]> granicus.if.org Git - postgresql/log
postgresql
24 years agoAdd numeric <-> int8 and numeric <-> int2 conversion functions, as well
Tom Lane [Thu, 24 Feb 2000 02:05:30 +0000 (02:05 +0000)]
Add numeric <-> int8 and numeric <-> int2 conversion functions, as well
as a unary minus operator for numeric.  Now that long numeric constants
will get converted to NUMERIC in early parsing, it's essential to have
numeric->int8 conversion to avoid 'can't convert' errors on undecorated
int8 constants.  Threw in the rest for completeness while I was in the
area.
I did not force an initdb for this, since the system will still run
without the new pg_proc/pg_operator entries.  Possibly I should've.

24 years agoMake make_const() check the size and precision of a T_Float Value,
Tom Lane [Thu, 24 Feb 2000 01:59:17 +0000 (01:59 +0000)]
Make make_const() check the size and precision of a T_Float Value,
and produce either FLOAT8 or NUMERIC output depending on whether the
value fits in a float8 or not.  This is almost back to the way the
code was before I changed T_Float, but there is a critical difference:
now, when a numeric constant doesn't fit in float8, it will be treated
as type NUMERIC instead of type UNKNOWN.

24 years agoint8in failed to detect overflow; it really should.
Tom Lane [Thu, 24 Feb 2000 01:54:40 +0000 (01:54 +0000)]
int8in failed to detect overflow; it really should.

24 years agoFix plpsql for bsdi.
Bruce Momjian [Wed, 23 Feb 2000 22:24:46 +0000 (22:24 +0000)]
Fix plpsql for bsdi.

24 years ago*** empty log message ***
Michael Meskes [Wed, 23 Feb 2000 19:26:05 +0000 (19:26 +0000)]
*** empty log message ***

24 years ago1. miscadmin.h needs to include sys/types.h for a definition of pid_t
Bruce Momjian [Wed, 23 Feb 2000 15:46:15 +0000 (15:46 +0000)]
1. miscadmin.h needs to include sys/types.h for a definition of pid_t

2. Regression tests fail for types int2 and int4 (which can easily be
fixed by adding entries to resultmap) aswell as float8 and geometry,
where floating point numbers appear to be rounded a little differently
than in your expected results (besides that I also need the positive
zeros file). I'm including a patch for the first 2, but I don't know
whether the latter two are actually a bug in postgres or a bug in the
OS or even allowed difference. I'm including my results for reference.

Rolf Grossmann

24 years agoWell, here's the first pass on regression
Bruce Momjian [Tue, 22 Feb 2000 20:58:31 +0000 (20:58 +0000)]
Well, here's the first pass on regression
tests for the Foreign Key support in 7.0 which
was made against a CVS copy from this
afternoon.

This modifies
 src/test/regress/sql/run_check.tests
 src/test/regress/sql/alter_table.sql
 src/test/regress/expected/alter_table.out
 src/test/regress/sql/foreign_key.sql
 src/test/regress/expected/foreign_key.out

sszabo@bigpanda.co

24 years ago*** empty log message ***
Michael Meskes [Tue, 22 Feb 2000 19:57:12 +0000 (19:57 +0000)]
*** empty log message ***

24 years agoUpdate TODO list.
Bruce Momjian [Tue, 22 Feb 2000 14:16:36 +0000 (14:16 +0000)]
Update TODO list.

24 years agoUpdate TODO list.
Bruce Momjian [Tue, 22 Feb 2000 14:08:12 +0000 (14:08 +0000)]
Update TODO list.

24 years agoChange cancel while waiting-for-lock stuff.
Hiroshi Inoue [Tue, 22 Feb 2000 09:55:26 +0000 (09:55 +0000)]
Change cancel while waiting-for-lock stuff.

24 years agoUpdate TODO list.
Bruce Momjian [Tue, 22 Feb 2000 00:58:51 +0000 (00:58 +0000)]
Update TODO list.

24 years agoFirst post-beta1 bug fix :-(. Silly typo in new coding for doNegate
Tom Lane [Tue, 22 Feb 2000 00:05:04 +0000 (00:05 +0000)]
First post-beta1 bug fix :-(.  Silly typo in new coding for doNegate
failed to negate a negative value back to positive, so '- - 123.45'
did the wrong thing.

