Tom Lane [Sun, 19 Apr 2009 18:52:58 +0000 (18:52 +0000)]
Revise plpgsql's scanner to process comments and string literals in a way
more nearly matching the core SQL scanner. The user-visible effects are:
* Block comments (slash-star comments) now nest, as per SQL spec.
* In standard_conforming_strings mode, backslash as the last character of a
non-E string literal is now correctly taken as an ordinary character;
formerly it was misinterpreted as escaping the ending quote. (Since the
string also had to pass through the core scanner, this invariably led
to syntax errors.)
* Formerly, backslashes in the format string of RAISE were always treated as
quoting the next character, regardless of mode. Now, they are ordinary
characters with standard_conforming_strings on, while with it off, they
introduce the same set of escapes as in the core SQL scanner. Also,
escape_string_warning is now effective for RAISE format strings. These
changes make RAISE format strings work just like any other string literal.
This is implemented by copying and pasting a lot of logic from the core
scanner. It would be a good idea to look into getting rid of plpgsql's
scanner entirely in favor of using the core scanner. However, that involves
more change than I can justify making during beta --- in particular, the core
scanner would have to become re-entrant.
In passing, remove the kluge that made the plpgsql scanner emit T_FUNCTION or
T_TRIGGER as a made-up first token. That presumably had some value once upon
a time, but now it's just useless complication for both the scanner and the
grammar.
Tom Lane [Sun, 19 Apr 2009 15:49:34 +0000 (15:49 +0000)]
Mention as a potential incompatibility the fact that SELECT DISTINCT, UNION,
etc are no longer guaranteed to produce sorted output; per gripe from Ian
Barwick. Also improve the release note entries about to_timestamp(), per
Brendan Jurd.
Tom Lane [Thu, 16 Apr 2009 20:42:16 +0000 (20:42 +0000)]
Fix planner to restore its previous level of intelligence about pushing
constants through full joins, as in
select * from tenk1 a full join tenk1 b using (unique1)
where unique1 = 42;
which should generate a fairly cheap plan where we apply the constraint
unique1 = 42 in each relation scan. This had been broken by my patch of
2008-06-27, which is now reverted in favor of a more invasive but hopefully
less incorrect approach. That patch was meant to prevent incorrect extraction
of OR'd indexclauses from OR conditions above an outer join. To do that
correctly we need more information than the outerjoin_delay flag can provide,
so add a nullable_relids field to RestrictInfo that records exactly which
relations are nulled by outer joins that are underneath a particular qual
clause. A side benefit is that we can make the test in create_or_index_quals
more specific: it is now smart enough to extract an OR'd indexclause into the
outer side of an outer join, even though it must not do so in the inner side.
The old coding couldn't distinguish these cases so it could not do either.
Tom Lane [Tue, 14 Apr 2009 22:18:47 +0000 (22:18 +0000)]
Fix broken {xufailed} production that made HEAD fail on
select u&42 from table-with-a-u-column;
Also fix missing SET_YYLLOC() in the {dolqfailed} production that I suppose
this was based on. The latter is a pre-existing bug, but the only effect
is to misplace the error cursor by one token, so probably not worth
backpatching.
Andrew Dunstan [Sun, 12 Apr 2009 21:02:44 +0000 (21:02 +0000)]
Fix logic to detect conflicts or blocks involving exclusive locks in parallel restore items.
If a currently running item needs an exclusive lock on any item that the candidate items needs
any sort of lock on, or vice versa, then the candidate item is not allowed to run now, and
must wait till later.
Tom Lane [Sat, 11 Apr 2009 20:23:05 +0000 (20:23 +0000)]
Fix pg_dumpall so that when --clean is specified, it drops roles and
tablespaces in an order that has some chance of working.
Per a complaint from Kevin Bailey.
This is a pre-existing bug, but given the lack of prior complaints I'm
not sure it's worth back-patching. In most cases failure of the DROP
commands wouldn't be that important anyway.
In passing, fix syntax errors in dumpCreateDB()'s queries for old servers;
these were apparently introduced in recent binary_upgrade patch.
Tom Lane [Thu, 9 Apr 2009 20:50:42 +0000 (20:50 +0000)]
Update time zone data files to tzdata release 2009e: DST law changes in
Argentina/San_Luis, Cuba, Jordan (historical correction only), Morocco,
Palestine, Syria, Tunisia.
Tom Lane [Thu, 9 Apr 2009 17:39:50 +0000 (17:39 +0000)]
Remove SQL-compatibility function cardinality(). It is not exactly clear
how this ought to behave for multi-dimensional arrays. Per discussion,
not having it at all seems better than having it with what might prove
to be the wrong behavior. We can always add it later when we have consensus
on the correct behavior.
