ads' corner

== Angewandte Patches ==

Tom Lane committed:

- In pgsql/src/backend/utils/adt/arrayfuncs.c, revert dubious message
  wording change.

- Make to_timestamp and friends skip leading spaces before an integer
  field, even when not in Fujii Masao mode.  This improves
  compatibility with Oracle and with our pre-8.4 behavior, as per bug
  #4862.  Add a couple of regression test cases for this.  In passing,
  get rid of the labeling of the individual test cases; doesn't seem
  to be good for anything except causing extra work when inserting a
  test...  Brendan Jurd and Tom Lane

- For bulk write operations (eg COPY IN), use a ring buffer of 16MB
  instead of the 256KB limit originally enforced by a patch committed
  2008-11-06.  Per recent test results, the smaller size resulted in
  an undesirable decrease in bulk data loading speed, due to COPY
  processing frequently getting blocked for WAL flushing.  This area
  might need more tweaking later, but this setting seems to be good
  enough for 8.4.

- Fix the makefiles to fail cleanly if Perl is needed but not present.
  This used to work as intended, but got broken some time ago (a
  quoted empty string is not an empty string), and got broken some
  more by the changes to generate ecpg's preproc.y automatically.
  Given all the unprotected uses of $(PERL) elsewhere, it seems best
  to make use of the $(missing) script rather than trying to ensure
  each such use is protected individually.  Also fix various bits of
  documentation that omitted to mention Perl as a requirement for
  building from a CVS pull.  Per a complaint from Robert Haas.

- In pgsql/src/backend/utils/adt/geo_ops.c, fix an ancient error in
  dist_ps (distance from point to line segment), which a number of
  other geometric operators also depend on.  It miscalculated the
  slope of the perpendicular to the given line segment anytime that
  slope was other than 0, infinite, or +/-1.  In some cases the error
  would be masked because the true closest point on the line segment
  was one of its endpoints rather than the intersection point, but in
  other cases it could give an arbitrarily bad answer.  Per bug #4872
  from Nick Roosevelt.  Bug goes clear back to Berkeley days, so patch
  all supported branches.  Make a couple of cosmetic adjustments while
  at it.

- In pgsql/doc/src/sgml/dml.sgml, provide a link to the UPDATE
  reference page in the 'Updating Data' section of the docs.  Per
  suggestion from Brad Bowman.

- In pgsql/contrib/pg_standby/pg_standby.c, make prior patch
  pgindent-safe.

- In pgsql/src/backend/utils/error/elog.c, add
  __attribute__((format_arg(1))) to the declaration of err_gettext(),
  to restore gcc's ability to crosscheck format arguments within
  elog.c.  Noted in a test compilation with -Wformat-nonliteral
  enabled.

- Cleanup and code review for the patch that made bgwriter active
  during archive recovery.  Invent a separate state variable and
  inquiry function for XLogInsertAllowed() to clarify some tests and
  make the management of writing the end-of-recovery checkpoint less
  klugy.  Fix several places that were incorrectly testing InRecovery
  when they should be looking at RecoveryInProgress or
  XLogInsertAllowed (because they will now be executed in the bgwriter
  not startup process).  Clarify handling of bad LSNs passed to
  XLogFlush during recovery.  Use a spinlock for setting/testing
  SharedRecoveryInProgress.  Improve quite a lot of comments.  Heikki
  Linnakangas and Tom Lane.

- In pgsql/doc/src/sgml/backup.sgml, marginal improvement of
  description of recovery_end_command.

- In pgsql/doc/src/sgml/monitoring.sgml, add a note about SystemTap vs
  DTrace spelling of probe names.

- In pgsql/doc/src/sgml/release-8.4.sgml, update release notes in
  final preparation for 8.4.0.

- In pgsql/doc/Makefile, revert addition of "o" to tar options.  This
  was intended to fix bug #4883, but the cure appears to be worse than
  the disease.  It turns out that GNU tar versions 1.14.x misinterpret
  -o as --same-owner, not --no-same-owner, leading to exactly the
  wrong behavior for both root and nonroot users.  While that bug has
  been fixed for nearly five years, these tar versions are still found
  in the wild, notably in OS X 10.4.  Given that #4883 was the first
  complaint we'd heard, it's definitely not worth fixing at the risk
  of breaking things for other users.  Perhaps revisit at a later date
  when we're not up against a release deadline.