24 years agoQuick hack solution so that pg_dump of views works. Needs repair after
Tom Lane [Mon, 21 Feb 2000 20:18:11 +0000 (20:18 +0000)]
Quick hack solution so that pg_dump of views works.  Needs repair after
Thomas gets back, but better this than nonfunctional pg_dump in the beta.

24 years agoFixed psql's Control-C handling when COPY in progress
Peter Eisentraut [Mon, 21 Feb 2000 19:40:42 +0000 (19:40 +0000)]
Fixed psql's Control-C handling when COPY in progress

24 years agoOops, commited a test version of this file by accident. Revert.
Tom Lane [Mon, 21 Feb 2000 18:49:54 +0000 (18:49 +0000)]
Oops, commited a test version of this file by accident.  Revert.

24 years agoChange parse-time representation of float literals (which include oversize
Tom Lane [Mon, 21 Feb 2000 18:47:12 +0000 (18:47 +0000)]
Change parse-time representation of float literals (which include oversize
integers) to be strings instead of 'double'.  We convert from string form
to internal representation only after type resolution has determined the
correct type for the constant.  This eliminates loss-of-precision worries
and gets rid of the change in behavior seen at 17 digits with the
previous kluge.

24 years agoTo avoid confusion during early beta testing, commit the current rules
Tom Lane [Mon, 21 Feb 2000 18:42:35 +0000 (18:42 +0000)]
To avoid confusion during early beta testing, commit the current rules
test output as expected output.  We'll probably want to change this again
after something's done about the verbosity of column alias display.

24 years agoUpdate TODO list.
Bruce Momjian [Mon, 21 Feb 2000 17:06:26 +0000 (17:06 +0000)]
Update TODO list.

24 years agoClean up temp file in 'make clean'.
Tom Lane [Mon, 21 Feb 2000 16:46:39 +0000 (16:46 +0000)]
Clean up temp file in 'make clean'.

24 years agoAvoid race conditions in detection of EINPROGRESS during connect().
Jan Wieck [Mon, 21 Feb 2000 12:26:19 +0000 (12:26 +0000)]
Avoid race conditions in detection of EINPROGRESS during connect().

Jan

24 years agoFix "Invalid XID in t_cmin" error in vacuum.
Hiroshi Inoue [Mon, 21 Feb 2000 07:49:40 +0000 (07:49 +0000)]
Fix "Invalid XID in t_cmin" error in vacuum.

24 years agoClean up some really grotty coding in catcache.c, improve hashing
Tom Lane [Mon, 21 Feb 2000 03:36:59 +0000 (03:36 +0000)]
Clean up some really grotty coding in catcache.c, improve hashing
performance in catcache lookups.

24 years agofix the TODO
Hiroshi Inoue [Mon, 21 Feb 2000 02:42:37 +0000 (02:42 +0000)]
fix the TODO
* Allow PQrequestCancel() to terminate when in waiting-for-lock state
Changes are limited to BACKEND,however.

24 years agoFixes for \encoding command.
Tatsuo Ishii [Mon, 21 Feb 2000 02:05:12 +0000 (02:05 +0000)]
Fixes for \encoding command.
1) freeing null pointer
2) invalid encoding info may be stored into psql variable
3) fix indentation

24 years agoUpdate discussion of aggregate syntax to mention ALL.
Tom Lane [Mon, 21 Feb 2000 01:20:41 +0000 (01:20 +0000)]
Update discussion of aggregate syntax to mention ALL.

24 years agoUpdate SELECT documentation to have a more thorough discussion
Tom Lane [Mon, 21 Feb 2000 01:13:52 +0000 (01:13 +0000)]
Update SELECT documentation to have a more thorough discussion
of LIMIT, and fix a few other glitches too.

24 years agoTweak planner to use OFFSET+LIMIT, not just LIMIT, as estimate of the
Tom Lane [Mon, 21 Feb 2000 01:13:04 +0000 (01:13 +0000)]
Tweak planner to use OFFSET+LIMIT, not just LIMIT, as estimate of the
portion of the query result that will be retrieved.  As far as I could
tell, the consensus was that we should let the planner do the best it
can with a LIMIT query, and require the user to add ORDER BY if he
wants consistent results from different LIMIT values.

