This patch merges the PostgreSQL 9.5 implementation of SSE4.2 into GPDB.
The SSE4.2 implementation was lifted right out of PostgreSQL without
change to make merging later PostgreSQL releases easier.

Update win32 configuration to support SSE4.2 runtime checks

This change was pulled from "src/include/pg_config.h.win32" from the
commits below.

configure.in changes, determining if cpu instruction for perfomring runtime
checks are availible, and moving some of the code to port SSE4.2 came from the
following PostgreSQL commits:

commit 3dc2d62d
Author: Heikki Linnakangas <heikki.linnakangas@iki.fi>
Date:   Tue Apr 14 17:05:03 2015 +0300

    Use Intel SSE 4.2 CRC instructions where available.

    Modern x86 and x86-64 processors with SSE 4.2 support have special
    instructions, crc32b and crc32q, for calculating CRC-32C. They greatly
    speed up CRC calculation.

    Whether the instructions can be used or not depends on the compiler and the
    target architecture. If generation of SSE 4.2 instructions is allowed for
    the target (-msse4.2 flag on gcc and clang), use them. If they are not
    allowed by default, but the compiler supports the -msse4.2 flag to enable
    them, compile just the CRC-32C function with -msse4.2 flag, and check at
    runtime whether the processor we're running on supports it. If it doesn't,
    fall back to the slicing-by-8 algorithm. (With the common defaults on
    current operating systems, the runtime-check variant is what you get in
    practice.)

    Abhijit Menon-Sen, heavily modified by me, reviewed by Andres Freund.

commit b4eb2d16
Author: Heikki Linnakangas <heikki.linnakangas@iki.fi>
Date:   Tue Apr 14 19:56:00 2015 +0300
	On gcc and clang, the _mm_crc32_u8 and _mm_crc32_u64 intrinsics are not
	defined at all, when not building with -msse4.2. But on icc, they are.
	So we cannot assume that if those intrinsics are defined, we can always use
	them safely, we might still need the runtime check.

	To fix, check if the __SSE_4_2__ preprocessor symbol is defined. That's
	supposed to be defined only when the compiler is targeting a processor that
	has SSE 4.2 support.

	Per buildfarm members fulmar and okapi.

I already removed support from the backend and from tools in src/bin
earlier, but missed these tools in contrib.

I removed everything else from gpupgrade earlier, but missed this file
for reason.

1. Remove unnecessary ORDER BYs (one of them still needed, as gpdiff.pl got
confused by the ORDER BY within subquery, thinking that the order of the
overall query was well-ordered).

2. Move GPDB-added test queries to separate test file, union_gp.sql.

This is part of long-standing effort ot eliminate the separate bugbuster
suite.

There are similar tests in the partition1 test already, so keep them
together.

These exactly same queries from bugbuster 'partition' test were already in
the main suite's 'partition' test (and even had some comments there).

This test takes a very long time, so by splitting it and running other tests
in parallel, we can hopefully utilize multiple CPUs better during regression
tests, and finish the tests earlier.

There's no big principle on where exactly I split this, but I did take care
that the new 'partition1' test case doesn't need different expected output
files for the ORCA and non-ORCA case. That always makes maintenance easier.

These tests query pg_locks and fail if there are concurrent transactions
holding locks. Isolate them to separate file, so that we can safely run
other tests in parallel.

When gethostbyaddr fails in gpssh-exkeys, print the name
of the host that failed lookup for easier troubleshooting.

When the cluster has multiple segments, planner now mistakenly calculates the
row number of catalog tables. One example is: SELECT COUNT(*) FROM pg_class may
say the table has 200 rows of records, while EXPLAIN SELECT * FROM pg_class
would say there are 200 * N rows of records, even if you have already issued ANALYZE
upon pg_class and restarted the session(to reload the relation cache), where N
is the number of primary segments.

Another bug is for query like SELECT * FROM gp_dist_random('pg_class'), in this
case, planner should assume the size of pg_class to be 200 * N, instead of the
current 200.

Only minor formatting and comment changes, everything else in this merge
commit we had already backported earlier.

