]> granicus.if.org Git - postgresql/commit
Wrap appendrel member outputs in PlaceHolderVars in additional cases.
authorTom Lane <tgl@sss.pgh.pa.us>
Wed, 9 Nov 2011 02:14:28 +0000 (21:14 -0500)
committerTom Lane <tgl@sss.pgh.pa.us>
Wed, 9 Nov 2011 02:14:28 +0000 (21:14 -0500)
commit7097e6c4a48bc02155a9a52c570be67b30b30c9a
treedc312612d44c80753d22b2a4e89bc3a28ed8271b
parentb1c701c90922d828a9fa4335dd36ab2588bad3a6
Wrap appendrel member outputs in PlaceHolderVars in additional cases.

Add PlaceHolderVar wrappers as needed to make UNION ALL sub-select output
expressions appear non-constant and distinct from each other.  This makes
the world safe for add_child_rel_equivalences to do what it does.  Before,
it was possible for that function to add identical expressions to different
EquivalenceClasses, which logically should imply merging such ECs, which
would be wrong; or to improperly add a constant to an EquivalenceClass,
drastically changing its behavior.  Per report from Teodor Sigaev.

The only currently known consequence of this bug is "MergeAppend child's
targetlist doesn't match MergeAppend" planner failures in 9.1 and later.
I am suspicious that there may be other failure modes that could affect
older release branches; but in the absence of any hard evidence, I'll
refrain from back-patching further than 9.1.
src/backend/optimizer/plan/planmain.c
src/backend/optimizer/prep/prepjointree.c
src/backend/optimizer/util/placeholder.c
src/include/optimizer/placeholder.h
src/test/regress/expected/inherit.out
src/test/regress/sql/inherit.sql