24 years agoFurther cleanups for type coercion: treat the locution typename(argument)
Tom Lane [Sun, 20 Feb 2000 23:04:06 +0000 (23:04 +0000)]
Further cleanups for type coercion: treat the locution typename(argument)
as representing a type coercion request in more cases than we did before.
It will work now whenever no underlying function is required, ie if the
coercion is binary-compatible or if the argument is a previously untyped
string constant.  Otherwise, you still need a real function to exist.

24 years agoCreate a new expression node type RelabelType, which exists solely to
Tom Lane [Sun, 20 Feb 2000 21:32:16 +0000 (21:32 +0000)]
Create a new expression node type RelabelType, which exists solely to
represent the result of a binary-compatible type coercion.  At runtime
it just evaluates its argument --- but during type resolution, exprType
will pick up the output type of the RelabelType node instead of the type
of the argument.  This solves some longstanding problems with dropped
type coercions, an example being 'select now()::abstime::int4' which
used to produce date-formatted output, not an integer, because the
coercion to int4 was dropped on the floor.

24 years agoForgot to include the documentation for \encoding ...
Peter Eisentraut [Sun, 20 Feb 2000 14:29:21 +0000 (14:29 +0000)]
Forgot to include the documentation for \encoding ...

24 years agoMoved psql \eset and \eshow to \encoding
Peter Eisentraut [Sun, 20 Feb 2000 14:28:28 +0000 (14:28 +0000)]
Moved psql \eset and \eshow to \encoding
Improved psql's Ctrl-C handling
Fixed configure test for sigsetjmp that now even recognizes it if it's a macro

24 years agoFix broken list-slinging logic in func_select_candidate and
Tom Lane [Sun, 20 Feb 2000 06:35:08 +0000 (06:35 +0000)]
Fix broken list-slinging logic in func_select_candidate and
agg_select_candidate, which could cause them to keep more candidates
than they should and thus fail to select a single match.  I had
previously fixed the identical bug in oper_select_candidate, but
didn't realize that the same error was repeated over here.
Also, repair func_select_candidate's curious notion that it could
scribble on the input type-OID vector.  That was causing failure to
apply necessary type coercion later on, leading to malfunction of
examples such as select date('now').

24 years agoEven after the great date/time consolidation, TypeCategory() was still
Tom Lane [Sun, 20 Feb 2000 06:28:42 +0000 (06:28 +0000)]
Even after the great date/time consolidation, TypeCategory() was still
a few bricks shy of a load concerning knowing all the date/time types.
This is real bad because it interferes with func_select_candidate()'s
willingness to disambiguate functions --- func_select_candidate() will
punt unless all the available choices have the same type category.
I think this whole mechanism needs redesigned, but in the meantime
this is a needed patch.

24 years agoReduce proc_exit(1) to proc_exit(0) for errors detected in backend
Tom Lane [Sun, 20 Feb 2000 04:26:35 +0000 (04:26 +0000)]
Reduce proc_exit(1) to proc_exit(0) for errors detected in backend
command line processing.  As it stood, a bogus PGOPTIONS value from
a client would force a database system restart.  Not bad as a denial-
of-service attack...

24 years agoFix missing \n in some psql_error calls.
Tom Lane [Sun, 20 Feb 2000 02:37:40 +0000 (02:37 +0000)]
Fix missing \n in some psql_error calls.

24 years agoAdd a simple regress test for SERIAL --- it's not much,
Tom Lane [Sun, 20 Feb 2000 02:16:02 +0000 (02:16 +0000)]
Add a simple regress test for SERIAL --- it's not much,
but it's better than no test at all...

24 years agoFix SERIAL columns, which Thomas inadvertently broke parsing of.
Tom Lane [Sun, 20 Feb 2000 02:14:58 +0000 (02:14 +0000)]
Fix SERIAL columns, which Thomas inadvertently broke parsing of.
Revised code probably accepts some silly combinations, but that's better
than not accepting valid ones.

