Robert Haas <robertmh...@gmail.com> writes:
> On Thu, Mar 14, 2019 at 3:13 AM Amit Langote
> <langote_amit...@lab.ntt.co.jp> wrote:
>> I'm curious why DestroyPartitionDirectory doesn't do
>> hash_destroy(pdir->pdir_hash)?
> What would be the point? It's more efficient to let context teardown
> take care of it.
Agreed, but the comments in this area are crap. Why doesn't
CreatePartitionDirectory say something like
* The object lives inside the given memory context and will be
* freed when that context is destroyed. Nonetheless, the caller
* must *also* ensure that (unless the transaction is aborted)
* DestroyPartitionDirectory is called before that happens, else
* we may leak some relcache reference counts.
It's completely not acceptable that every reader of this code should
have to reverse-engineer these design assumptions, especially given
how shaky they are.
There's an independent question as to whether the planner's use of
the feature is specifying a safe memory context. Has this code been
exercised under GEQO?
regards, tom lane