Peter Eisentraut committed:

- In pgsql/doc/src/sgml/ref/move.sgml, add more explicit note that the
  parameters of MOVE are identical to FETCH.  Per David Wheeler.

- In pgsql/src/backend/access/gist/gistsplit.c, correct grammar in
  picksplit debug messages

- In pgsql/doc/src/sgml/Makefile, man pages back to lowercase names,
  with additional fixup to make the .so links work.

- In pgsql/doc/Makefile, use $(TAR) instead of tar in makefiles.

- In pgsql/doc/Makefile, extract tarballs with "o" option (tar xof),
  to behave reasonably when run as root, as would usually be the case
  during make install.  Per bug #4883.

- Translation updates for 8.4 release.  File that are translated less
  than 80% have been removed, as per new translation team policy.

Marc Fournier committed:

- Bundle RC2.

- Bundle v8.4.0.

Magnus Hagander committed:

- Properly initialize SSL engines when used from libpq.  This is
  required for most external engines.  Per report and initial code
  from Lars Kanis.

- In pgsql/src/backend/libpq/hba.c, parse_ident_usermap() should use
  ereport(LOG) and not ERROR, and put the return value in the *error_p
  variable.  Noted by Tom Lane.

- In pgsql/doc/src/sgml/client-auth.sgml, fix incorrect LDAP example,
  noted by Tom Lane.

- In pgsql/src/backend/libpq/auth.c, disallow empty passwords in LDAP
  authentication, the same way we already do it for PAM.

Heikki Linnakangas committed:

- Disable pg_standby -l option because the backend doesn't expect the
  recovered file to be a symlink.  We tried to fix this issue with an
  earlier server-side patch, but it didn't fix the whole issue.  The
  same bug is present in older releases as well, but the 8.4 train is
  about to leave the station, and I'm not sure if have consensus on
  whether we can remove the -l option in back-branches or do we need
  to attempt a server-side fix to make symlinking safe.  Patch by
  Simon Riggs, per discussion on bug identified by Fujii Masao.

- In pgsql/src/backend/access/transam/twophase.c, the code to unlink
  dropped relations in FinishPreparedTransaction() was acting like
  runs inside WAL recovery, but it doesn't.  I must've copy-pasted
  this from a redo-function in the relation forks patch.  Noticed by
  Tom Lane while he was looking through callers of smgrdounlink().

- Fix some serious bugs in archive recovery, now that bgwriter is
  active during it: When bgwriter is active, the startup process can't
  perform mdsync() correctly because it won't see the fsync requests
  accumulated in bgwriter's private pendingOpsTable.  Therefore make
  bgwriter responsible for the end-of-recovery checkpoint as well,
  when it's active.  When bgwriter is active (= archive recovery), the
  startup process must not accumulate fsync requests to its own
  pendingOpsTable, since bgwriter won't see them there when it
  performs restartpoints.  Make startup process drop its
  pendingOpsTable when bgwriter is launched to avoid that.  Update
  minimum recovery point one last time when leaving archive recovery.
  It won't be updated by the end-of-recovery checkpoint because
  XLogFlush() sees us as out of recovery already.  This fixes bug
  #4879 reported by Fujii Masao.

== Abgelehnte Patches (bis jetzt) ==

No one was disappointed this week :-)

== Eingesandte Patches ==

ITAGAKI Takahiro sent in a patch to add missing_ok to GetUserMapping()
and makes dblink use it.

Robert Haas sent in a patch to make pg_listener.h conform better to
PostgreSQL coding standards.

Zoltan Boszormenyi sent in a patch to add dynamic cursors and SQLDA to
ECPG.

Fujii Masao and Magnus Hagander traded patches to remove pg_standby's
dependency on libpq.so.5.

KaiGai Kohei sent in a patch to add DAC security checks on large
objects.

Scara Maccai sent in a patch to add a query progress indicator.

Tsutomu Yamada sent in a patch to make PostgreSQL more portable to
64-bit platforms.

Nedyalko Borisov sent in a JOIN optimization for partitioned tables.

Robert Haas sent in a patch to make generated header files work better
with building subsystems.