24 years agoRepair longstanding violation of SQL92 semantics: GROUP BY would
Tom Lane [Sat, 19 Feb 2000 23:45:07 +0000 (23:45 +0000)]
Repair longstanding violation of SQL92 semantics: GROUP BY would
interpret a column name as an output column alias (targetlist AS name),
ather than a real column name as it ought to.  According to the spec,
only ORDER BY should look at output column names.  I left in GROUP BY's
willingness to use an output column number ('GROUP BY 2'), even though
this is also contrary to the spec --- again, only ORDER BY is supposed
to accept that.  But there is no possible reason to want to GROUP BY
an integer constant, so keeping this old behavior won't break any
SQL-compliant queries.  DISTINCT ON will behave the same as GROUP BY.

Change numerology regress test, which depended on the incorrect
behavior.

24 years agoApply Keith Park's updates for expected/horology-solaris-1947.out. Fix
Tom Lane [Sat, 19 Feb 2000 22:13:37 +0000 (22:13 +0000)]
Apply Keith Park's updates for expected/horology-solaris-1947.out.  Fix
erroneous expected output for RESET DateStyle: should be ISO now.
Fix run_check.sh so that test postmaster is started with PGDATESTYLE=ISO,
else the horology test won't pass.

24 years agoGet rid of postgres.c's separate parsing logic for PGDATESTYLE env.
Tom Lane [Sat, 19 Feb 2000 22:10:47 +0000 (22:10 +0000)]
Get rid of postgres.c's separate parsing logic for PGDATESTYLE env.
variable, instead calling same code in variable.c that is used to parse
SET DATESTYLE.  Fix bug: although backend's startup datestyle had been
changed to ISO, 'RESET DATESTYLE' and 'SET DATESTYLE TO DEFAULT' didn't
know about it.  For consistency I have made the latter two reset to the
PGDATESTYLE-defined initial value, which may not be the same as the
compiled-in default of ISO.

24 years agoAccept the noise-word ALL in aggregate function invocations for SQL92
Tom Lane [Sat, 19 Feb 2000 19:37:21 +0000 (19:37 +0000)]
Accept the noise-word ALL in aggregate function invocations for SQL92
compliance.  Wish they were all that easy...

24 years agoplperl's makefile tried to use perl's choice of compiler with
Tom Lane [Sat, 19 Feb 2000 19:02:20 +0000 (19:02 +0000)]
plperl's makefile tried to use perl's choice of compiler with
postgres's choice of compiler options.  Tres uncool.

24 years agoHP's compiler objects (quite rightly too) to 'static void foo();'
Tom Lane [Sat, 19 Feb 2000 18:58:37 +0000 (18:58 +0000)]
HP's compiler objects (quite rightly too) to 'static void foo();'
followed by 'extern void foo() { ... }'.

24 years agoFirst workaround solution for Jan's column constraint attributes.
Thomas G. Lockhart [Sat, 19 Feb 2000 08:25:49 +0000 (08:25 +0000)]
First workaround solution for Jan's column constraint attributes.

24 years agoClean up markup so docs will compile.
Thomas G. Lockhart [Sat, 19 Feb 2000 08:14:54 +0000 (08:14 +0000)]
Clean up markup so docs will compile.

24 years agoUpdate for new Irix info.
Thomas G. Lockhart [Sat, 19 Feb 2000 08:14:33 +0000 (08:14 +0000)]
Update for new Irix info.

24 years agoFix nasty bug in setting client_encoding
Tatsuo Ishii [Sat, 19 Feb 2000 05:04:54 +0000 (05:04 +0000)]
Fix nasty bug in setting client_encoding

24 years agoAdd new backslash command \eset and \eshow.
Tatsuo Ishii [Sat, 19 Feb 2000 05:01:16 +0000 (05:01 +0000)]
Add new backslash command \eset and \eshow.

