Re: Compilation errors using Gnulib's time.h on MinGW

[Bruno Haible]

Hi Eli,

> The problem described below was found when building the latest pretest
> of wget2 with mingw.org's MinGW system headers and libraries.

Oh, there are still people who use the (old) mingw.org system?
I thought everyone by now had switched to mingw-w64. This (newer) mingw
does not exhibit the behaviour you describe:
<time.h>        
https://github.com/mirror/mingw-w64/blob/master/mingw-w64-headers/crt/time.h
<sys/types.h>   
https://github.com/mirror/mingw-w64/blob/master/mingw-w64-headers/crt/sys/types.h

Hmm. mingw.org points me to this source:
<time.h>   
https://sourceforge.net/p/mingw/mingw-org-wsl/ci/21762bb4a1bd0c88c38eead03f59e8d994349e83/tree/include/time.h
which also does not fit your description: it doesn't include <sys/types.h>.

> This assumes that the system header sys/types.h can be included just
> once, anywhere in the program, and that is sufficient.  (MinGW's
> time.h includes sys/types.h, which causes it to include Gnulib's
> sys/types.h.)  Is that a portable assumption?

In general, it is a portable assumption. There are very few cases where
such an assumption does not hold, and it requires special handling with
special #ifs in gnulib.

> What happens with MinGW headers is that when MinGW's sys/types.h is
> included from time.h, only a small portion of sys/types.h is
> processed, the one that is relevant to time.h types, and that portion
> doesn't include the types mentioned in the above error messages.  The
> next time sys/types.h is included by the program, the Gnulib header
> decides it doesn't need to delegate to the system header, because
> address@hidden@_SYS_TYPES_H is already defined.  So the relevant
> portions of MinGW's sys/types.h are skipped, and we get the error
> messages.

Please can you attach
  - a copy of the mingw <time.h>,
  - a copy of the mingw <sys/types.h>,
  - the result of preprocessing utimens.c with "-E"?

Thanks.

          Bruno