postgresql version 7.3, but yea... this patch adds full IPv6
support to postgres. I've tested it out on 7.2.3 and has
been running perfectly stable.
CREDITS:
The KAME Project (Initial patch)
Nigel Kukard <nkukard@lbsd.net>
Johan Jordaan <johanj@lando.co.za>
cleaning up locale names and nothing else. Since all the locale names
are in plain ASCII I think it will be safe to use ASCII-only lower-case
conversion.
Nicolai Tufar
problems that occur if sublink is referenced via a join alias variable.
Perhaps this can be improved later, but a simple and safe fix is needed
for 7.3.1.
to plan nodes, not vice-versa. All executor state nodes now inherit from
struct PlanState. Copying of plan trees has been simplified by not
storing a list of SubPlans in Plan nodes (eliminating duplicate links).
The executor still needs such a list, but it can build it during
ExecutorStart since it has to scan the plan tree anyway.
No initdb forced since no stored-on-disk structures changed, but you
will need a full recompile because of node-numbering changes.
('SELECT expression') inline, like macros, during the constant-folding
phase of planning. The actual expansion is not difficult, but checking
that we're not changing the semantics of the call turns out to be more
subtle than one might think; in particular must pay attention to
permissions issues, strictness, and volatility.
well as function calls. This is needed for cases where the planner has
constant-folded or inlined the original function call. Possibly we should
back-patch this change into 7.3 branch as well.
and eliminate its manual pfree() calls. This solves the encoding-conversion
bug recently reported, and should be faster and more robust than the
original coding anyway. For example, we are no longer at risk if
datatype output routines leak memory or choose to return a constant string.
logic, dissuade planner from thinking that 'x IS DISTINCT FROM 42' may
be optimized into 'x = 42' (!!), cause dependency on = operator to be
recorded correctly, minor other improvements.
instead of only one. This should speed up planning (only one hash path
to consider for a given pair of relations) as well as allow more effective
hashing, when there are multiple hashable joinclauses.
operations: make sure we use operators that are compatible, as determined
by a mergejoin link in pg_operator. Also, add code to planner to ensure
we don't try to use hashed grouping when the grouping operators aren't
marked hashable.
sublink results and COPY's domain constraint checking. A Const that
isn't really constant is just a Bad Idea(tm). Remove hacks in
parse_coerce and other places that were needed because of the former
klugery.
just done for copyfuncs/equalfuncs. Read functions in particular get
a lot shorter than before, and it's much easier to compare an out function
with the corresponding read function to make sure they agree.
initdb forced due to small changes in nodestring format (regularizing
a few cases that were formerly idiosyncratic).
reliance on macros, in hopes of eliminating silly typos (like copying
to the wrong field) and just generally making it easier to see the forest
instead of the trees. As an example, here is the new code for A_Indices:
static A_Indices *
_copyAIndices(A_Indices *from)
{
A_Indices *newnode = makeNode(A_Indices);
COPY_NODE_FIELD(lidx);
COPY_NODE_FIELD(uidx);
return newnode;
}
static bool
_equalAIndices(A_Indices *a, A_Indices *b)
{
COMPARE_NODE_FIELD(lidx);
COMPARE_NODE_FIELD(uidx);
return true;
}
I plan to redo outfuncs/readfuncs in a similar style, but am committing
what I've got.
joinclauses is determined accurately for each join. Formerly, the code only
considered joinclauses that used all of the rels from the outer side of the
join; thus for example
FROM (a CROSS JOIN b) JOIN c ON (c.f1 = a.x AND c.f2 = b.y)
could not exploit a two-column index on c(f1,f2), since neither of the
qual clauses would be in the joininfo list it looked in. The new code does
this correctly, and also is able to eliminate redundant clauses, thus fixing
the problem noted 24-Oct-02 by Hans-Jürgen Schönig.
-hackers a couple days ago.
Notes/caveats:
- added regression tests for the new functionality, all
regression tests pass on my machine
- added pg_dump support
- updated PL/PgSQL to support per-statement triggers; didn't
look at the other procedural languages.
- there's (even) more code duplication in trigger.c than there
was previously. Any suggestions on how to refactor the
ExecXXXTriggers() functions to reuse more code would be
welcome -- I took a brief look at it, but couldn't see an
easy way to do it (there are several subtly-different
versions of the code in question)
- updated the documentation. I also took the liberty of
removing a big chunk of duplicated syntax documentation in
the Programmer's Guide on triggers, and moving that
information to the CREATE TRIGGER reference page.