24 years agoFix compiler warning
Tatsuo Ishii [Sat, 19 Feb 2000 04:59:30 +0000 (04:59 +0000)]
Fix compiler warning

24 years agoClean up scan.l's handling of \r vs \n --- they are reliably treated as
Tom Lane [Sat, 19 Feb 2000 04:17:25 +0000 (04:17 +0000)]
Clean up scan.l's handling of \r vs \n --- they are reliably treated as
equivalent now, which should make Windows and Mac clients happier.
Also fix failure to handle SQL comments between segments of a multiline
quoted literal.

24 years agoRemove gcc warnings. The Postgres world isn't really safe
Tom Lane [Sat, 19 Feb 2000 02:29:07 +0000 (02:29 +0000)]
Remove gcc warnings.  The Postgres world isn't really safe
for 'const' qualifiers yet ...

24 years agoPlug some more memory leaks in the planner. It still leaks like a sieve,
Tom Lane [Fri, 18 Feb 2000 23:47:31 +0000 (23:47 +0000)]
Plug some more memory leaks in the planner.  It still leaks like a sieve,
but this is as good as it'll get for this release...

24 years ago*** empty log message ***
Michael Meskes [Fri, 18 Feb 2000 16:02:49 +0000 (16:02 +0000)]
*** empty log message ***

24 years ago*** empty log message ***
Michael Meskes [Fri, 18 Feb 2000 14:34:12 +0000 (14:34 +0000)]
*** empty log message ***

24 years agoImplement reindex command
Hiroshi Inoue [Fri, 18 Feb 2000 09:30:20 +0000 (09:30 +0000)]
Implement reindex command

24 years ago*** empty log message ***
Michael Meskes [Fri, 18 Feb 2000 08:49:07 +0000 (08:49 +0000)]
*** empty log message ***

24 years agoReplace inefficient _bt_invokestrat calls with direct calls to the
Tom Lane [Fri, 18 Feb 2000 06:32:39 +0000 (06:32 +0000)]
Replace inefficient _bt_invokestrat calls with direct calls to the
appropriate btree three-way comparison routine.  Not clear why the
three-way comparison routines were being used in some paths and not
others in btree --- incomplete changes by someone long ago, maybe?
Anyway, this makes for a nice speedup in CREATE INDEX.

24 years agoFixed handling of old --with-mb option using newly gained wisdom from
Peter Eisentraut [Fri, 18 Feb 2000 00:21:38 +0000 (00:21 +0000)]
Fixed handling of old --with-mb option using newly gained wisdom from
autoconf mailing list. ;)

24 years ago*** empty log message ***
Michael Meskes [Thu, 17 Feb 2000 19:48:58 +0000 (19:48 +0000)]
*** empty log message ***

24 years agoFix assert bug(was my fault)
Hiroshi Inoue [Thu, 17 Feb 2000 05:00:38 +0000 (05:00 +0000)]
Fix assert bug(was my fault)

24 years agoFinish repairing 6.5's problems with r-tree indexes: create appropriate
Tom Lane [Thu, 17 Feb 2000 03:40:02 +0000 (03:40 +0000)]
Finish repairing 6.5's problems with r-tree indexes: create appropriate
selectivity functions and make the r-tree operators use them.  The
estimation functions themselves are just stubs, unfortunately, but
perhaps someday someone will make them compute realistic estimates.
Change pg_am so that the optimizer can reliably tell the difference
between ordered and unordered indexes --- before it would think that
an r-tree index can be scanned in '<<' order, which is not right AFAIK.
Repair broken negator links for network_sup and related ops.
Initdb forced.  This might be my last initdb force for 7.0 ... hope so
anyway ...

24 years agoUpdate expected/horology-no-DST-before-1970.out to match
Tom Lane [Thu, 17 Feb 2000 02:20:59 +0000 (02:20 +0000)]
Update expected/horology-no-DST-before-1970.out to match
new datetime regress test.  Someone needs to deal with fixing
expected/horology-solaris-1947.out too, assuming we still need it.

24 years agoPackage: postgresql
Bruce Momjian [Wed, 16 Feb 2000 23:07:31 +0000 (23:07 +0000)]
Package: postgresql
Version: 6.5.3-11
Severity: important

'char' is not a signed type by default on powerpc; therefore a character
can
never be equal to EOF (-1).  A patch is attached.

