KO Myung-Hun wrote:
> --- a/lib/freopen.c
> +++ b/lib/freopen.c
> @@ -26,10 +26,26 @@
> #include <stdio.h>
> #undef __need_FILE
>
> +#include <errno.h>
> +
> static FILE *
> orig_freopen (const char *filename, const char *mode, FILE *stream)
> {
> - return freopen (filename, mode, stream);
> + FILE *result;
> +
> + /* Clear errno to check the success of freopen() with it */
> + errno = 0;
> +
> + result = freopen (filename, mode, stream);
> +
> +#ifdef __KLIBC__
> + /* On OS/2 kLIBC, freopen() returns NULL even if it is successful
> + if filename is NULL. */
> + if (!filename && !result && !errno)
> + result = stream;
> +#endif
> +
> + return result;
> }
>
> /* Specification. */
Just a question of coding style: We have a chain of invocations
rpl_freopen -> orig_freopen -> freopen
While, in theory, you can put workarounds in either rpl_freopen or orig_freopen,
the function name 'orig_freopen' and the comment
/* Get the original definition of freopen.
indicate that 'orig_freopen' should remain a simple invocation of freopen,
and the workaround should go into rpl_freopen. This is the canonical way we've
been using in gnulib (see fopen.c, fstat.c, lstat.c, open.c, openat.c etc.).
I see no reason to abandon this convention.
Bruno