- I also included some spelling fixes and similar small
cleanups I noticed while making the changes. If you'd like
me to split those into a separate patch, let me know.
Neil Conway
one more row from the subplan than the COUNT would appear to require.
This costs a little more logic but a number of people have complained
about the old implementation.
results due to doing arithmetic on uninitialized values. Add some
documentation about the AT TIME ZONE construct. Update some other
date/time documentation that seemed out of date for 7.3.
database access outside a transaction; revert bogus performance improvement
in SIBackendInit(); improve comments; add documentation (this part courtesy
Neil Conway).
parameter to allow it to be forced off for comparison purposes.
Add ORDER BY clauses to a bunch of regression test queries that will
otherwise produce randomly-ordered output in the new regime.
of groups produced by GROUP BY. This improves the accuracy of planning
estimates for grouped subselects, and is needed to check whether a
hashed aggregation plan risks memory overflow.
"canceled", so I changed the one remaining usage of the British
spelling ("cancelled") over to the former, and updated the translation
files appropriately.
Neil Conway
The code was not making TupleConstr structs for such catalogs in
several places; with the consequence that the not-null constraint
wasn't actually enforced. With this change,
INSERT INTO pg_proc VALUES('sdf');
generates a 'Fail to add null value' error instead of a core dump.
- CLUSTER ALL clusters all the tables that have some index with
indisclustered set and the calling user owns.
- CLUSTER tablename clusters the named table, using the index with
indisclustered set. If no index has the bit set, throws elog(ERROR).
- The multi-relation version (CLUSTER ALL) uses a multitransaction
approach, similar to what VACUUM does.
Alvaro Herrera
anymore given the mktime() workaround now done in DetermineLocalTimeZone.
This has now been confirmed by Robert Bruccoleri for Irix, and I'm going
to extrapolate to AIX as well.
before commit, not after :-( --- the original coding is not only unsafe
if an error occurs while it's processing, but it generates an invalid
sequence of WAL entries. Resurrect 7.2 logic for deleting items when
no longer needed. Use an enum instead of random macros. Editorialize
on names used for routines and constants. Teach backend/nodes routines
about new field in CreateTable struct. Add a regression test.
>
> ... he is now about to write an inlined version that can go into
> s_lock.h . I'll send the new patch later on...
OK, here it comes:
An inlined version of tas(), that works for both, powerpc and
powerpc64. The patch is against 7.3b5 and passes the test suite on
both architectures.
Reinhard Max
postgres.h or c.h includes a system header (such as stdio.h or
stdlib.h), there's no need to specifically include it in any of the .c
files in the backend.
Neil Conway
in hopes of reducing platform-to-platform variations in its results.
This will cause the geometry regression test to start failing on some
platforms. I plan to update the test later today.
precision for float4, float8, and geometric types. Set it in pg_dump
so that float data can be dumped/reloaded exactly (at least on platforms
where the float I/O support is properly implemented). Initial patch by
Pedro Ferreira, some additional work by Tom Lane.
now)" item on the open items, and subsequent plpgsql function I sent in,
made me realize it was too hard to get the upper and lower bound of an
array. The attached creates two functions that I think will be very
useful when combined with the ability of plpgsql to return sets.
array_lower(array, dim_num)
- and -
array_upper(array, dim_num)
They return the value (as an int) of the upper and lower bound of the
requested dim in the provided array.
Joe Conway
node now does its own grouping of the input rows, and has no need for a
preceding GROUP node in the plan pipeline. This allows elimination of
the misnamed tuplePerGroup option for GROUP, and actually saves more code
in nodeGroup.c than it costs in nodeAgg.c, as well as being presumably
faster. Restructure the API of query_planner so that we do not commit to
using a sorted or unsorted plan in query_planner; instead grouping_planner
makes the decision. (Right now it isn't any smarter than query_planner
was, but that will change as soon as it has the option to select a hash-
based aggregation step.) Despite all the hackery, no initdb needed since
only in-memory node types changed.
a child table already has a matching column. Acquire appropriate
lock on child table; do the right thing with any CHECK constraints
attached to the new parent column.
where it's safe to do database access. Along the way, fix core dump
for 'DEFAULT' parameters to CREATE DATABASE. initdb forced due to
change in pg_proc entry.
