Horton

Re: TupleTableSlot abstraction

Andres Freund <andres@anarazel.de>

View thread View on postgr.es
From: Andres Freund <andres@anarazel.de>
To: Ashutosh Bapat <ashutosh.bapat@enterprisedb.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>, Robert Haas <robertmhaas@gmail.com>, Haribabu Kommi <kommi.haribabu@gmail.com>, Alvaro Herrera <alvherre@2ndquadrant.com>, Amit Kapila <amit.kapila16@gmail.com>
Date: 2018-08-06T04:45:45Z
Lists: pgsql-hackers 
Hi,

On 2018-08-06 10:08:24 +0530, Ashutosh Bapat wrote:
> I think, I explained why getattr() needs to be a separate callback.
> There's a reason why slot_getattr() more code than just calling
> slot_getsomeattrs() and return the required one - the cases when the
> requested attribute is NULL or is missing from a tuple.  Depending
> upon the tuple layout access to a simple attribute can be optimized
> without spending cycles to extract attributes prior to that one.
> Columnar store (I haven't seen Haribabu's patch), for example, stores
> columns from multiple rows together and thus doesn't have a compact
> version of tuple. In such cases extracting an individual attribute
> directly is far cheaper than extracting all the previous attributes.

OTOH, it means we continually access the null bitmap instead of just
tts_isnull[i].

Your logic about not deforming columns in this case would hold for *any*
deforming of previous columns as well. That's an optimization that we
probably want to implement at some point (e.g. by building a bitmap of
needed columns in the planner), but I don't think we should do it
together with this already large patchset.


> Why should we force the storage API to extract all the attributes in
> such a case?

Because there's no need yet, and it complicates the API without
corresponding benefit.

Greetings,

Andres Freund

Commits

  1. Fix memory leak when inserting tuple at relation creation for CTAS

  2. Build HashState's hashkeys expression with the correct parent.

  3. Move TupleTableSlots boolean member into one flag variable.

  4. Move generic slot support functions from heaptuple.c into execTuples.c.

  5. Remove absolete function TupleDescGetSlot().

  6. Remove function list from prologue of execTuples.c.

  7. Split ExecStoreTuple into ExecStoreHeapTuple and ExecStoreBufferHeapTuple.

  8. Change TupleTableSlot->tts_nvalid to type AttrNumber.

  9. Deduplicate code between slot_getallattrs() and slot_getsomeattrs().

  10. Mention ownership requirements for REFRESH MATERIALIZED VIEW in docs

  11. Spell "partitionwise" consistently.

  12. Fix WHERE CURRENT OF when the referenced cursor uses an index-only scan.