On 10/11/2011 12:54 PM, Charles Wilson wrote:
> Consensus does appear to be unanimous on what to do; I just need to
> review all the postings and figure out exactly /how/ to do it.

I have uploaded the new packages.  There are three new patches:

1) modified localename.c significantly.  No longer "ignores" 
LANG=C.UTF-8; also does not try to do any parsing of the Windows I18N 
settings.  Basically, acts like linux -- if the value of the locale 
string isn't supported by the underlying setlocale() implementation, 
then it is ignored (e.g. default back to "C.UTF-8" or "C") -- libintl 
doesn't try to 'be smart' -- or to second-guess.  Also, relies on 
cygwin's glibc-like setlocale(LC_*, NULL) behavior -- which has been 
supported by newlib since the cygwin 1.5 days (even if it always 
returned "C" back then).

2) Fixes to the test suite related to the above changes.  From the 
CHECK_NOTES file:

NOTE: there were four other failures that arose because of the
       changes to localename.  These were:
          format-c-3
          format-c-4
          plural-1
          plural-2
       but the test suite was patched to avoid these failures.
       The problem is rather complicated; gettext-0.18.1.1 now has a
       libintl_setlocale function that is used by libintl clients,
       instead of the system setlocale.  The localename patch left
       that in place, but ensured that it did far less work -- was less
       obtrusive -- in its interposition between clients and the system
       setlocale.  However, some of that eliminated 'work' also allowed
       the gettext test suite to 'cheat' a bit -- it used non-existent
       locales (like 'fr' rather than 'fr_FR'), and *re-hijacked* the
       setlocale function itself.  None of that works anymore -- and
       *actual* clients are unlikely to need/want to do it -- so the
       test suite was patched to use real locales in all cases, and to
       not re-hijack.

3) Adopted Bruno's upstream changes to relocatable.c, turning off 
"expensive" relocation support in libintl.

Odds of #1 and #2 being adopted upstream are effectively nil, so...

--
Chuck