[PATCH v21 18/27] test: remove use of VLAs for Windows built code in bitset tests

[David Marchand]

Hello André,

On Tue, Feb 4, 2025 at 9:57 PM Andre Muezerie
<andremue at linux.microsoft.com> wrote:
> @@ -168,10 +169,20 @@ test_flip_size(test_fun test_fun, assign_fun assign_fun, flip_fun flip_fun, size
>         rand_bitset(bitset, size);
>
>         for (i = 0; i < size; i++) {
> -               RTE_BITSET_DECLARE(reference, size);
> +               RTE_BITSET_DECLARE(reference, RAND_SET_MAX_SIZE);
> +
> +#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION >= 110000)
> +#pragma GCC diagnostic push
> +#pragma GCC diagnostic ignored "-Warray-bounds"
> +#endif
>
> +               /* gcc is giving false positives here when code is optimized */

Why not simply alloca(te the right size)?

I tested with my gcc 14 (for which I could reproduce the array bound warning).
By replacing with uint64_t *reference = alloca(RTE_BITSET_SIZE(size)),
gcc seems to be less smart and won't inspect 'reference' and 'bitset'
arrays boundaries.


>                 rte_bitset_copy(reference, bitset, size);
>
> +#if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION >= 110000)
> +#pragma GCC diagnostic pop
> +#endif
> +
>                 bool value = test_fun(bitset, i);
>
>                 flip_fun(bitset, i);

The rest of the series lgtm and the plan is to merge it for rc1.

Just beware that, if you send a new revision, new drivers (net/xsc and
net/zxdh) landed in main.
Both use VLA, so both require cflags += no_wvla_cflag in their meson.build.


-- 
David Marchand