(usually bison output files), not as standalone files. This hack
works around flex's insistence on including <stdio.h> before we are
able to include postgres.h; postgres.h will already be read before
the compiler starts to read the flex output file. Needed for largefile
support on some platforms.
core file to be produced for debugging, and avoids trying to run the
normal proc-exit cleanup hooks, which are likely to cause additional
problems if the system is hosed.
between signal handler and enable/disable code, avoid accumulation of
timing error due to trying to maintain remaining-time instead of
absolute-end-time, disable timeout before commit not after.
Only affects machines where MAXALIGN > 4, and is a boundary-condition
case even there, but still surprising that it's not been identified
before. Also reduce tuple chain move give-up messages from WARNING
to DEBUG1, since they are not unexpected conditions.
coercions, not implicit ones. For example, 'select abstime(1035497293)'
should succeed because there is an explicit binary coercion from int4
to abstime.
principled order; in particular ensure that all shared resources
are released before we release transaction locks. The code used
to release locks before buffer pins, which might explain an ancient
note I have about a bufmgr assertion failure I'd seen once several
years ago, and been unable to reproduce since. (Theory: someone
trying to drop a relation might be able to reach FlushRelationBuffers
before the last user of the relation had gotten around to dropping
his buffer pins.)
item, if the page containing the current item is split while the indexscan
is stopped and holds no read-lock on the page. The current item might
move right onto a page that the indexscan holds no pin on. In the prior
code this would allow btbulkdelete to reach and possibly delete the item,
causing 'my bits moved right off the end of the world!' when the indexscan
finally resumes. Fix by chaining read-locks to the right during
_bt_restscan and requiring btbulkdelete to LockBufferForCleanup on every
page it scans, not only those with deletable items. Per my pghackers
message of 25-May-02. (Too bad no one could think of a better way.)
whose conditions might yield NULL. The negated qual to attach to the
original query is properly 'x IS NOT TRUE', not 'NOT x'. This fix
produces correct behavior, but we may be taking a performance hit because
the planner is much stupider about IS NOT TRUE than it is about NOT
clauses. Future TODO: teach prepqual, other parts of planner how to
cope with BooleanTest clauses more effectively.
all utility statement types *except* a short list, per discussion a few
days ago. Add missing SetQuerySnapshot calls in VACUUM and REINDEX,
and guard against calling REINDEX DATABASE from a function (has same
problem as VACUUM).
rather than being reordered according to INSTEAD attribute for
implementation convenience.
Also, increase compiled-in recursion depth limit from 10 to 100 rewrite
cycles. 10 seems pretty marginal for situations where multiple rules
exist for the same query. There was a complaint about this recently,
so I'm going to bump it up. (Perhaps we should make the limit a GUC
parameter, but that's too close to being a new feature to do in beta.)
Ray Ontko 28-June-02. Also, fix prefix_selectivity for NAME lefthand
variables (it was bogusly assuming binary compatibility), and adjust
make_greater_string() to not call pg_mbcliplen() with invalid multibyte
data (this last per bug report that I can't find at the moment, but it
was in July '02).
> in the position that attislocal should be reset. I'll clean everything
> up and submit the patch I had originally made.
All right, this is it. This patch merely checks if child tables have
the column. If atttypid and atttypmod are the same, the attributes'
attinhcount is incremented; else the operation is aborted. If child
tables don't have the column, recursively add it.
attislocal is not touched in any case.
Alvaro Herrera
specifically ceil(), floor(), and sign(). There may be other functions
that need to be added, but this is a start. I've included some simple
regression tests.
Neil Conway
been bit by the fact that the locale functions return pointers to
modifiable variables. I added some comments that might help us avoid
the mistake in future.
command status at the interactive level. SPI_processed, etc are set
in the same way as the returned command status would have been set if
the same querystring were issued interactively. Per gripe from
Michael Paesold 25-Sep-02.
query that uses it. This ensures that triggers will be applied consistently
throughout a query even if someone commits changes to the relation's
pg_class.reltriggers field meanwhile. Per crash report from Laurette Cisneros.
While at it, simplify memory management in relcache.c, which no longer
needs the old hack to try to keep trigger info in the same place over
a relcache entry rebuild. (Should try to fix rd_att and rewrite-rule
access similarly, someday.) And make RelationBuildTriggers simpler and
more robust by making it build the trigdesc in working memory and then
CopyTriggerDesc() into cache memory.
