Re: default locale considered harmful? (was Re: [GENERAL]

Tom Lane wrote:
> Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> > So, my understanding is that you would create something such as:
> > CREATE INDEX iix ON tab (LIKE col)
> > and that does LIKE lookups and knows how to do col LIKE 'abc%', but it
> > can't be used for >= or ORDER BY, but it can be used for equality tests?
>
> Hm. Right at the moment, it wouldn't be used for equality tests unless
> you spelled equality as "a ~=~ b". I wonder whether that's necessary
> though; couldn't we dispense with that operator and use ordinary
> equality as the BTEqual member of these opclasses? Are there any
> locales that claim that not-physically-identical strings are equal?

Let me see if I understand.

Our default indexes will be able to do =, >, <, ORDER BY, and the
special index will be able to do LIKE, ORDER BY, and maybe equals. Do I
have that correct?

Looking at CVS, I see the warning about non-C locales has been removed.
Should we instead mention the new LIKE index method?

# (Be sure to maintain the correspondence with locale_is_like_safe() in selfuncs.c.)
if test x`pg_getlocale COLLATE` != xC && test x`pg_getlocale COLLATE` != xPOSIX; then
echo "This locale setting will prevent the use of indexes for pattern matching"
echo "operations. If that is a concern, rerun $CMDNAME with the collation order"
echo "set to \"C\". For more information see the Administrator's Guide."
fi

Doing LIKE with single-byte encodings would be easy because it would be
only 256 compares to find the min/max char values, but that doesn't work
with multi-byte encodings, right?

This LIKE/encoding problem is a tricky one because it gives poor
performance with little warning to users.

--
Bruce Momjian | http://candle.pha.pa.us
pgman(at)candle(dot)pha(dot)pa(dot)us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073