BUG #15285: Query used index over field with ICU collation in some cases wrongly return 0 rows

Hi,

I'm bumping this thread on pgsql-hacker, hopefully it will drag some more
opinions/discussions.

Should we try to fix this issue or not? This is clearly an upstream bug. It has
been reported, including regression tests, but this doesn't move since 2 years
now.

If we choose not to fix it on our side using eg a workaround (see patch), I
suppose this small bug should be documented somewhere so people are not lost
alone in the wild.

Opinions?

Regards,

Begin forwarded message:

Date: Sat, 13 Jun 2020 00:43:22 +0200
From: Jehan-Guillaume de Rorthais <jgdr(at)dalibo(dot)com>
To: Thomas Munro <thomas(dot)munro(at)gmail(dot)com>, Peter Geoghegan <pg(at)bowt(dot)ie>
Cc: Роман Литовченко <roman(dot)lytovchenko(at)gmail(dot)com>, PostgreSQL mailing lists
<pgsql-bugs(at)lists(dot)postgresql(dot)org> Subject: Re: BUG #15285: Query used index
over field with ICU collation in some cases wrongly return 0 rows

On Fri, 12 Jun 2020 18:40:55 +0200
Jehan-Guillaume de Rorthais <jgdr(at)dalibo(dot)com> wrote:

> On Wed, 10 Jun 2020 00:29:33 +0200
> Jehan-Guillaume de Rorthais <jgdr(at)dalibo(dot)com> wrote:
> [...]
> > After playing with ICU regression tests, I found functions ucol_strcollIter
> > and ucol_nextSortKeyPart are safe. I'll do some performance tests and report
> > here.
>
> I did some benchmarks. See attachment for the script and its header to
> reproduce.
>
> It sorts 935895 french phrases from 0 to 122 chars with an average of 49.
> Performance tests were done on current master HEAD (buggy) and using the patch
> in attachment, relying on ucol_strcollIter.
>
> My preliminary test with ucol_getSortKey was catastrophic, as we might
> expect. 15-17x slower than the current HEAD. So I removed it from actual
> tests. I didn't try with ucol_nextSortKeyPart though.
>
> Using ucol_strcollIter performs ~20% slower than HEAD on UTF8 databases, but
> this might be acceptable. Here are the numbers:
>
> DB Encoding HEAD strcollIter ratio
> UTF8 2.74 3.27 1.19x
> LATIN1 5.34 5.40 1.01x
>
> I plan to add a regression test soon.

Please, find in attachment the second version of the patch, with a
regression test.

Regards,

--
Jehan-Guillaume de Rorthais
Dalibo