ANALYZE is not quite clear when branches of the query are never
executed. So this tiny patch fixes that.
The patch is attached and can also be found at:
http://svana.org/kleptog/pgsql/pgsql-explain.patch
Martijn van Oosterhout
> > > > found in the postmaster and not included from elsewhere)
> >
> > shared libs on AIX need to be able to resolve all symbols at linkage time.
> > Those two symbols are in backend/utils/SUBSYS.o but not in the postgres
> > executable.
>
> They are defined in backend/utils/mb/conv.c and declared in
> include/mb/pg_wchar.h. They're also linked into the
> postmaster. I don't see anything unusual.
Attached is a patch to fix the mb linking problems on AIX. As a nice side effect
it reduces the duplicate symbol warnings to linking libpq.so and libecpg.so
(all shlibs that are not postmaster loadable modules).
Please apply to current (only affects AIX).
The _LARGE_FILES problem is unfortunately still open, unless Peter
has fixed it per his recent idea.
Zeugswetter Andreas SB SD
into postgres.c; make sure it happens for all cases that seem to need it.
Perhaps it would be better to explicitly exclude just a few utility
statement types from setting a snapshot?
in our write/flush operation any WAL entries that got queued while we
were waiting to get the WALWriteLock. This improves throughput when
transactions are small enough that several can be committed per WAL
write (ie, per disk revolution).
discussion some weeks ago. Also, add a check that two types to be
binary-equivalenced match as to typlen, typbyval, and typalign; if
they don't then it's surely a mistake to equivalence them.
just the significant fields of FunctionCallInfoData, rather than MemSet'ing
the whole struct to zero. Unused positions in the arg[] array will
thereby contain garbage rather than zeroes. This buys back some of the
performance hit from increasing FUNC_MAX_ARGS. Also tweak tuplesort.c
code for more speed by marking some routines 'inline'. All together
these changes speed up simple sorts, like count(distinct int4column),
by about 25% on a P4 running RH Linux 7.2.
client
utilities (libpq.dll and psql.exe) for win32 (missing defines,
adjustments to
includes, pedantic casting, non-existent functions) per:
http://developer.postgresql.org/docs/postgres/install-win32.html.
It compiles cleanly under Windows 2000 using Visual Studio .net. Also
compiles clean and passes all regression tests (regular and contrib)
under Linux.
In addition to a review by the usual suspects, it would be very
desirable for someone well versed in the peculiarities of win32 to take
a look.
Joe Conway
so that precision of result is always at least as good as you'd get from
float8 arithmetic (ie, always at least 16 digits of accuracy). Per
pg_hackers discussion a few days ago.
Vacuum must not advance pg_database.datvacuumxid nor truncate CLOG
unless it's processed *all* tables in the database. Vacuums run by
unprivileged users don't count.
(Beats head against nearest convenient wall...)
no reason to worry about the tuple commit status bits until the tuple
is inserted in a relation by heapam.c. Also, improve comments for
heap_addheader().
VACUUM FULL tuple moves. Store full-width t_infomask in WAL, rather
than storing low 8 bits and expecting to be able to reconstruct upper
bits. While at it, remove redundant t_oid field from WAL headers
(the OID, if present, is now recorded in the data portion of the tuple).
WAL version number bumped --- this does not force an initdb, you can
instead run pg_resetxlog after a clean shutdown of the old postmaster.
ProcKill instead, where we still have a PGPROC with which to wait on
LWLocks. This fixes 'can't wait without a PROC structure' failures
occasionally seen during backend shutdown (I'm surprised they weren't
more frequent, actually). Add an Assert() to LWLockAcquire to help
catch any similar mistakes in future. Fix failure to update MyProcPid
for standalone backends and pgstat processes.
executor should not return the tuple as successfully marked, because in
fact it's been deleted. Not clear that this case has ever been seen
in practice (I think you'd have to write a SELECT FOR UPDATE that calls
a function that deletes some row the SELECT will visit later...) but we
should be consistent. Also add comments to several other places that
got it right but didn't explain what they were doing.
contains the correct statistics. This is a partial solution for the
problem of allowing concurrent CREATE INDEX commands: unless they commit
at nearly the same instant, the second one will see the first one's
pg_class updates as committed, and won't try to update again, thus
avoiding the 'tuple concurrently updated' failure.
even when dealing with a nailed-in-cache relation; otherwise, following
VACUUM truncation of a system catalog, other backends might have
unreasonably large values of these fields.
