Compiling getdate on mingw32:
getdate.y: In function 'get_date':
getdate.y:1325: warning: 'tzname' redeclared without dllimport attribute:
previous dllimport ignored
Linking fails:
Info: resolving _tzname by linking to __imp__tzname (auto-import)
Creating library file: .libs/libshishi.dll.a
fu000001.o:(.idata$2+0xc): undefined reference to `_libmoldname_a_iname'
fu000002.o:(.idata$2+0xc): undefined reference to `_libmoldname_a_iname'
nmth000000.o:(.idata$4+0x0): undefined reference to `__nm__tzname'
collect2: ld returned 1 exit status
make[3]: *** [libshishi.la] Error 1
getdate.y:1325 reads:
# ifndef tzname
extern char *tzname[];
# endif
tzname is declared in in time.h, although exactly which prototype is
used I'm not sure, but this could be it:
/* CRTDLL is royally messed up when it comes to these macros.
TODO: import and alias these via oldnames import library instead
of macros. */
#define daylight _daylight
/* NOTE: timezone not defined as macro because it would conflict with
struct timezone in sys/time.h.
Also, tzname used to a be macro, but now it's in moldname. */
__MINGW_IMPORT char *tzname[2];
Anyway, as far as I can tell, nothing says that tzname has to be a
#define. OpenGroup spec says it should be 'extern char *tzname[2];':
http://www.opengroup.org/onlinepubs/009695399/functions/tzname.html
Is there any platform where tzname isn't defined in time.h? If not, I
would suggest this patch:
diff --git a/gl/getdate.y b/gl/getdate.y
index 1ed914f..8de85d0 100644
--- a/gl/getdate.y
+++ b/gl/getdate.y
@@ -1321,9 +1321,6 @@ get_date (struct timespec *result, char const *p, struct
timespec const *now)
#else
#if HAVE_TZNAME
{
-# ifndef tzname
- extern char *tzname[];
-# endif
int i;
for (i = 0; i < 2; i++)
{
If there are platforms that doesn't declare tzname properly, I suppose
we should detect that using AC_CHECK_DECLS and work around it in the
time-h module.
Thoughts?
Maybe we could apply the patch, and wait for bug reports. Then we will
find out which platform lacks a proper tztime prototype.
/Simon
