Re: Current testsuite failure analysis on i386-unknown-freebsd4.7

[Martin Sebor <sebor at roguewave dot com>]

Loren James Rittle wrote:
> Every current (newly recent) failure on mainline appears related to
> locale support in some way or another.

Let me offer a suggestion to test suite writers for an approach
that has worked well for us on all platforms we support:

Tests such as num_get/put, and money_get/put should create and
install their own punct and ctype facets and exercise the get/put
facets in isolation from any implementation issues they would
otherwise be dependent on. That way the expected results of
the tests can be hardcoded and must match regardless of the
locales installed on a system. This will allow you to exercise
wacky cases such as groupings like "\017\xff" or decimal_point
of '\0' that you won't run into in any standard locale but that
users can create themselves.

For, codecvt, collate, ctype, punct and time facets that are
inherently tied to the underlying implementation, one option is
to run "locale -a" on each system and exercise only (and/or all)
those locales that are installed on the system. Instead of
hardcoding the expected results, the tests can compare them to
the results of the corresponding POSIX and libc functions such
as wctomb(), wcscoll(), isalpha(), localeconv(), wcsftime(),
and strptime().

Regards
Martin