Re: The question of LOCALE at the time of a regression test.

Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>   
>> Since we rely on the OS to supply locale settings, getting a reliable 
>> set of regression tests that depended on the locale would be close to 
>> impossible. We really have to run the regression tests under --no-locale.
>>     
>
> This is quite untrue; we have variant regression files that exist
> specifically to support running the tests under various popular locales.
> As Hiroshi-san points out, --no-locale isn't even the default in the
> Unix makefile.
>
> Perhaps the Windows locales are different enough that what we have
> doesn't cover them?
>
>   

I stand corrected.

I know I had enough trouble even before we started doing Windows builds 
that we had to use --no-locale in the buildfarm (or at least that was 
the solution I adopted).

For example, I know of cases where FBSD and Linux don't agree even on 
collation order for quite common locales.

But by all means let's see what happens on Windows when we take the flag 
out. Hiroshi-san, perhaps you would like to test it and see?

cheers

andrew