Conflicts:
	configure
	configure.in
	doc/TODO
	doc/src/FAQ/TODO.html
	doc/src/sgml/Makefile
	doc/src/sgml/installation.sgml
	doc/src/sgml/ref/initdb.sgml
	doc/src/sgml/release.sgml
	src/Makefile.shlib
	src/backend/executor/execQual.c
	src/backend/optimizer/path/costsize.c
	src/backend/optimizer/plan/initsplan.c
	src/backend/parser/parse_expr.c
	src/backend/parser/parser.c
	src/backend/port/Makefile
	src/backend/utils/adt/float.c
	src/backend/utils/adt/xml.c
	src/bin/initdb/initdb.c
	src/include/utils/xml.h
	src/interfaces/ecpg/test/expected/compat_informix-charfuncs.c
	src/interfaces/ecpg/test/expected/compat_informix-dec_test.c
	src/interfaces/ecpg/test/expected/compat_informix-rfmtdate.c
	src/interfaces/ecpg/test/expected/compat_informix-rfmtlong.c
	src/interfaces/ecpg/test/expected/compat_informix-rnull.c
	src/interfaces/ecpg/test/expected/compat_informix-test_informix.c
	src/interfaces/ecpg/test/expected/compat_informix-test_informix2.c
	src/interfaces/ecpg/test/expected/connect-test1.c
	src/interfaces/ecpg/test/expected/connect-test2.c
	src/interfaces/ecpg/test/expected/connect-test3.c
	src/interfaces/ecpg/test/expected/connect-test4.c
	src/interfaces/ecpg/test/expected/connect-test5.c
	src/interfaces/ecpg/test/expected/pgtypeslib-dt_test.c
	src/interfaces/ecpg/test/expected/pgtypeslib-dt_test2.c
	src/interfaces/ecpg/test/expected/pgtypeslib-num_test.c
	src/interfaces/ecpg/test/expected/pgtypeslib-num_test2.c
	src/interfaces/ecpg/test/expected/preproc-comment.c
	src/interfaces/ecpg/test/expected/preproc-define.c
	src/interfaces/ecpg/test/expected/preproc-init.c
	src/interfaces/ecpg/test/expected/preproc-type.c
	src/interfaces/ecpg/test/expected/preproc-variable.c
	src/interfaces/ecpg/test/expected/preproc-whenever.c
	src/interfaces/ecpg/test/expected/sql-array.c
	src/interfaces/ecpg/test/expected/sql-binary.c
	src/interfaces/ecpg/test/expected/sql-code100.c
	src/interfaces/ecpg/test/expected/sql-copystdout.c
	src/interfaces/ecpg/test/expected/sql-define.c
	src/interfaces/ecpg/test/expected/sql-desc.c
	src/interfaces/ecpg/test/expected/sql-dynalloc.c
	src/interfaces/ecpg/test/expected/sql-dynalloc2.c
	src/interfaces/ecpg/test/expected/sql-dyntest.c
	src/interfaces/ecpg/test/expected/sql-execute.c
	src/interfaces/ecpg/test/expected/sql-fetch.c
	src/interfaces/ecpg/test/expected/sql-func.c
	src/interfaces/ecpg/test/expected/sql-indicators.c
	src/interfaces/ecpg/test/expected/sql-quote.c
	src/interfaces/ecpg/test/expected/sql-show.c
	src/interfaces/ecpg/test/expected/thread-thread.c
	src/interfaces/ecpg/test/expected/thread-thread_implicit.c
	src/interfaces/libpq/Makefile
	src/test/regress/expected/xml.out
	src/test/regress/expected/xml_1.out
	src/test/regress/sql/xml.sql

Just the bump in copyright year in this, nothing interesting. And almost
every file already carried a higher year already anyway, so I didn't
touch those. I did update the PostgreSQL CVS tags where applicable, though,
just to reduce the number of merge conflicts in future.

Noth much interesting here, but I wanted to stop before the next upstream
commit, which is a mechanical Copyright year bump.

