Re: [PATCH 0/2] nolibc: Add fread() and fseek()

[Willy Tarreau <w@1wt.eu>]

Hi Daniel,

On Sun, Jan 04, 2026 at 08:12:11PM +0900, Daniel Palmer wrote:
> Hi Thomas,
> 
> On Sun, 4 Jan 2026 at 18:11, Thomas Weißschuh <linux@weissschuh.net> wrote:
> > Please also do add some tests.
> 
> Would a single function test that exercises the new functions be
> enough? Something like this:
> 
> --- a/tools/testing/selftests/nolibc/nolibc-test.c
> +++ b/tools/testing/selftests/nolibc/nolibc-test.c
> @@ -877,6 +877,45 @@ int test_file_stream(void)
>        return 0;
> }
> 
> +int test_file_stream_wsr(void)
> +{
> +       const char dataout[] = "foo";
> +       const size_t datasz = sizeof(dataout);
> +       char datain[datasz];
> +       FILE *f;
> +       int r;
> +
> +       f = fopen("/tmp/file_stream_test", "w+");
> +       if (!f)
> +               return -1;
> +
> +       r = fwrite(dataout, 1, datasz, f);
> +       if (r != datasz)
> +               goto fail;
> +
> +       r = fseek(f, 0, SEEK_SET);
> +       if (r)
> +               goto fail;
> +
> +       r = fread(datain, 1, datasz, f);
> +       if (r != datasz)
> +               goto fail;
> +
> +       if (memcmp(datain, dataout, datasz) != 0)
> +               goto fail;
> +
> +       r = fclose(f);
> +       if (r == EOF)
> +               return -1;
> +
> +       return 0;
> +
> +fail:
> +       fclose(f);
> +       return -1;
> +}
> +

In general it's preferable when tests focus on a specific syscall or
function, so that when a failure is reported, it's easier to figure
what we broke. But I agree it's not always easy, especially when
resources are allocated and released, like here. I can hardly propose
better. Please also keep in mind that sometimes we also want to validate
certain error cases (e.g. if opening a file in read-only mode, writing
to it should fail).

Hoping this helps,
Willy