Tom Lane [Thu, 9 Apr 2009 02:57:53 +0000 (02:57 +0000)]
Fix the plpgsql memory leak exhibited in bug #4677. That leak was introduced
by my patch of 2007-01-28 to use per-subtransaction ExprContexts/EStates:
since we re-prepared any expression tree when the current subtransaction ID
changed, we'd accumulate more and more leaked expression state trees in the
outermost subtransaction if the same function was executed at multiple levels
of subtransaction nesting. To fix, go back to the previous scheme where
there was only one EState per transaction for simple plpgsql expressions.
We really only need an ExprContext per subtransaction, not a whole EState,
so it's possible to keep prepared expression state trees in the one EState
throughout the transaction. This should be more efficient as well as not
leaking memory for cases involving lots of subtransactions.
The added regression test is the case that inspired the 2007-01-28 patch in
the first place, just to make sure we didn't go backwards. The current
memory leak complaint is unfortunately hard to test for in the regression
test framework, though manual testing shows it's fixed.
Although this is a pre-existing bug, I'm not back-patching because I'd like to
see this method get some field testing first. Consider back-patching if it
gets through 8.4beta unscathed.
Tom Lane [Wed, 8 Apr 2009 22:29:30 +0000 (22:29 +0000)]
Remove psql's ancient hack that suppressed functions taking or returning
cstring from the output of \df. Now that the default behavior is to
exclude all system functions, the de-cluttering rationale for this behavior
seems pretty weak; and it was always quite confusing/unhelpful if you were
actually looking for I/O functions. (Not to mention if you were looking
for encoding converters or other cases that might take or return cstring.)
XMLATTRIBUTES() should send the attribute values through
map_sql_value_to_xml_value() instead of directly through the data type output
function. This is per SQL standard, and consistent with XMLELEMENT().
Tell gettext which codeset to use by calling bind_textdomain_codeset(). We
already did that on Windows, but it's needed on other platforms too when
LC_CTYPE=C. With other locales, we enforce (or trust) that the codeset of
the locale matches the server encoding so we don't need to bind it
explicitly. It should do no harm in that case either, but I don't have
full faith in the PG encoding -> OS codeset mapping table yet. Per recent
discussion on pgsql-hackers.
Tom Lane [Tue, 7 Apr 2009 15:53:54 +0000 (15:53 +0000)]
Defend against non-ASCII letters in fuzzystrmatch code. The functions
still don't behave very sanely for multibyte encodings, but at least
they won't be indexing off the ends of static arrays.
Tom Lane [Tue, 7 Apr 2009 00:31:26 +0000 (00:31 +0000)]
Add an optional parameter to pg_start_backup() that specifies whether to do
the checkpoint in immediate or lazy mode. This is to address complaints
that pg_start_backup() takes a long time even when there's no need to minimize
its I/O consumption.
Tom Lane [Sun, 5 Apr 2009 22:28:59 +0000 (22:28 +0000)]
Change cardinality() into a C-code function, instead of a SQL-language
alias for array_length(v,1). The efficiency gain here is doubtless
negligible --- what I'm interested in is making sure that if we have
second thoughts about the definition, we will not have to force a
post-beta initdb to change the implementation.
Tom Lane [Sun, 5 Apr 2009 20:32:06 +0000 (20:32 +0000)]
Make ExecInitExpr build the list of SubPlans found in a plan tree in order
of discovery, rather than reverse order. This doesn't matter functionally
(I suppose the previous coding dates from the time when lcons was markedly
cheaper than lappend). However now that EXPLAIN is labeling subplans with
IDs that are based on order of creation, this may help produce a slightly
less surprising printout.
Tom Lane [Sun, 5 Apr 2009 19:59:40 +0000 (19:59 +0000)]
Change EXPLAIN output so that subplans and initplans (particularly CTEs)
are individually labeled, rather than just grouped under an "InitPlan"
or "SubPlan" heading. This in turn makes it possible for decompilation of
a subplan reference to usefully identify which subplan it's referencing.
I also made InitPlans identify which parameter symbol(s) they compute,
so that references to those parameters elsewhere in the plan tree can
be connected to the initplan that will be executed. Per a gripe from
Robert Haas about EXPLAIN output of a WITH query being inadequate,
plus some longstanding pet peeves of my own.
Fix infinite loop while checking of partial match in pending list.
Improve comments. Now GIN-indexable operators should be strict.
Per Tom's questions/suggestions.
Tom Lane [Sun, 5 Apr 2009 04:19:59 +0000 (04:19 +0000)]
Remove a boatload of useless definitions of 'int optreset'. If we
are using our own ports of getopt or getopt_long, those will define
the variable for themselves; and if not, we don't need these, because
we never touch the variable anyway.
Tom Lane [Sun, 5 Apr 2009 04:09:01 +0000 (04:09 +0000)]
I had always wondered why pg_config.h.win32 claimed that Windows
provides optreset. Current mastodon results prove that in fact it
does not; it was only because getopt.c defined the variable anyway
that things failed to fall over.