the SQL99 standard. (I'm not sure that the character-class features are
quite right, but that can be fixed later.) Document SQL99 and POSIX
regexps as being different features; provide variants of SUBSTRING for
each.
parse analysis and into the execution code (in tablecmds.c). This
eliminates a lot of unreasonably complex code that needed to have two
or more execution paths in case it was dealing with a not-yet-created
table column vs. an already-existing one. The execution code is always
dealing with already-created tables and so needs only one case. This
also eliminates some potential race conditions (the table wasn't locked
between parse analysis and execution), makes it easy to fix the gripe
about wrong referenced-column names generating a misleading error message,
and lets us easily add a dependency from the foreign-key constraint to
the unique index that it requires the referenced table to have. (Cf.
complaint from Kris Jurka 12-Sep-2002 on pgsql-bugs.)
Also, third try at building a deletion mechanism that is not sensitive
to the order in which pg_depend entries are visited. Adding the above-
mentioned dependency exposed the folly of what dependency.c had been
doing: it failed for cases where B depends on C while both auto-depend
on A. Dropping A should succeed in this case, but was failing if C
happened to be visited before B. It appears the only solution is two
separate walks over the dependency tree.
with OPAQUE. CREATE LANGUAGE, CREATE TRIGGER, and CREATE TYPE will all
accept references to functions declared with OPAQUE --- but they will
issue a NOTICE, and will modify the function entries in pg_proc to have
the preferred type-safe argument or result types instead of OPAQUE.
Per recent pghackers discussions.
> I see in your recent bytea-LIKE patch
>
> if (datatype != BYTEAOID && pg_database_encoding_max_length()
> 1)
> len = pg_mbcliplen((const unsigned char *) workstr, len,
len - 1);
> else
> len -= -1;
>
> Surely there's one too many minus signs in that last?
Joe Conway
> moment, but they used to be used; I think the correct response is to
> put back the missing counter increments, not rip out the counters.
Ok, fair enough. It's worth noting that they've been broken for a
while -- for example, the HashJoin counter increments were broken when
you comitted r1.20 of executor/nodeHashJoin.c in May of '99.
I've attached a revised patch that doesn't remove the counters (but
doesn't increment them either: I'm not sure of all the places where
the counter should be incremented).
Neil Conway
composite types. Add a couple more lsyscache.c routines to support this,
and make use of them in some other places that were doing lookups the
hard way.
ruleutils display is not such a great idea. For arguments of functions
and operators I think we'd better keep the historical behavior of showing
such casts explicitly, to ensure that the function/operator is reparsed
the same way when the rule is reloaded. This also makes the output of
EXPLAIN less obscurantist about exactly what's happening.
to be flexible about assignment casts without introducing ambiguity in
operator/function resolution. Introduce a well-defined promotion hierarchy
for numeric datatypes (int2->int4->int8->numeric->float4->float8).
Change make_const to initially label numeric literals as int4, int8, or
numeric (never float8 anymore).
Explicitly mark Func and RelabelType nodes to indicate whether they came
from a function call, explicit cast, or implicit cast; use this to do
reverse-listing more accurately and without so many heuristics.
Explicit casts to char, varchar, bit, varbit will truncate or pad without
raising an error (the pre-7.2 behavior), while assigning to a column without
any explicit cast will still raise an error for wrong-length data like 7.3.
This more nearly follows the SQL spec than 7.2 behavior (we should be
reporting a 'completion condition' in the explicit-cast cases, but we have
no mechanism for that, so just do silent truncation).
Fix some problems with enforcement of typmod for array elements;
it didn't work at all in 'UPDATE ... SET array[n] = foo', for example.
Provide a generalized array_length_coerce() function to replace the
specialized per-array-type functions that used to be needed (and were
missing for NUMERIC as well as all the datetime types).
Add missing conversions int8<->float4, text<->numeric, oid<->int8.
initdb forced.
(notify/SI-overrun interrupt) while it is in process of doing proc_exit,
it is possible for Async_NotifyHandler() to try to start a transaction
when one is already running. This leads to Asserts() or worse. I think
it may only be possible to occur when frontend synchronization is lost
(ie, the elog(FATAL) in SocketBackend() fires), but that is a standard
occurrence after error during COPY. In any case, I have seen this
failure occur during regression tests, so it is definitely possible.
Bruce Momjian
Tom Lane
Tatsuo Ishii
Peter Eisentraut