Dan

/--------------------------------\  /--------------------------------\
|       Daniel Jacobowitz        |__|        SCS Class of 2002       |

24 years agoREADME not needed anymore.
Bruce Momjian [Wed, 16 Feb 2000 21:26:00 +0000 (21:26 +0000)]
README not needed anymore.

24 years agoAll regression tests pass except for rules.sql (unrelated).
Thomas G. Lockhart [Wed, 16 Feb 2000 18:17:02 +0000 (18:17 +0000)]
All regression tests pass except for rules.sql (unrelated).
Implement "date/time grand unification".
 Transform datetime and timespan into timestamp and interval.
 Deprecate datetime and timespan, though translate to new types in gram.y.
 Transform all datetime and timespan catalog entries into new types.
 Make "INTERVAL" reserved word allowed as a column identifier in gram.y.
 Remove dt.h, dt.c files, and retarget datetime.h, datetime.c as utility
  routines for all date/time types.
 date.{h,c} now deals with date, time types.
 timestamp.{h,c} now deals with timestamp, interval types.
 nabstime.{h,c} now deals with abstime, reltime, tinterval types.
Make NUMERIC a known native type for purposes of type coersion. Not tested.

24 years agoAll regression tests pass except for rules.sql (unrelated).
Thomas G. Lockhart [Wed, 16 Feb 2000 17:27:27 +0000 (17:27 +0000)]
All regression tests pass except for rules.sql (unrelated).
Implement "date/time grand unification".
 Transform datetime and timespan into timestamp and interval.
 Deprecate datetime and timespan, though translate to new types in gram.y.
 Transform all datetime and timespan catalog entries into new types.
 Make "INTERVAL" reserved word allowed as a column identifier in gram.y.
 Remove dt.h, dt.c files, and retarget datetime.h, datetime.c as utility
  routines for all date/time types.
 date.{h,c} now deals with date, time types.
 timestamp.{h,c} now deals with timestamp, interval types.
 nabstime.{h,c} now deals with abstime, reltime, tinterval types.
Make NUMERIC a known native type for purposes of type coersion. Not tested.

24 years agoImplement "date/time grand unification".
Thomas G. Lockhart [Wed, 16 Feb 2000 17:26:26 +0000 (17:26 +0000)]
Implement "date/time grand unification".
 Transform datetime and timespan into timestamp and interval.
 Deprecate datetime and timespan, though translate to new types in gram.y.
 Transform all datetime and timespan catalog entries into new types.
 Make "INTERVAL" reserved word allowed as a column identifier in gram.y.
 Remove dt.h, dt.c files, and retarget datetime.h, datetime.c as utility
  routines for all date/time types.
 date.{h,c} now deals with date, time types.
 timestamp.{h,c} now deals with timestamp, interval types.
 nabstime.{h,c} now deals with abstime, reltime, tinterval types.
Make NUMERIC a known native type for purposes of type coersion. Not tested.

24 years agoMake ISO date style (e.g. "2000-02-16 09:33") the default.
Thomas G. Lockhart [Wed, 16 Feb 2000 17:25:49 +0000 (17:25 +0000)]
Make ISO date style (e.g. "2000-02-16 09:33") the default.
Implement "date/time grand unification".
 Transform datetime and timespan into timestamp and interval.
 Deprecate datetime and timespan, though translate to new types in gram.y.
 Transform all datetime and timespan catalog entries into new types.
 Make "INTERVAL" reserved word allowed as a column identifier in gram.y.
 Remove dt.h, dt.c files, and retarget datetime.h, datetime.c as utility
  routines for all date/time types.
 date.{h,c} now deals with date, time types.
 timestamp.{h,c} now deals with timestamp, interval types.
 nabstime.{h,c} now deals with abstime, reltime, tinterval types.
Make NUMERIC a known native type for purposes of type coersion. Not tested.

24 years ago*** empty log message ***
Michael Meskes [Wed, 16 Feb 2000 16:18:29 +0000 (16:18 +0000)]
*** empty log message ***

24 years agoinclude cleanup
Bruce Momjian [Wed, 16 Feb 2000 13:22:00 +0000 (13:22 +0000)]
include cleanup

24 years agoClean up include files use in psql.
Bruce Momjian [Wed, 16 Feb 2000 13:15:26 +0000 (13:15 +0000)]
Clean up include files use in psql.