Tom Lane [Sun, 5 Apr 2009 00:40:35 +0000 (00:40 +0000)]
Remove contrib/intarray's definitions of the <@ and @> operators, so that they
don't cause confusion with the built-in anyarray versions of those operators.
Adjust the module's index opclasses to support the built-in operators in place
of the private ones.
The private implementations are still available under their historical
names @ and ~, so no functionality is lost. Some quick testing suggests
that they offer no real benefit over the core operators, however.
Tom Lane [Sat, 4 Apr 2009 21:55:50 +0000 (21:55 +0000)]
Make an attempt at fixing our current Solaris 11 breakage: add a configure
probe for opterr (exactly like the one for optreset) and have getopt.c
define the variables only if configure doesn't find them in libc.
Tom Lane [Sat, 4 Apr 2009 21:12:31 +0000 (21:12 +0000)]
Remove the recently added node types ReloptElem and OptionDefElem in favor
of adding optional namespace and action fields to DefElem. Having three
node types that do essentially the same thing bloats the code and leads
to errors of confusion, such as in yesterday's bug report from Khee Chin.
Tom Lane [Sat, 4 Apr 2009 17:40:36 +0000 (17:40 +0000)]
A session that does not have any live snapshots does not have to be waited for
when we are waiting for old snapshots to go away during a concurrent index
build. In particular, this rule lets us avoid waiting for
idle-in-transaction sessions.
This logic could be improved further if we had some way to wake up when
the session we are currently waiting for goes idle-in-transaction. However
that would be a significantly more complex/invasive patch, so it'll have to
wait for some other day.
Tom Lane [Sat, 4 Apr 2009 04:53:25 +0000 (04:53 +0000)]
Rewrite interval_hash() so that the hashcodes are equal for values that
interval_eq() considers equal. I'm not sure how that fundamental requirement
escaped us through multiple revisions of this hash function, but there it is;
it's been wrong since interval_hash was first written for PG 7.1.
Per bug #4748 from Roman Kononov.
Backpatch to all supported releases.
This patch changes the contents of hash indexes for interval columns. That's
no particular problem for PG 8.4, since we've broken on-disk compatibility
of hash indexes already; but it will require a migration warning note in
the next minor releases of all existing branches: "if you have any hash
indexes on columns of type interval, REINDEX them after updating".
To implement this without almost duplicating the reloption table, treat
relopt_kind as a bitmask instead of an integer value. This decreases the
range of allowed values, but it's not clear that there's need for that much
values anyway.
This patch also makes heap_reloptions explicitly a no-op for relation kinds
other than heap and TOAST tables.
Patch by ITAGAKI Takahiro with minor edits from me. (In particular I removed
the bit about adding relation kind to an error message, which I intend to
commit separately.)
Tom Lane [Sat, 4 Apr 2009 00:41:11 +0000 (00:41 +0000)]
Make \dt \di and friends more consistent about the treatment of
TOAST tables and indexes; to wit, never show either. (You can
examine them with plain \d if you're really so inclined.)
Tom Lane [Fri, 3 Apr 2009 23:27:17 +0000 (23:27 +0000)]
Use (unsigned char) cast in argument of pg_tolower(). Maybe it works on
Windows without that, but we shouldn't put bad examples where people might
copy them. Also, reformat slightly to improve the odds that pgindent
won't go nuts on this.
Tom Lane [Fri, 3 Apr 2009 18:17:43 +0000 (18:17 +0000)]
Add a comment documenting the question of whether PrefetchBuffer should
try to protect an already-existing buffer from being evicted. This was
left as an open issue when the posix_fadvise patch was committed. I'm
not sure there's any evidence to justify more work in this area, but we
should have some record about it in the source code.
Tom Lane [Fri, 3 Apr 2009 16:59:43 +0000 (16:59 +0000)]
Defend against possible crash if a plpython function does not specify names
for its arguments. Also add a regression test, since someone apparently
changed every single plpython test case to use only named parameters; else
we'd have noticed this sooner.
Euler Taveira de Oliveira, per a report from Alvaro
Magnus Hagander [Fri, 3 Apr 2009 11:52:08 +0000 (11:52 +0000)]
Make directory name comparisons on Win32 case insensitive.
This method will not catch all different ways since the locale
handling in NTFS doesn't provide an easy way to do that, but it
will hopefully solve the most common cases causing startup
problems when the backend is found in the system PATH.
Tom Lane [Thu, 2 Apr 2009 22:39:30 +0000 (22:39 +0000)]
Refactor ExecProject and associated routines so that fast-path code is used
for simple Var targetlist entries all the time, even when there are other
entries that are not simple Vars. Also, ensure that we prefetch attributes
(with slot_getsomeattrs) for all Vars in the targetlist, even those buried
within expressions. In combination these changes seem to significantly
reduce the runtime for cases where tlists are mostly but not exclusively
Vars. Per my proposal of yesterday.