On Fri, Dec 31, 2004 at 09:38:00PM -0500, [EMAIL PROTECTED] wrote:
> O O F O O F
> O O F O O F -Duse64bitint
> O O F O O F -Duselongdouble
> O O F O O F -Dusemorebits
> O O F O O F -Duseithreads
> O O F O O F -Duseithreads -Duse64bitint
> O O F O O F -Duseithreads -Duselongdouble
> O X F O O F -Duseithreads -Dusemorebits
> ../t/io/layers.t........................FAILED 1
The Fs were due to change 23714, which corrected the test to actually *test*
things :-)
It seems that t/io/layers.t was never aware that STDIN might have 2 layers
on it; stdio and utf8:
$ LC_ALL=en_GB.UTF-8 ./perl -C t/io/layers.t
1..36
# PERLIO = (undef)
# DOSISH = 0
# NONSTDIO = 0
# FASTSTDIO = 0
not ok 1 - STDIN - layers == 1
# Failed at t/io/layers.t line 96
# got '2'
# expected '1'
I've checked in change 23723 as a quick "fix" for the smokes, which I hope
will be sufficient to make the smokes pass. However, it's not really a
complete solution, as the presence/ab sense of the :utf8 layer for some
values of PERL_UNICODE/-C is determined by whether the locale is /UTF8/. So
I think that the regression tests need perl space access to PL_utf8locale
to be reliable. This would probably mean creating ${^UTF8LOCALE}. Is this
sane?
Nicholas Clark