Conflicts:
	contrib/btree_gist/btree_gist.sql.in
	doc/FAQ
	doc/FAQ_DEV
	doc/FAQ_japanese
	doc/bug.template
	doc/src/FAQ/FAQ.html
	doc/src/FAQ/FAQ_DEV.html
	doc/src/FAQ/FAQ_japanese.html
	doc/src/sgml/ref/pg_ctl-ref.sgml
	doc/src/sgml/ref/postgres-ref.sgml
	doc/src/sgml/ref/reindex.sgml
	doc/src/sgml/release.sgml
	src/backend/access/gin/ginvacuum.c
	src/backend/access/heap/tuptoaster.c
	src/backend/main/main.c
	src/backend/postmaster/postmaster.c
	src/backend/tcop/postgres.c
	src/backend/utils/adt/float.c
	src/backend/utils/adt/regexp.c
	src/bin/pg_ctl/pg_ctl.c
	src/bin/pg_dump/dumputils.c
	src/include/catalog/pg_type.h
	src/include/pg_config.h.win32
	src/include/utils/builtins.h
	src/interfaces/ecpg/compatlib/Makefile
	src/interfaces/ecpg/ecpglib/Makefile
	src/interfaces/ecpg/pgtypeslib/Makefile
	src/interfaces/ecpg/preproc/Makefile
	src/interfaces/libpq/Makefile
	src/interfaces/libpq/libpq.rc.in
	src/port/win32ver.rc
	src/test/regress/pg_regress.c
	src/tools/RELEASE_CHANGES
	src/tools/copyright
	src/tools/msvc/Project.pm

This batch includes:

* A series of tweaks to handling float corner case. Our code is now roughly
  the same as upstream tip of 8.3 stable branch. As part of this merge,
  I reverted some unnecessary changes to the float regression test, so that
  it now closely matches upstream too.

* Widen money datatype to 64 bits. Not that exciting, but noteworthy because
  that changes the on-disk format, and would need to be dealt with if we
  tried to implement in-place upgrade crossing this commit. (There are more
  such changes upcoming from PostgreSQL 8.3, but this is the first)

* Changes to the API of smgr.c and md.c. The big change is that mdread() on
  a block beoynd EOF no longer extends the file with zeros, but throws an
  error. A couple of places had to be tweaked for that. Also, I started
  getting errors from that from a TID scan test case in portals_updatable
  test case. If a table was partitioned, and the current TID was pointing e.g.
  to block 10, we would try to fetch block 10 on all partitions, not only the
  partition the current row came from. That now lead to an ERROR in mdread().
  I suspect this has been wrong all along, but no-one's noticed.

Conflicts:
	configure
	src/backend/access/hash/hashpage.c
	src/backend/access/nbtree/nbtsort.c
	src/backend/commands/tablecmds.c
	src/backend/storage/smgr/md.c
	src/backend/storage/smgr/smgr.c
	src/backend/utils/adt/float.c
	src/backend/utils/adt/int.c
	src/backend/utils/adt/network.c
	src/include/catalog/catversion.h
	src/include/catalog/pg_type.h
	src/include/port/win32.h
	src/include/storage/smgr.h
	src/test/regress/expected/float4.out
	src/test/regress/expected/float8.out
	src/test/regress/expected/rules.out

Move a comment outside local block, so that the indentation matches
upstream. Remove some unnecessary #includes.

This batch includes a bunch of misc fixes, but the most notable change is
the support for type modifiers of user-defined types. There were also a
bunch of XML-type related fixes, but most of those were not applicable
(i.e. had them already), as I backported the 9.1 XML implementation in toto
already.

