intl: On native Windows, actually define the *wprintf functions.
In dcaf8c4d7 (Cygwin portability., 2003-09-17), the *wprintf() family of
functions in `gettext-runtime/intl/printf.c` were no longer guarded by
the constant `HAVE_WPRINTF`, but instead by `HAVE_FWPRINTF`.
This apparently worked even if the corresponding part in
`gettext-runtime/intl/libgnuintl.h.in` uses `HAVE_WPRINTF` to guard the
declarations of those functions.
However, in d84f20745 (Make sure that libintl.h declares the *wprintf
overrides on Windows., 2019-04-08), gettext introduced a change where it
would look for `wprintf()` instead of `fwprintf()`. As a consequence it
would no longer define the `HAVE_FWPRINTF` constant at all.
GCC apparently interprets `#if HAVE_FWPRINTF` as `#if 0` if the constant
has not even been defined.
This leads to the funny situation that previously, the *wprintf()
functions would be defined, but not be declared. Whereas now the
functions are not defined, but declared.
Also funny: Cygwin did not even export the `fwprintf()` function until
2009, as per Cygwin's 45e20e47ba (cygwin.din: Export wprintf,
fwprintf, swprintf, vwprintf, vfwprintf, vswprintf. [...] , 2009-03-06),
while the `wprintf()` function was exported already in 2003. So the
Cygwin portability patch from 2003 seems to have turned off all of the
*wprintf() functions in gettext.
Let's revert that "portability" patch, as it now only does harm and has
no benefit anymore.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
[Bruno's summary]
Thank you for the analysis and explanation, Johannes!
The commits dcaf8c4d7 and f5ca7696d (2003-09-17) introduced a test for
fwprintf rather than wprintf. This was designed as a workaround to a
Cygwin problem that existed from 2003 to 2009.
The commit d84f20745 (2019-04-08) switched back from testing fwprintf
to testing wprintf, but did so incompletely: I forgot to modify printf.c.
So, since then, the *wprintf functions are declared but not defined.