Обсуждение: Is this an oversight in reparameterizing Memoize path?
When reviewing other patch I noticed there might be an oversight for
MemoizePath in reparameterize_path. In reparameterize_path we are
supposed to increase the path's parameterization to required_outer.
However, AFAICS for MemoizePath we just re-create the same path thus its
parameterization does not get increased.
I'm not sure if this has consequences in practice. Just from reading
the codes, it seems this may cause assertion failure after the call of
reparameterize_path.
Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
Thanks
Richard
MemoizePath in reparameterize_path. In reparameterize_path we are
supposed to increase the path's parameterization to required_outer.
However, AFAICS for MemoizePath we just re-create the same path thus its
parameterization does not get increased.
I'm not sure if this has consequences in practice. Just from reading
the codes, it seems this may cause assertion failure after the call of
reparameterize_path.
Assert(bms_equal(PATH_REQ_OUTER(path), required_outer));
Thanks
Richard
Richard Guo <guofenglinux@gmail.com> writes: > When reviewing other patch I noticed there might be an oversight for > MemoizePath in reparameterize_path. In reparameterize_path we are > supposed to increase the path's parameterization to required_outer. > However, AFAICS for MemoizePath we just re-create the same path thus its > parameterization does not get increased. Yeah, that sure looks wrong. At minimum we should be recursively fixing the subpath. (It looks like doing that and re-calling create_memoize_path might be sufficient.) According to [1] our code coverage for reparameterize_path is just awful. MemoizePath in reparameterize_pathlist_by_child isn't tested either ... regards, tom lane [1] https://coverage.postgresql.org/src/backend/optimizer/util/pathnode.c.gcov.html