Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220629235326.480858-1-justinstitt@google.com> <20220718172934.4155189-1-justinstitt@google.com>
In-Reply-To: <20220718172934.4155189-1-justinstitt@google.com>
From:   Nick Desaulniers <ndesaulniers@google.com>
Date:   Mon, 18 Jul 2022 15:00:24 -0700
Message-ID: <CAKwvOdnN67xFL22oWfqrKjvES4vjSB3KNhm401LtoC7igO+4Vg@mail.gmail.com>
Subject: Re: [PATCH v3] lib/test_printf.c: fix clang -Wformat warnings
To:     Justin Stitt <justinstitt@google.com>
Cc:     andriy.shevchenko@linux.intel.com, linux-kernel@vger.kernel.org,
        llvm@lists.linux.dev, nathan@kernel.org, pmladek@suse.com,
        rostedt@goodmis.org, senozhatsky@chromium.org, trix@redhat.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

/On Mon, Jul 18, 2022 at 10:29 AM Justin Stitt <justinstitt@google.com> wro=
te:
>
> see warnings:
> | lib/test_printf.c:157:52: error: format specifies type 'unsigned char'
> | but the argument has type 'int' [-Werror,-Wformat]
> test("0|1|1|128|255",
> | "%hhu|%hhu|%hhu|%hhu|%hhu", 0, 1, 257, 128, -1);
> -
> | lib/test_printf.c:158:55: error: format specifies type 'char' but the
> | argument has type 'int' [-Werror,-Wformat] test("0|1|1|-128|-1",
> | "%hhd|%hhd|%hhd|%hhd|%hhd", 0, 1, 257, 128, -1);
> -
> | lib/test_printf.c:159:41: error: format specifies type 'unsigned
> short'
> | but the argument has type 'int' [-Werror,-Wformat]
> | test("2015122420151225", "%ho%ho%#ho", 1037, 5282, -11627);
>
> There's an ongoing movement to eventually enable the -Wformat flag for
> clang. Previous patches have targeted incorrect usage of
> format specifiers. In this case, however, the "incorrect" format
> specifiers are intrinsically part of the test cases. Hence, fixing them
> would be misaligned with their intended purpose. My proposed fix is to
> simply disable the warnings so that one day a clean build of the kernel
> with clang (and -Wformat enabled) would be possible. It would also keep
> us in the green for alot of the CI bots.
>
> Link: https://github.com/ClangBuiltLinux/linux/issues/378
> Suggested-by: Nathan Chancellor <nathan@kernel.org>
> Suggested-by: Nick Desaulniers <ndesaulniers@google.com>
> Signed-off-by: Justin Stitt <justinstitt@google.com>
> ---
> changes from v1 -> v2:
> * moved NOWARN macro definition to a more appropriate location
> * using __diag_ignore_all (thanks Nathan)
> * using local scoping for code blocks instead of __VA_ARGS__ (thanks Nick=
)
> * indented affected test cases (thanks Andy)
>
> changes from v2 -> v3:
> * reinserted commit message
> * remove Andy's Suggested-by tag
> * add issue tracker link
>
>  lib/test_printf.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/lib/test_printf.c b/lib/test_printf.c
> index 07309c45f327..1b1755ce9fa7 100644
> --- a/lib/test_printf.c
> +++ b/lib/test_printf.c
> @@ -30,6 +30,9 @@
>  #define PAD_SIZE 16
>  #define FILL_CHAR '$'
>
> +#define NOWARN(option, comment, block) \
> +               __diag_push() __diag_ignore_all(#option, comment) block _=
_diag_pop()

Do you mind putting these 4 on distinct lines with \ at the end of the
lines, and terminate the statements with an additional ; if that
doesn't produce new warnings about empty statements? Maybe something
like:

+#define NOWARN(option, comment, block) \
+  __diag_push(); \
+  __diag_ignore_all(#option, comment); \
+ block \
+ __diag_pop();

(sorry for not formatting that myself, remember to use tabs vs
spaces).  To me, it seems more readable from a quick glance to have
these be distinct lines.  You'll find open coded uses of __diag_push
and friends add the semicolon, even if not strictly necessary.

> +
>  KSTM_MODULE_GLOBALS();
>
>  static char *test_buffer __initdata;
> @@ -154,9 +157,11 @@ test_number(void)
>         test("0x1234abcd  ", "%#-12x", 0x1234abcd);
>         test("  0x1234abcd", "%#12x", 0x1234abcd);
>         test("0|001| 12|+123| 1234|-123|-1234", "%d|%03d|%3d|%+d|% d|%+d|=
% d", 0, 1, 12, 123, 1234, -123, -1234);
> -       test("0|1|1|128|255", "%hhu|%hhu|%hhu|%hhu|%hhu", 0, 1, 257, 128,=
 -1);
> -       test("0|1|1|-128|-1", "%hhd|%hhd|%hhd|%hhd|%hhd", 0, 1, 257, 128,=
 -1);
> -       test("2015122420151225", "%ho%ho%#ho", 1037, 5282, -11627);
> +       NOWARN(-Wformat, "Disables clang -Wformat warning", {

Technically, this is disabling -Wformat for all compilers.  How about
a comment string like "Intentionally test narrowing conversion
specifiers"?

> +               test("0|1|1|128|255", "%hhu|%hhu|%hhu|%hhu|%hhu", 0, 1, 2=
57, 128, -1);
> +               test("0|1|1|-128|-1", "%hhd|%hhd|%hhd|%hhd|%hhd", 0, 1, 2=
57, 128, -1);
> +               test("2015122420151225", "%ho%ho%#ho", 1037, 5282, -11627=
);
> +       })
>         /*
>          * POSIX/C99: =C2=BBThe result of converting zero with an explici=
t
>          * precision of zero shall be no characters.=C2=AB Hence the outp=
ut
> --
> 2.37.0.rc0.161.g10f37bed90-goog
>


--=20
Thanks,
~Nick Desaulniers