Re: PG-10 + ICU and abbreviated keys

On 11/13/2017 09:27 AM, Andreas Joseph Krogh wrote:
>  
> In PG-10, with ICU enabled, is abbreviated keys now enabled?
>  

Hello,

I think yes :

src/backend/utils/adt/varlena.c

1876 /*
1877 * Unfortunately, it seems that abbreviation for non-C collations is
1878 * broken on many common platforms; testing of multiple versions of glibc
1879 * reveals that, for many locales, strcoll() and strxfrm() do not return
1880 * consistent results, which is fatal to this optimization. While no
1881 * other libc other than Cygwin has so far been shown to have a problem,
1882 * we take the conservative course of action for right now and disable
1883 * this categorically. (Users who are certain this isn't a problem on
1884 * their system can define TRUST_STRXFRM.)
1885 *
1886 * Even apart from the risk of broken locales, it's possible that there
1887 * are platforms where the use of abbreviated keys should be disabled at
1888 * compile time. Having only 4 byte datums could make worst-case
1889 * performance drastically more likely, for example. Moreover, macOS's
1890 * strxfrm() implementation is known to not effectively concentrate a
1891 * significant amount of entropy from the original string in earlier
1892 * transformed blobs. It's possible that other supported platforms are
1893 * similarly encumbered. So, if we ever get past disabling this
1894 * categorically, we may still want or need to disable it for particular
1895 * platforms.
1896 */
1897 #ifndef TRUST_STRXFRM
1898 if (!collate_c && !(locale && locale->provider == COLLPROVIDER_ICU))
1899 abbreviate = false;
1900 #endif

But I did not do any test to compare performances.

Regards,

--
Adrien NAYRAT

http://dalibo.com - http://dalibo.org