Conflicts:
	contrib/pgbench/pgbench.c
	contrib/tsearch2/dict_thesaurus.c
	doc/TODO
	doc/src/FAQ/TODO.html
	doc/src/sgml/catalogs.sgml
	doc/src/sgml/ddl.sgml
	doc/src/sgml/errcodes.sgml
	doc/src/sgml/installation.sgml
	doc/src/sgml/keywords.sgml
	doc/src/sgml/maintenance.sgml
	doc/src/sgml/plpgsql.sgml
	doc/src/sgml/ref/create_type.sgml
	doc/src/sgml/ref/vacuum.sgml
	doc/src/sgml/rules.sgml
	src/backend/access/common/tupdesc.c
	src/backend/access/nbtree/README
	src/backend/access/nbtree/nbtutils.c
	src/backend/commands/tablecmds.c
	src/backend/commands/typecmds.c
	src/backend/executor/execMain.c
	src/backend/executor/execQual.c
	src/backend/executor/execUtils.c
	src/backend/executor/nodeBitmapHeapscan.c
	src/backend/executor/nodeFunctionscan.c
	src/backend/executor/nodeIndexscan.c
	src/backend/executor/nodeResult.c
	src/backend/executor/nodeSeqscan.c
	src/backend/executor/nodeSubplan.c
	src/backend/executor/nodeSubqueryscan.c
	src/backend/executor/nodeTidscan.c
	src/backend/executor/nodeValuesscan.c
	src/backend/executor/spi.c
	src/backend/nodes/copyfuncs.c
	src/backend/nodes/equalfuncs.c
	src/backend/nodes/makefuncs.c
	src/backend/nodes/outfuncs.c
	src/backend/nodes/readfuncs.c
	src/backend/optimizer/util/clauses.c
	src/backend/parser/gram.y
	src/backend/parser/keywords.c
	src/backend/parser/parse_coerce.c
	src/backend/parser/parse_expr.c
	src/backend/parser/parse_relation.c
	src/backend/parser/parse_target.c
	src/backend/parser/parse_type.c
	src/backend/storage/buffer/localbuf.c
	src/backend/utils/adt/arrayutils.c
	src/backend/utils/adt/date.c
	src/backend/utils/adt/float.c
	src/backend/utils/adt/format_type.c
	src/backend/utils/adt/ruleutils.c
	src/backend/utils/adt/timestamp.c
	src/backend/utils/adt/varbit.c
	src/backend/utils/adt/varchar.c
	src/backend/utils/adt/xml.c
	src/backend/utils/cache/lsyscache.c
	src/backend/utils/cache/relcache.c
	src/backend/utils/mb/mbutils.c
	src/backend/utils/misc/guc.c
	src/backend/utils/mmgr/aset.c
	src/bin/pg_dump/pg_dump.c
	src/bin/psql/command.c
	src/bin/psql/input.c
	src/bin/psql/mbprint.c
	src/include/catalog/catversion.h
	src/include/catalog/pg_class.h
	src/include/catalog/pg_proc.h
	src/include/catalog/pg_type.h
	src/include/mb/pg_wchar.h
	src/include/nodes/nodes.h
	src/include/nodes/parsenodes.h
	src/include/nodes/primnodes.h
	src/include/parser/parse_coerce.h
	src/include/parser/parse_type.h
	src/include/utils/array.h
	src/include/utils/builtins.h
	src/include/utils/timestamp.h
	src/include/utils/varbit.h
	src/include/utils/xml.h
	src/interfaces/libpq/Makefile
	src/pl/plpgsql/src/pl_exec.c
	src/test/regress/expected/create_type.out
	src/test/regress/expected/horology.out
	src/test/regress/expected/opr_sanity.out
	src/test/regress/expected/xml.out
	src/test/regress/expected/xml_1.out
	src/test/regress/sql/create_type.sql
	src/test/regress/sql/opr_sanity.sql
	src/tools/findoidjoins/README
	src/tools/msvc/Solution.pm
	src/tools/pgindent/pgindent

Fix: Immutable function mregr_coef returns different results

The immutable function was returning different values, because it was allocating memory and using a table whose entries may have a garbage initial value.
The fix initiates the entries of the table to zero.

In the gpmgmt1 bugbuster test, the gpstate -f line makes a
connection to the template1 database which can cause the
other tests in gpmgmt1's parallel test group to fail.  An
example of this is when metadata_track tries to create
databases and fails as template1 is being accessed.

gpstate -f in gpmgmt1 bugbuster test has been removed.

The gpdbrestore -G flag has been modified from a boolean flag
to one that accepts an optional argument.  Passing it the
argument "include" will carry out the old -G functionality,
restoring global objects in addition to performing a normal
restore; passing it no argument will do the same, to retain
backwards compatibility.  Passing it the argument "only" will
restore global objects but will not restore anything else.

Authors: James McAtamney and Jimmy Yih

Currently gpdbrestore does not allow tables to be restored from one schema to another.
Eg:
testschema.testtable => anotherschema.testtable

This merges operator families from upstream.

