Re: add more frame types in window functions (ROWS)

2009/11/19 Hitoshi Harada <umi(dot)tanuki(at)gmail(dot)com>:
> 2009/11/19 Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk>:
>> Small nitpicks: there are some comments not updated (e.g. the
>> big one at the start of eval_windowaggregates). A couple of lines are
>> commented-out using C++ comments.

Fixed. Document patch is included as well.

>> - missing _readWindowFrameDef function (all nodes that are output
>> from parse analysis must have both _read and _out functions,
>> otherwise views can't work)

I changed my mind and WindowFrameDef is alive only in initial parser
stage so that _readWindowFrameDef is now unnecessary. Information of
startOffset and endOffset will be copied to WindowClause members.

>> - the A_Const nodes should probably be transformed to Const nodes in
>> parse analysis, since A_Const has no _read/_out functions, which
>> means changing the corresponding code in the executor.

Fixed.

>> - ruleutils.c not updated to deparse the newly added window options

Fixed.

>> - leaks memory like it's going out of style

As earlier mail, I added aggcontext to WindowAggState.

Still it doesn't contain RANGE ... PRECEDING / FOLLOWING. If it's not
acceptable for commit without RANGE value support, I'd agree with
that. I'm planning to work on that until the next CommitFest.

Regards,

--
Hitoshi Harada