ads' corner

== Angewandte Patches ==

Magnus Hagander committed:

- In pgsql/src/tools/msvc/Project.pm, enable the use of multiple
  CPUs/cores when building on MSVC.  This only affects the C compiler
  step - we still only build one target at a time.

- In pgsql/src/port/exec.c, fix minor memory leak in Win32 SID
  handling functions. Not a big issue since it's only called during
  process startup, thus no backpatch.  Found by TAKATSUKA Haruka,
  patch by Magnus Hagander and Andrew Chernow.

- In pgsql/contrib/pgbench/pgbench.c, make sure FD_SETSIZE is set
  before we include any Windows header files.  Josh Williams.

Tom Lane committed:

- Add system catalog columns pg_constraint.conindid and
  pg_trigger.tgconstrindid.  conindid is the index supporting a
  constraint.  We can use this not only for unique/primary-key
  constraints, but also foreign-key constraints, which depend on the
  unique index that constrains the referenced columns.  tgconstrindid
  is just copied from the constraint's conindid field, or is zero for
  triggers not associated with constraints.  This is mainly intended
  as infrastructure for upcoming patches, but it has some virtue in
  itself, since it exposes a relationship that you formerly had to
  grovel in pg_depend to determine.  I simplified one
  information_schema view accordingly.  (There is a pg_dump query that
  could also use conindid, but I left it alone because it wasn't clear
  it'd get any faster.)

- In pgsql/src/backend/storage/ipc/procarray.c, fix a thinko
  introduced into CountActiveBackends by a recent patch: we should
  ignore NULL array entries, not non-NULL ones.  This had the effect
  of disabling commit_delay, and could have caused a crash in the rare
  race condition the patch was intended to fix.  Bug report and
  diagnosis by Jeff Janes, in bug #4952.

- Support deferrable uniqueness constraints.  The current
  implementation fires an AFTER ROW trigger for each tuple that looks
  like it might be non-unique according to the index contents at the
  time of insertion.  This works well as long as there aren't many
  conflicts, but won't scale to massive unique-key reassignments.
  Improving that case is a TODO item.  Dean Rasheed.

- In pgsql/src/backend/utils/adt/date.c, fix time_part and timetz_part
  (ie, EXTRACT() for those datatypes) to include a fractional part in
  the output for MILLISECOND and SECOND cases, rather than truncating
  the source value.  This is what the float-timestamp code has always
  done, and it was clearly the code author's intent to do the same for
  integer timestamps, but he forgot about integer division in C.  The
  other datatypes supported by EXTRACT() already do this correctly.
  Backpatch to 8.4, so that the default (integer) behavior of that
  branch will match the default (float) behavior of older branches.
  Arguably we should patch further back, but it's possible that
  applications are expecting the broken behavior in older branches.
  8.4 is new enough that expectations shouldn't be too settled.  Per
  report from Greg Stark.

- Merge the Constraint and FkConstraint node types into a single type.
  This was foreseen to be a good idea long ago, but nobody had got
  round to doing it.  The recent patch for deferred unique constraints
  made transformConstraintAttrs() ugly enough that I decided it was
  time.  This change will also greatly simplify parsing of deferred
  CHECK constraints, if anyone ever gets around to implementing that.
  While at it, add a location field to Constraint, and use that to
  provide an error cursor for some of the constraint-related error
  messages.

- Create a multiplexing structure for signals to Postgres child
  processes.  This patch gets us out from under the Unix limitation of
  two user-defined signal types.  We already had done something
  similar for signals directed to the postmaster process; this adds
  multiplexing for signals directed to backends and auxiliary
  processes (so long as they're connected to shared memory).  As proof
  of concept, replace the former usage of SIGUSR1 and SIGUSR2 for
  backends with use of the multiplexing mechanism.  There are still
  some hard-wired definitions of SIGUSR1 and SIGUSR2 for other process
  types, but getting rid of those doesn't seem interesting at the
  moment.  Fujii Masao.

- Improve unique-constraint-violation error messages to include the
  exact values being complained of.  In passing, also remove the
  arbitrary length limitation in the similar error detail message for
  foreign key violations.  Itagaki Takahiro.

- Department of second thoughts: let's show the exact key during
  unique index build failures, too.  Refactor a bit more since that
  error message isn't spelled the same.

- Add ALTER TABLE ... ALTER COLUMN ... SET STATISTICS DISTINCT.
  Robert Haas

Teodor Sigaev committed:

- In pgsql/src/backend/utils/adt/tsquery_rewrite.c, fix incorrect
  cleanup of tsquery in ts_rewrite().  Per bug #4933 by Aaron
  Marcuse-Kubitza.

- Correct calculations of overlap and contains operations over
  polygons.

== Abgelehnte Patches (bis jetzt) ==

Fernando Ike de Oliveira's patch to list languages.  Not updated in
time, returned with feedback.

Tsutomu Yamada's patch to support 64-bit platforms more portably.
Incomplete, returned with feedback.

Dickson S. Guedes's patch to display client and server versions in
psql prompt.  Not updated in time, returned with feedback.

== Eingesandte Patches ==

Teodor Sigaev sent in two patches to split up Oleg Bartunov's patch
for filtering dictionaries and unaccent in textsearch.

Josh Williams sent in a fix to the multi-threaded pgbench patch.

Sergey V. Karpov submitted two versions and Robert Haas and Tom Lane
reworked another revision of the patch to improve dict_xsyn.

Robert Haas sent in another revision of his machine-readable explain
output patch.

Steve Prentice sent in two revisions of a patch to make PL/pgsql's
input parameters writeable.

Zoltan Boszormenyi sent in a patch adding a lock_timeout GUC which
sets how long SELECT ... FOR UPDATE will wait.  This defaults to 0,
meaning "wait forever if needed."

Bernd Helmle sent in a revised patch for mixed named notation in
PL/pgsql.

Tom Lane sent in an improvement to Kevin Grittner's patch to
revise parallel pg_restore's scheduling heuristic by getting rid of
the find_ready_items() scans, thus avoiding O(n^2) behavior.

Dimitri Fontaine sent in an updated version of the ALTER TABLE...ALTER
COLUMN...SET DISTINCT patch.

Marko (johto) Tiikkaja sent in another WIP patch to allow write
operations inside CTEs.

Mark Kirkwood sent in an updated patch to show lock wait statistics.

Joe Conway sent in an updated version of Abhijit Menon-Sen's patch to
implement has_sequence_privilege().

Jeff Davis sent in a rework of Oleg Bartunov's patch to support prefix
in synonym dictionaries.