Conflicts:
	contrib/intarray/_int.sql.in
	contrib/intarray/uninstall__int.sql
	doc/src/sgml/catalogs.sgml
	doc/src/sgml/indexam.sgml
	doc/src/sgml/indices.sgml
	doc/src/sgml/xoper.sgml
	src/backend/access/hash/hashfunc.c
	src/backend/access/index/indexam.c
	src/backend/access/nbtree/nbtsearch.c
	src/backend/catalog/Makefile
	src/backend/catalog/dependency.c
	src/backend/catalog/namespace.c
	src/backend/commands/indexcmds.c
	src/backend/commands/opclasscmds.c
	src/backend/commands/operatorcmds.c
	src/backend/commands/tablecmds.c
	src/backend/executor/nodeIndexscan.c
	src/backend/executor/nodeMergejoin.c
	src/backend/nodes/copyfuncs.c
	src/backend/nodes/outfuncs.c
	src/backend/nodes/readfuncs.c
	src/backend/optimizer/path/indxpath.c
	src/backend/optimizer/path/joinpath.c
	src/backend/optimizer/plan/createplan.c
	src/backend/optimizer/plan/planagg.c
	src/backend/optimizer/plan/subselect.c
	src/backend/optimizer/util/clauses.c
	src/backend/optimizer/util/pathnode.c
	src/backend/optimizer/util/predtest.c
	src/backend/optimizer/util/restrictinfo.c
	src/backend/utils/adt/ruleutils.c
	src/backend/utils/adt/selfuncs.c
	src/backend/utils/cache/catcache.c
	src/backend/utils/cache/lsyscache.c
	src/backend/utils/cache/relcache.c
	src/backend/utils/cache/syscache.c
	src/backend/utils/sort/tuplesort.c
	src/bin/initdb/initdb.c
	src/bin/pg_dump/pg_dump.c
	src/include/catalog/catversion.h
	src/include/catalog/indexing.h
	src/include/catalog/pg_amop.h
	src/include/catalog/pg_amproc.h
	src/include/catalog/pg_opclass.h
	src/include/catalog/pg_operator.h
	src/include/nodes/plannodes.h
	src/include/nodes/relation.h
	src/include/optimizer/pathnode.h
	src/include/utils/lsyscache.h
	src/include/utils/selfuncs.h
	src/include/utils/syscache.h
	src/test/regress/expected/opr_sanity.out
	src/test/regress/expected/sanity_check.out
	src/tools/findoidjoins/README

We should be more careful with string substitutions. The old code generates
stack trace without giving out the information which host has the problem
due to the bug in code Zendesk ticket
https://discuss.zendesk.com/agent/tickets/17783

To have more parallelism and hence speed up the overall test suite a tiny
bit.

Also, remove unnecessary CREATE DATABASE and \c command from DMLstmnt test.

This speeds up the overall runtime, when you have multiple CPUs so that
the tests can run in parallel.

The test in the main suite is identical to the one in bugbuster, except that
the bugbuster test had to set up tenk1 table while in the main suite it's
created by the Postgres tests. No point in testing the same thing twice.

The long-term goal is to move all relevant tests from bugbuster, remove
any that are not relevant, and get rid of bugbuster as a separate test
suite. This is one small step in that direction.

Rename regex to regex_gp while we're at it. Later versions of PostgreSQL
have a test case for regular expressions called 'regex', and having both
'regex' and 'regexp' would be confusing.

Try to put tests into larger groups, and try to run long-running tests
in parallel. This utilizes multiple CPUs better, making the overall suite
finish faster. On my laptop, this reduced the overall runtime of the
greenplum_schedule (i.e. excluding the Postgres tests in parallel_schedule)
from about 4 minutes to 3 minutes.

This is fixed in PostgreSQL and I found parts of those fixes pulled in already.
This commit pulls in the remnants.  The SEGV was found at a customer site
where default_tablespace was configured for a role.  While new connections were
made by the role, GRANT/REVOKE on the same tablespace was running concurrently.
The assign hook for default_tablespace invoked elog(ERROR) but target for
siglongjmp was not initialized, causing a segv and PANIC.

Some unit tests had to be changed to accommodate the fix.

Original commits from PostgreSQL:
88f1fd29
79ca7ffe

Currently GPtransfer has count validation which runs after every table transfer. It does not do any checking at the end of the transfer to the destination. Therefore it does not validate that the tables haven't changed in the meantime.

To resolve the issue, an additional table row count validation has been added with all the existing validation options. This count validation does a final pass after the transfer is completely done to make sure the tables haven't changed in the meantime. Therefore, the default is to compare the row count of tables copied to the destination databases with the tables in the source database.

Also an option '--no-final-count’ has beed introduced for users who do not want to run the final count validation by default. The option '--no-final-count’ will disable table row count validation that is performed after GPtransfer completes copying database objects to the target database.

In a sequence, if NO CYCLE is specified, any calls to nextval after the sequence has reached its maximum value will return an error.

Commit 9114c65f moved dumping
tablespaces under the -F option to ensure that tablespaces would
only be dumped when the filespaces upon which they depended were
dumped.  It was decided that, for backwards compatibility, the
original behavior would be restored.

Authors: Jimmy Yih and James McAtamney