calcite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Haisheng Yuan" <h.y...@alibaba-inc.com>
Subject Re: Re: [DISCUSS] IS NOT DISTINCT FROM rewrite
Date Thu, 06 Jun 2019 00:34:56 GMT
>  if sending SQL to a database that does not understand IS NOT DISTINCT FROM
That is sadly true. I found that Hive just starts to support INDF since 3.0.0. But dy default
expanding INDF is still questionable. Their incapability should not force Calcite expand INDF
by default.

> If both arguments are not null, it probably makes sense to rewrite “x IS NOT DISTINCT
FROM y” to “x = y”,
I agree. But it is can be done with RexSimplify easily, expanding it or using a rule FilterRemoveIsNotDistinctFromRule
is an overkill, IMHO.

- Haisheng

------------------------------------------------------------------
发件人:Julian Hyde<jhyde@apache.org>
日 期:2019年06月06日 05:09:14
收件人:dev<dev@calcite.apache.org>
主 题:Re: [DISCUSS] IS NOT DISTINCT FROM rewrite

My instinct is that we should leave it unexpanded. And that we should recognize “equals-like
operators”, so that a planner rule originally written for ‘=‘ could easily be expanded
to also apply to ‘is not distinct from’.

Of course there would be a way of expanding it that we could use if circumstances required
it — e.g. if sending SQL to a database that does not understand IS NOT DISTINCT FROM —
but we would not expand it by default.

If both arguments are not null, it probably makes sense to rewrite “x IS NOT DISTINCT FROM
y” to “x = y”, because the latter is more common and no less simple.

Julian



> On Jun 5, 2019, at 1:46 PM, Haisheng Yuan <h.yuan@alibaba-inc.com> wrote:
> 
> I see INDF is rewritten to OR, and FilterRemoveIsNotDistinctFromRule rewrites INDF to
CASE expression. Why do we want to do that? To simplify expression like "a is not distinct
from b or a = b"? Then we spend a lot effort to convert OR/CASE back to INDF. 
> 
> I am curious what is the motivation to rewrite INDF. Does it really help a lot in production?
I would like to hear the use cases if it does.
> 
> - Haisheng
> 
Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message