full TimestampTz precision, instead of only
time_t as at present.  That's not done in this patch, but will be committed
separately.

17 years agoMarginal performance hack: use a dedicated routine instead of copyObject
Tom Lane [Mon, 30 Apr 2007 00:16:43 +0000 (00:16 +0000)]
Marginal performance hack: use a dedicated routine instead of copyObject
to copy nodes that are known to be Vars during plan reference adjustment.
Saves useless memzero operation as well as the big switch in copyObject.

17 years agoMarginal performance hack: avoid unnecessary work in expression_tree_mutator.
Tom Lane [Mon, 30 Apr 2007 00:14:54 +0000 (00:14 +0000)]
Marginal performance hack: avoid unnecessary work in expression_tree_mutator.
We can just palloc, instead of using makeNode, when we are going to
overwrite the whole node anyway in the FLATCOPY macro.  Also, use
FLATCOPY instead of copyObject for common node types Var and Const.

17 years agoMarginal performance hack: remove the loop that used to be needed to
Tom Lane [Mon, 30 Apr 2007 00:12:08 +0000 (00:12 +0000)]
Marginal performance hack: remove the loop that used to be needed to
look through a freelist for a chunk of adequate size.  For a long time
now, all elements of a given freelist have been exactly the same
allocated size, so we don't need a loop.  Since the loop never iterated
more than once, you'd think this wouldn't matter much, but it makes a
noticeable savings in a simple test --- perhaps because the compiler
isn't optimizing on a mistaken assumption that the loop would repeat.
AllocSetAlloc is called often enough that saving even a couple of
instructions is worthwhile.

17 years agoUpdate to 8.4:
Bruce Momjian [Sun, 29 Apr 2007 07:17:12 +0000 (07:17 +0000)]
Update to 8.4:

< * Make standard_conforming_strings the default in 8.3?
> * Make standard_conforming_strings the default in 8.4?

17 years agoPl/pgsql MOVE done:
Bruce Momjian [Sun, 29 Apr 2007 06:48:11 +0000 (06:48 +0000)]
Pl/pgsql MOVE done:

<  o Add support for MOVE and SCROLL cursors
<
<    PL/pgSQL cursors should support the same syntax as
<    backend cursors.
<
>  o -Add support for MOVE cursors
>  o Add support for SCROLL cursors