Commits backported from upstream are listed below.

	commit 87f2ad13
	Author: Tom Lane <tgl@sss.pgh.pa.us>
	Date:   Sun Mar 27 12:51:04 2011 -0400

	    Fix plpgsql to release SPI plans when a function or DO block is freed.

	    This fixes the gripe I made a few months ago about DO blocks getting
	    slower with repeated use.  At least, it fixes it for the case where
	    the DO block isn't aborted by an error.  We could try running
	    plpgsql_free_function_memory() even during error exit, but that seems
	    a bit scary since it makes a lot of presumptions about the data
	    structures being in good shape.  It's probably reasonable to assume
	    that repeated failures of DO blocks isn't a performance-critical case.

	commit 42b2907d
	Author: Tom Lane <tgl@sss.pgh.pa.us>
	Date:   Sun Nov 29 03:02:27 2009 +0000

	    Add support for anonymous code blocks (DO blocks) to PL/Perl.

	    Joshua Tolley, reviewed by Brendan Jurd and Tim Bunce

	commit 9048b731
	Author: Tom Lane <tgl@sss.pgh.pa.us>
	Date:   Tue Sep 22 23:43:43 2009 +0000

	    Implement the DO statement to support execution of PL code without having
	    to create a function for it.

	    Procedural languages now have an additional entry point, namely a function
	    to execute an inline code block.  This seemed a better design than trying
	    to hide the transient-ness of the code from the PL.  As of this patch, only
	    plpgsql has an inline handler, but probably people will soon write handlers
	    for the other standard PLs.

	    In passing, remove the long-dead LANCOMPILER option of CREATE LANGUAGE.

	    Petr Jelinek

Update to avoid misunderstanding. `--with-perl` means building Perl modules (PL/Perl) instead of perl runtime, `--with-python` means building Python modules (PL/Python) instead of python runtime.

The trgm2int() function was backported from PostgreSQL 8.4, but never used.
And the TRGMINT() macro was removed as part of that backport, without
changing the callers.

This makes the code to compile, and the regression tests to pass, except
for the fact that GPDB doesn't support GIN, which produces an error.

Fixes issue #491, reported by @digoal.

It doesn't seem to be needed anymore. Window and Motion nodes have a
suitable targetlist nowadays, so no need to reach out to their child node
for one.

I'm not sure what the original problem was, because PostgreSQL is happy
without the rtoffset arg, and everything seems to work when I revert it.
I dug the original reproduction script referenced from when that change
was originally made and tested that with this revert, and it also worked
fine. Keeping the code as close to upstream as possible makes merging
easier.

Makes merging a bit easier.

ReadBuffer_Resync() and BufferAlloc_Resync() were not materially different
from plain ReadBuffer() and BufferAlloc(), so just use the latter directly.

In the passing, also:
* rename BufferAlloc_Internal to BufferAlloc_common, like it's called in
  newer upstream versions.
* revert some spurious whitespace and comment differences vs. upstream.

* the "if (smgr_relation == NULL)" is not needed, because smgr_relation is
  always NULL at this point.
* smgropen() cannot return NULL, so the code that checked for that
  was dead.
* smgropen() sets smgr_rnode. No need for the caller to set it.

Now s3ext supports all amazon s3 regions.
http://docs.aws.amazon.com/general/latest/gr/rande.html#s3_region

fix read_callback()
fix memleaks, thanks to cppcheck
check every pointer returned from malloc()
remove unused functions, add uri encoding codes
strftime()'s argument max includes the terminating null byte

This is pretty trivial, but every little helps.

We don't particularly care about Windows, but this makes merging slightly
easier.

Resurrect the make_database_relative() function, for constructing a
temporary file's path. This refactors the code back to the way it is in
the upstream, to make merging easier. No change in behaviour.

All callers passed 1, so it didn't do anything interesting.

Will make merging code to fd.c easier.

Closes #449

Closes #488.

Most of these test additions are inspired from Pivotal's internal
testing and needed to be added to the open source installcheck to
give the community more test coverage.

If a PathKey contains a constant member, it can be evaluated without
any entries in the target list, and can always be returned in
cdbpullup_findPathKeyExprInTargetList. This fixes the "Unexpected intarnal
error" you got with the included test query.

Closes issue #348, reported by liruto. Thanks for the report!

This check ensured that distributed transaction log does not have
missing page, and if it does, initialize to zero.  This was only
relevant in old Greeplum releases and is no longer required.

Closes #476.

Changing WorkfileQueryspace_ReleaseEntry assert to check querySpaceNestingLevel == 0 when we release the entry.

When DELETEing or UPDATEing an inherited table, some tables in the
inheritance tree might need an explicit Motion node to bring the targeted
tuples back to the segment where they reside, and some might not. The code
to build the plan handled that correctly, but this assertion incorrecly
assumed that it's all or nothing.

Remove the assertion, as it doesn't seem very useful in the first place.
The code that inserted the Motion nodes is just above the assertion, and
the assertion was basically just testing the same thing that the code
just did, and not some general invariant that should always hold.

Fixes issue #332

It is left-behind work of "5b2af3cf"

This fix contains two part
1. Backport 30bf4689 from upstream
2. Sync the implementation of fillPGconn() with postgresql which is simpler and more robust.

QD should not expect end-of-stream comes from QEs who is not members of
direct dispatch and should not report warning message.

More fallout from the equivalence classes merge.

It's easy to see that the code was broken, as it would always return 'true'
regardless of what happened in the for-loop.