24 years ago*** empty log message ***
Michael Meskes [Wed, 16 Feb 2000 11:52:25 +0000 (11:52 +0000)]
*** empty log message ***

24 years agoIf we don't have any stats for a boolean column, assume
Tom Lane [Wed, 16 Feb 2000 01:00:23 +0000 (01:00 +0000)]
If we don't have any stats for a boolean column, assume
the disbursion is 0.5, not something small.

24 years agoMake eqsel produce better results for boolean columns,
Tom Lane [Wed, 16 Feb 2000 00:59:27 +0000 (00:59 +0000)]
Make eqsel produce better results for boolean columns,
and make scalarltsel a little more forgiving at the boundaries of the
known range of a column value.

24 years agoUpdate EXPLAIN documentation to reflect the fact that the
Tom Lane [Tue, 15 Feb 2000 23:37:49 +0000 (23:37 +0000)]
Update EXPLAIN documentation to reflect the fact that the
planner now produces two cost numbers instead of one.

24 years agoRemove long-dead code.
Tom Lane [Tue, 15 Feb 2000 23:12:26 +0000 (23:12 +0000)]
Remove long-dead code.

24 years agoFix missing lfirst() in ListTableAsAttrs(). This code
Tom Lane [Tue, 15 Feb 2000 23:09:08 +0000 (23:09 +0000)]
Fix missing lfirst() in ListTableAsAttrs().  This code
doesn't seem to be used at the moment, but as long as I'm looking at it...

24 years agoNew cost model for planning, incorporating a penalty for random page
Tom Lane [Tue, 15 Feb 2000 20:49:31 +0000 (20:49 +0000)]
New cost model for planning, incorporating a penalty for random page
accesses versus sequential accesses, a (very crude) estimate of the
effects of caching on random page accesses, and cost to evaluate WHERE-
clause expressions.  Export critical parameters for this model as SET
variables.  Also, create SET variables for the planner's enable flags
(enable_seqscan, enable_indexscan, etc) so that these can be controlled
more conveniently than via PGOPTIONS.

Planner now estimates both startup cost (cost before retrieving
first tuple) and total cost of each path, so it can optimize queries
with LIMIT on a reasonable basis by interpolating between these costs.
Same facility is a win for EXISTS(...) subqueries and some other cases.

Redesign pathkey representation to achieve a major speedup in planning
(I saw as much as 5X on a 10-way join); also minor changes in planner
to reduce memory consumption by recycling discarded Path nodes and
not constructing unnecessary lists.

Minor cleanups to display more-plausible costs in some cases in
EXPLAIN output.

Initdb forced by change in interface to index cost estimation
functions.

24 years agoChange to report COPY line number in error messages affects
Tom Lane [Tue, 15 Feb 2000 19:45:31 +0000 (19:45 +0000)]
Change to report COPY line number in error messages affects
results of constraints regress test.

24 years ago*** empty log message ***
Michael Meskes [Tue, 15 Feb 2000 19:23:20 +0000 (19:23 +0000)]
*** empty log message ***

24 years agoSilence gcc warning about uninitialized var.
Tom Lane [Tue, 15 Feb 2000 18:17:33 +0000 (18:17 +0000)]
Silence gcc warning about uninitialized var.

24 years agoRemoved unused var to silence gcc warning.
Tom Lane [Tue, 15 Feb 2000 18:15:12 +0000 (18:15 +0000)]
Removed unused var to silence gcc warning.

24 years ago*** empty log message ***
Michael Meskes [Tue, 15 Feb 2000 12:15:57 +0000 (12:15 +0000)]
*** empty log message ***

24 years agoRepair bogus rule display of attr lists.
Tom Lane [Tue, 15 Feb 2000 08:24:12 +0000 (08:24 +0000)]
Repair bogus rule display of attr lists.

24 years agoRemove some // comments, which are not ANSI C last I heard.
Tom Lane [Tue, 15 Feb 2000 07:47:37 +0000 (07:47 +0000)]
Remove some // comments, which are not ANSI C last I heard.

