Re: [PATCH] regexec: fix unintentinal fallthrough warning with GCC7

[Eric Blake]

On 03/23/2017 11:16 AM, Andrei Borzenkov wrote:
> [  104s] In file included from ../../grub-core/gnulib/regex.c:72:0:
> [  104s] ../../grub-core/gnulib/regexec.c: In function 'check_node_accept':
> [  104s] ../../grub-core/gnulib/regexec.c:4100:10: error: this statement may 
> fall through [-Werror=implicit-fallthrough=]
> [  104s]        if (ch >= ASCII_CHARS)
> [  104s]           ^
> [  104s] ../../grub-core/gnulib/regexec.c:4104:5: note: here
> [  104s]      case OP_PERIOD:
> [  104s]      ^~~~
> 
> The code in question does have FALLTHROUGH annotation; unfortunately GCC7
> ignores it, because it is separated from case statement by #endif.
> 
> The patch fixes it by using new attribute that is honored even inside 
> #ifdef'ed
> code.
> 
> ---
>  lib/regexec.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/lib/regexec.c b/lib/regexec.c
> index ef52b24..4588e13 100644
> --- a/lib/regexec.c
> +++ b/lib/regexec.c
> @@ -4078,6 +4078,9 @@ check_node_accept (const re_match_context_t *mctx, 
> const re_token_t *node,
>      case OP_UTF8_PERIOD:
>        if (ch >= ASCII_CHARS)
>          return false;
> +#if defined __GNUC__ && __GNUC__ >= 7
> +      __attribute__ ((fallthrough));
> +#endif
>        /* FALLTHROUGH */
>  #endif
>      case OP_PERIOD:


Would it also work if you did:

    case OP_UTF8_PERIOD:
      if (ch >= ASCII_CHARS)
        return false;
#endif
      /* FALLTHROUGH */
    case OP_PERIOD:

(that is, swap the #endif and FALLTHROUGH lines)?

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature