'volatile' pointers to access those structures, so that optimizing
compilers will not decide to move the structure accesses outside of the
spinlock-acquire-to-spinlock-release sequence. There are no known bugs
in these uses at present, but based on bad experience with lwlock.c,
it seems prudent to ensure that we protect these other uses too.
Per pghackers discussion around 12-Dec. (Note: it should not be
necessary to worry about structures protected by LWLocks, since the
LWLock acquire and release operations are not inline macros.)
machines. I have just been observing some scenarios where set_ps_display
accounts for more than 10% of the backend CPU, and this loop has to be
the reason.
do not use the undo pointer anyway. This is a quick-hack solution for
the three-way deadlock condition discussed in pghackers 17-Dec-01.
Need to find a better way of doing it.
requests for implicit trivial coercions. Prevents sillinesses like
this one:
regression=# select x.int8_tbl.q1 from int8_tbl x;
ERROR: fmgr_info: function 270997776: cache lookup failed
wasn't updated to handle more than two decimal digits for fractional seconds
that now are possible in 7.2. This patch fixes the timestamp parsing logic.
I have built and tested on both jdbc1 and jdbc2.
to be used, causing extra sql statements to be executed. This was a
significant performance problem with the database meta data classes.
The fix is a simple one liner.
set-returning functions in its target list. This ensures that we
won't rewrite the query in a way that places set-returning functions
into quals (WHERE clauses). Cf. bug reports from Joe Conway.
- Fixed bug that caused segfault when given incorrect DB name.
- Fixed bug in ecpglib causing indicator to list the size of the
variable instead of the size of the data.
to use and significantly faster. This tweak saves 25% (!) of the runtime
of COPY IN in a test with 8000-character lines. I wouldn't normally
commit a performance improvement this late in the cycle, but 25% got
my attention...
the latest version and I wanted to make sure that there was a clean release.
I also change the build files as I discussed in my letter of Nov 6, 2001. At
the time I was asked to hold off until after the release.
Magnus Hagander that DLL only contains error strings for the Net***
functions, *not* WinSock. We need to look for a workable solution for
older Windows flavors ... but it won't happen for PG 7.2.
to give more useful error messages. Stephen Szabo's example of this
morning ('loop' used as a variable name inside a subselect) works
correctly now, and a FOR that is misinterpreted as an integer FOR will
draw 'missing .. at end of SQL expression', which is at least
marginally helpful.
remove brain-dead rule that double quotes are needed if and only if the
datatype is pass-by-reference; neither direction of the implication holds
water. Instead, examine the actual data string to see if it contains
any characters that force us to quote it.
Add some documentation about quoting of array values, which was previously
explained nowhere AFAICT.
never did inflateEnd, thus leaking some tens of KB per call. Which
added up *real fast* when dealing with, say, thousands of BLOBs.
Thanks to Lane Rollins for the bug report.
per bug report from Stefan Hadjistoytchev. There are some cases
where the dot notation works, but there are more where it doesn't.
Eventually ought to consider fixing the parser to allow cases like
func().field, but for now this is the simplest patch.
The bug was that any insert or update would fail if the returned oid was
larger than a signed int. Since OIDs are unsigned int's it was
a bug that the code used a java signed int to deal with the values. The bug
would result in the error message: "Unable to fathom update count".
While fixing the bug, it became apparent that other code made a similar
assumption about OIDs being signed ints. Therefore some methods that returned
or took OIDs are arguements also needed to be changed.
Since we are so close to the 7.2 release I have added new methods that
return longs and deprecated the old methods returning ints. Therefore all
old code should still work without requiring a code change to cast from long to int. Also note that the methods below are PostgreSQL specific extensions to
the JDBC api are are not part of the spec from Sun, thus it is unlikely that
they are used much or at all.
The deprecated methods are:
ResultSet.getInsertedOID()
Statement.getInsertedOID()
Serialize.store()
Connection.putObject()
and are replaced by:
ResultSet.getLastOID()
Statement.getLastOID()
Serialize.storeObject()
Connection.storeObject()
All the deprecated methods returned int, while their replacements return long
This patch also fixes two comments in MD5Digest that the author Jeremy Wohl
submitted.
--Barry
where rightmost index page splits while we are waiting to obtain exclusive
lock on it. Not clear this would actually hurt (probably the callback
would always fail), but better safe than sorry.
Also, improve comments describing concurrency considerations in this code.
send patches to pgsql-patches list.
the zh_CN NLS patch is about 80K,
but sended twice and still can emerge on list.
so I've put it at:
http://laser.zhengmai.com.cn/download/zh_CN.po.diff.tar.gz
If possible, please download it and apply it.
(for current CVS).
regards laser
search lists was broken in such a way that only the most recent
instance of a given hash code would ever be searched, thus possibly
missing longer matches further back. Fixing this gave 5 to 10%
compression improvement on some text test cases. Additional small
tweaks to improve speed of inner loops a little bit. There is no
compatibility issue created by this change, since the compressed data
format and decompression algorithm don't change.
This seems the right thing for most usages, but I notice two places
where it is the wrong thing. One is that the default permissions on
TOAST rels should be no-access, not world-readable; the other is that
PrepareForTupleInvalidation doesn't really need to spend time looking
at tuples of TOAST relations.
int8, int16, int32, int64 and separately uint8, uint16, uint32, uint64
The previous patch grouped:
int8, int16 and int32
uint8, uint16 and uint32
int64 and uint64 <-- this grouping is wrong on AIX 4.3.3 and below
If you prefer to make 4 groups out of this you could apply this patch.
Andreas
There's also a little fix for the getRow() method. While fixing
absolute(), I noticed that getRow() wasn't quite following the spec: it
wasn't returning 0 when the ResultSet wasn't positioned on a row. I've
started a ResultSet test case and included it as well.
Liam Stewart
bpchar, bit, numeric with typmod -1. Alter format_type so that this
representation is printed when the typmod is -1. This ensures that
tables having such columns can be pg_dump'd and reloaded correctly.
Also, remove the rather useless and non-SQL-compliant default
precision and scale for type NUMERIC. A numeric column declared as
such (with no precision/scale) will now have typmod -1 which means
that numeric values of any precision/scale can be stored in it,
without conversion to a uniform scale. This seems significantly
more useful than the former behavior. Part of response to bug #513.
typmod of a particular column, mark the output with that same typmod,
not -1 as formerly. -1 is still used if there is any disagreement.
Part of response to bug#513.
operators. Should report the declared oprresult type, not the return type
of the underlying proc, which might be only binary-compatible (cf.
textcat entries).
used inside fk constraints, since some of the checks
in the trigger did a SELECT oid. Since the oid wasn't
actually used, I changed this to SELECT 1. My test
case with non-oid tables now works and fk regression
appears to run fine on my machine.
Stephan Szabo
clauses per path key. Indeed, we *must* do so or we will be unable to
form a valid plan for FULL JOIN with overlapping join conditions, eg
select * from a full join b on
a.v1 = b.v1 and a.v2 = b.v2 and a.v1 = b.v2.
mergeclauses in RIGHT/FULL join cases, just like the other routines have.
I'm not quite sure why I thought it didn't need one --- but Nick
Fankhauser's recent bug report proves that it does.
though alas not as unquoted function names. De-reserve a bunch of
keywords that could have been in ColId rather than ColLabel all along.
Per recent proposal in pgsql-patches.
one fuzzy translation fix, some
other messages tweaking. Theoretically,
should be up-to-date by now.
Please apply to /src/interfaces/libpq/ru.po
--
Serguei A. Mokhov
should be accounted for in the PROC_SEM_MAP_ENTRIES() macro. Otherwise
the ports that rely on this macro to size data structures are broken.
Mea culpa.
to the target list in gram.y; it must wait till after expansion of the
target list in analyze.c. Per bug report 4-Nov:
lx=# CREATE TABLE abc (a char, b char, c char);
CREATE
lx=# CREATE TABLE xyz (x, y, z) AS SELECT * FROM abc;
ERROR: CREATE TABLE/AS SELECT has mismatched column count
so that only one signal number is used not three. Flags in shared
memory tell the reason(s) for the current signal. This method is
extensible to handle more signal reasons without chewing up even more
signal numbers, but the immediate reason is to keep pg_pwd reloads
separate from SIGHUP processing in the postmaster.
Also clean up some problems in the postmaster with delayed response to
checkpoint status changes --- basically, it wouldn't schedule a checkpoint
if it wasn't getting connection requests on a regular basis.
- Fix handling of {data/schema}-only restores when using a full
backup file; prior version was restoring schema in data-only
restores. Added enum to make code easier to understand.
analysis. This keeps stored rules from prematurely absorbing default
information, which is necessary for ALTER TABLE SET DEFAULT to work
unsurprisingly with rules. See pgsql-bugs discussion 24-Oct-01.
postmaster children before client auth step. Postmaster now rereads
pg_pwd on receipt of SIGHUP, the same way that pg_hba.conf is handled.
No cycles need be expended to validate password cache validity during
connection startup.
recreated since the start of our transaction, our first reference to it
errored out because we'd try to reuse our old relcache entry for it.
Do this by accepting SI inval messages just before relcache search in
heap_openr, so that dead relcache entries will be flushed before we
search. Also, break heap_open/openr into two pairs of routines,
relation_open(r) and heap_open(r). The relation_open routines make
no tests on relkind and so can be used to open anything that has a
pg_class entry. The heap_open routines are wrappers that add a relkind
test to preserve their established behavior. Use the relation_open
routines in several places that had various kluge solutions for opening
rels that might be either heap or index rels.
Also, remove the old 'heap stats' code that's been superseded by Jan's
stats collector, and clean up some inconsistencies in error reporting
between the different types of ALTER TABLE.
stored in pg_pwd, to guard against failures of the sort observed by
Tom Yackel. Note: in the case of encrypted passwords this is no
restriction, since the string we are interested in is the MD5 hash.
from Philip Warner. Side effect of change is that GROUP BY expressions
will not be re-evaluated at multiple plan levels anymore, whereas this
sometimes happened with old code.
(why bother dropping individual objects in a just-created database?)
as well as dangerous (as the code stands, the drops will be issued in
the wrong database, namely the one you were originally connected to).
never overwrite adjacent pages with copied data, even if page header
and/or item pointers are already corrupt. Change inspired by trouble
report from Alvaro Herrera.
environment strings need to be moved around, do so when called from
initial startup (main.c), not in init_ps_status. This eliminates the
former risk of invalidating saved environment-string pointers, since
no code has yet had a chance to grab any such pointers when main.c
is running.
in .:/home/postgres/testversion/bin:/opt/perl5.6.1/bin:/home/postgres/bin:/usr/local/bin:/usr/bin:/opt/ansic/bin:/usr/ccs/bin:/usr/contrib/bin:/opt/nettladm/bin:/opt/pd/bin:/usr/bin/X11:/usr/contrib/bin/X11:/opt/upgrade/bin:/opt/CC/bin:/opt/langtools/bin:/opt/graphics/phigs/bin:/opt/java/bin:/bin:/opt/imake/bin:/opt/hparray/bin:/opt/aCC/bin:/opt/lrom/bin:/usr/local/nmh/bin:. (I suppose the only common case for this is '.').
per suggestion from Peter. Simplify several APIs by transmitting the
original argv location directly from main.c to ps_status.c, instead of
passing it down through several levels of subroutines.
Mask both typmod subfields for INTERVAL to avoid setting the high bit,
per dire warning from Tom Lane.
Clear tmask for DTK_ISO_TIME case to avoid time zone troubles.
Symptom reported by Tom Lane.
Clean up checking for valid time zone info in output routine.
This should now work for both SQL99 and Unix-style time zones.
Put in explicit check for INTERVAL() typmod rounding to avoid accumulating
cruft in the lower bits. Not sure that this helps, but we'll need to do
something. The symptom is visible with a query like
select interval(2) '10000 days 01:02:03.040506';
Regression tests are patched to repair the Tom Lane symptom, and all pass.
subprocesses; perhaps this will fix portability problem just noted by
Lockhart. Also, move test for bad permissions of DataDir to a more
logical place.
bootstrap) check for a valid PG_VERSION file before looking at anything
else in the data directory. This fixes confusing error report when
trying to start current sources in a pre-7.1 data directory.
Per trouble report from Rich Shepard 10/18/01.
just after receipt of the startup packet. Now, postmaster children
that are waiting for client authentication response will show as
'postgres: user database host authentication'. Also, do an
init_ps_display for startup/shutdown/checkpoint subprocesses,
so that they are readily identifiable as well. Fix an obscure race
condition that could lead to Assert failure in the postmaster ---
attempting to start a checkpoint process before any connections have
been received led to calling PostmasterRandom before setting random_seed.
authentication failed' and a 'send() failed: Broken pipe' message
on every connection from psql in password auth mode. Problem is
that psql doesn't ask user for a password until it sees a password
challenge failure, and libpq just closes the connection unceremoniously
if it's challenged for a password when it hasn't got one to send.
Accordingly, EOF from the client after asking for a password is
normal behavior and should not result in postmaster log entries.
the entered password would get echoed on some platforms, eg HPUX.
We have enough copies of this code that I'm thinking it ought to be
moved into libpq, but that's a task for another day.
Modified the parser and the SET handlers to use full Node structures
rather than simply a character string argument.
Implement INTERVAL() YEAR TO MONTH (etc) syntax per SQL99.
Does not yet accept the goofy string format that goes along with, but
this should be fairly straight forward to fix now as a bug or later
as a feature.
Implement precision for the INTERVAL() type.
Use the typmod mechanism for both of INTERVAL features.
Fix the INTERVAL syntax in the parser:
opt_interval was in the wrong place.
INTERVAL is now a reserved word, otherwise we get reduce/reduce errors.
Implement an explicit date_part() function for TIMETZ.
Should fix coersion problem with INTERVAL reported by Peter E.
Fix up some error messages for date/time types.
Use all caps for type names within message.
Fix recently introduced side-effect bug disabling 'epoch' as a recognized
field for date_part() etc. Reported by Peter E. (??)
Bump catalog version number.
Rename "microseconds" current transaction time field
from ...Msec to ...Usec. Duh!
date/time regression tests updated for reference platform, but a few
changes will be necessary for others.
clause being added to a particular restriction-clause list is redundant
with those already in the list. This avoids useless work at runtime,
and (perhaps more importantly) keeps the selectivity estimation routines
from generating too-small estimates of numbers of output rows.
Also some minor improvements in OPTIMIZER_DEBUG displays.
view when using the aggregate function count() and function nextval
that returns an int8 value, but in python is represented like string:
>> db.query("select nextval('my_seq')").getresult()
[('2',)]
>> db.query("select count(*) from films").dictresult()
[{'count': '120'}]
Ricardo Caesar Lenzi
> ! $$ = cat_str(8, make_str("grant"), $2, make_str("on"), $4, $5,
> make_str("to"), $7, $8);
> ISTM your patch loses the opt_with_grant clause. (Of course the
> backend doesn't currently accept that clause anyway, but that's no
> reason for ecpg to drop it.)
My patch doesn't loose the option, it's never been passed on anyway:
opt_with_grant: WITH GRANT OPTION
{
mmerror(ET_ERROR, "WITH GRANT OPTION is not supported. Only relation owners can
set privileges");
}
| /*EMPTY*/
;
The existing code in ecpg/preproc/preproc.y to handle the WITH option
simply throws an error and aborts the processing... The patch below
prevents the segfault and also passes on the WITH option to the
backend, probably a better fix.
Lee Kindness
> - corrects a bit the UTF-8 code from Tatsuo to allow Unicode 3.1
> characters (characters with values >= 0x10000, which are encoded on
> four bytes).
Also, update mb/expected/unicode.out. This is necessary since the
patches affetc the result of queries using UTF-8.
---------------------------------------------------------------
Hi,
I should have sent the patch earlier, but got delayed by other stuff.
Anyway, here is the patch:
- most of the functionality is only activated when MULTIBYTE is
defined,
- check valid UTF-8 characters, client-side only yet, and only on
output, you still can send invalid UTF-8 to the server (so, it's
only partly compliant to Unicode 3.1, but that's better than
nothing).
- formats with the correct number of columns (that's why I made it in
the first place after all), but only for UNICODE. However, the code
allows to plug-in routines for other encodings, as Tatsuo did for
the other multibyte functions.
- corrects a bit the UTF-8 code from Tatsuo to allow Unicode 3.1
characters (characters with values >= 0x10000, which are encoded on
four bytes).
- doesn't depend on the locale capabilities of the glibc (useful for
remote telnet).
I would like somebody to check it closely, as it is my first patch to
pgsql. Also, I created dummy .orig files, so that the two files I
created are included, I hope that's the right way.
Now, a lot of functionality is NOT included here, but I will keep that
for 7.3 :) That includes all string checking on the server side (which
will have to be a bit more optimised ;) ), and the input checking on
the client side for UTF-8, though that should not be difficult. It's
just to send the strings through mbvalidate() before sending them to
the server. Strong checking on UTF-8 strings is mandatory to be
compliant with Unicode 3.1+ .
Do I have time to look for a patch to include iso-8859-15 for 7.2 ?
The euro is coming 1. january 2002 (before 7.3 !) and over 280
millions people in Europe will need the euro sign and only iso-8859-15
and iso-8859-16 have it (and unfortunately, I don't think all Unices
will switch to Unicode in the meantime)....
err... yes, I know that this is not every single person in Europe that
uses PostgreSql, so it's not exactly 280m, but it's just a matter of
time ! ;)
I'll come back (on pgsql-hackers) later to ask a few questions
regarding the full unicode support (normalisation, collation,
regexes,...) on the server side :)
Here is the patch !
Patrice.
--
Patrice HÉDÉ ------------------------------- patrice à islande org -----
-- Isn't it weird how scientists can imagine all the matter of the
universe exploding out of a dot smaller than the head of a pin, but they
can't come up with a more evocative name for it than "The Big Bang" ?
-- What would _you_ call the creation of the universe ?
-- "The HORRENDOUS SPACE KABLOOIE !" - Calvin and Hobbes
------------------------------------------ http://www.islande.org/ -----
