From: Tom Lane <tgl@sss.pgh.pa.us>
To: Oliver Ford <ojford@gmail.com>
cc: Erik Rijkers <er@xs4all.nl>, Pg Hackers <pgsql-hackers@postgresql.org>
Subject: Re: Add RANGE with values and exclusions clauses to the Window
 Functions
In-reply-to: 
 <CAGMVOdtGe=30u6D2WqwvWchE+-yNb9jKnEeG8BHyw7jHiGPR7w@mail.gmail.com>
References: 
 <CAGMVOdu9sivPAxbNN0X+q19Sfv9edEPv=HibOJhB14TJv_RCQg@mail.gmail.com>
 <b2374b9461c349b17b50c77f3a70e1a1@xs4all.nl>
 <CAGMVOdsopg5ZAuJh9+ceq-s0pU2qqZsZn-jbL4A1c=h1iWaZVA@mail.gmail.com>
 <CAGMVOdvrZCwJg0BngM9GMO+LK=j20vhSfdk1gTTBNNH311Vgsw@mail.gmail.com>
 <bc8a1d397fa81d6ae33c297d0224184b@xs4all.nl>
 <2c084c1c2bf599ceb2afa87874de8bc1@xs4all.nl>
 <CAGMVOdvETRCKpeT06Uoq5RsNUOdH7d1iYy7C1Pze=L5=gBzs-Q@mail.gmail.com>
 <cdbe62a6eaddae00fa770c09ed346fc0@xs4all.nl>
 <CAGMVOdttRSUU3=tjgdhr902HmUVL1GLQWG0ZTVXi7OY59Xo0Xw@mail.gmail.com>
 <13279.1515535159@sss.pgh.pa.us>
 <CAGMVOdtYSNJusckQ_9vBkMsZBQzan=Uw-Z-uYG2fD9rZ2TP7pA@mail.gmail.com>
 <4139.1515567978@sss.pgh.pa.us>
 <CAGMVOdvJMfOzm0v4psCoAaVGfCbdFwNKzaSEhfTj8rE9UCz-eQ@mail.gmail.com>
 <c30e6457a6de09d26425af997f71996e@xs4all.nl>
 <CAGMVOdtGe=30u6D2WqwvWchE+-yNb9jKnEeG8BHyw7jHiGPR7w@mail.gmail.com>
Comments: In-reply-to Oliver Ford <ojford@gmail.com>
	message dated "Sat, 27 Jan 2018 10:49:46 +0000"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <27787.1517069693.1@sss.pgh.pa.us>
Date: Sat, 27 Jan 2018 11:14:53 -0500
Message-ID: <27788.1517069693@sss.pgh.pa.us>
Precedence: bulk

Oliver Ford <ojford@gmail.com> writes:
> On Sat, Jan 27, 2018 at 7:40 AM, Erik Rijkers <er@xs4all.nl> wrote:
>> Regression tests only succeed for assert-disabled compiles; they fail when
>> assert-enabled:

> Problem seems to be with an existing Assert in catcache.c:1545:
> Assert(nkeys > 0 && nkeys < cache->cc_nkeys);
> The "<" needs to be "<=" (and is changed in the attached patch).

No, that Assert is correct, because it's in SearchCatCacheList.
It doesn't make any sense to use SearchCatCacheList for a lookup
that specifies all of the key columns, because then you necessarily
have at most one match; you might as well use regular SearchCatCache,
which is significantly more efficient.

> AFAICT this was never a problem before purely because no code before
> this patch called SearchSysCacheList4, so they always called with
> fewer keys than the number available.

Hm, probably we shouldn't even have that macro.

			regards, tom lane