24 years agoRename file to be consistant with other platform-specific FAQs.
Thomas G. Lockhart [Tue, 15 Feb 2000 05:22:36 +0000 (05:22 +0000)]
Rename file to be consistant with other platform-specific FAQs.

24 years agoFix extraneous ending tag on libpq.sgml.
Thomas G. Lockhart [Tue, 15 Feb 2000 03:57:02 +0000 (03:57 +0000)]
Fix extraneous ending tag on libpq.sgml.
Renormalize query.sgml and fix indenting.

24 years agoCarry column aliases from the parser frontend. Enables queries like
Thomas G. Lockhart [Tue, 15 Feb 2000 03:38:29 +0000 (03:38 +0000)]
Carry column aliases from the parser frontend. Enables queries like
  SELECT a FROM t1 tx (a);
Allow join syntax, including queries like
  SELECT * FROM t1 NATURAL JOIN t2;
Update RTE structure to hold column aliases in an Attr structure.

24 years agoUpdate "join syntax" test for new capabilities.
Thomas G. Lockhart [Tue, 15 Feb 2000 03:31:33 +0000 (03:31 +0000)]
Update "join syntax" test for new capabilities.

24 years agoAdd line to mention PST8PDT time zone. Seems useful, but maybe not?
Thomas G. Lockhart [Tue, 15 Feb 2000 03:30:57 +0000 (03:30 +0000)]
Add line to mention PST8PDT time zone. Seems useful, but maybe not?

24 years agoFix up error message to start with cap letter.
Thomas G. Lockhart [Tue, 15 Feb 2000 03:30:06 +0000 (03:30 +0000)]
Fix up error message to start with cap letter.

24 years agoFix up tabbing for _int8 entry.
Thomas G. Lockhart [Tue, 15 Feb 2000 03:28:31 +0000 (03:28 +0000)]
Fix up tabbing for _int8 entry.

24 years agoAdd "DEC" as synonym for "DECIMAL".
Thomas G. Lockhart [Tue, 15 Feb 2000 03:26:38 +0000 (03:26 +0000)]
Add "DEC" as synonym for "DECIMAL".
Add "SESSION_USER" as SQL92 keyword; equivalent to CURRENT_USER for now.
Implement column aliases (aka correlation names) and more join syntax.
Fix up indenting and tabbing.

24 years agoRecognize special case of POSIX time zone: "GMT+8" and "GMT-8".
Thomas G. Lockhart [Tue, 15 Feb 2000 03:17:09 +0000 (03:17 +0000)]
Recognize special case of POSIX time zone: "GMT+8" and "GMT-8".
Still needs to be done for the general case:
 "tz+/-#" where tz is a 3 char string.
This will probably involve moving code around to other places.

24 years agoFix up error and log messages.
Thomas G. Lockhart [Tue, 15 Feb 2000 03:00:37 +0000 (03:00 +0000)]
Fix up error and log messages.

24 years agoI suspect that you are not the person to send this to, but I wasn't sure
Bruce Momjian [Mon, 14 Feb 2000 12:33:19 +0000 (12:33 +0000)]
I suspect that you are not the person to send this to, but I wasn't sure
where else to mail it. I am the maintainer of unixODBC, and we have a
set of code in our project that started life as the Postgres windows
ODBC driver, which has been ported back to unix. Anyway I have just
fixed a memory leak in the driver, and I cant see any mention of the fix
being done in the main Postgres code, so I thougth I would let you know.

Its in the statement.c module, after the COMMIT statement has been
executed in SC_Execute, the code was

Nick Gorham

24 years agoUpdate README.NT with new NT ipc location.
Bruce Momjian [Mon, 14 Feb 2000 12:29:58 +0000 (12:29 +0000)]
Update README.NT with new NT ipc location.

24 years ago*** empty log message ***
Michael Meskes [Mon, 14 Feb 2000 10:00:06 +0000 (10:00 +0000)]
*** empty log message ***

24 years agoAdjusted psql echoing options (-a and -e)
Peter Eisentraut [Sun, 13 Feb 2000 21:45:15 +0000 (21:45 +0000)]
Adjusted psql echoing options (-a and -e)