Paolo Bonzini <pbonz...@redhat.com> writes:
> If expected_args is 0, qtest frees the argument vector and then returns it
> nevertheless. Coverity complains; in practice this is not an issue because
> expected_args == 0 means that the caller is not interested in the argument
> vector, but it would be a potential problem if somebody wanted to add
> commands with optional arguments to qtest.
>
> Suggested-by: Kamil Dudka <kdu...@redhat.com>
> Signed-off-by: Paolo Bonzini <pbonz...@redhat.com>
> ---
> tests/qtest/libqtest.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/tests/qtest/libqtest.c b/tests/qtest/libqtest.c
> index be0fb430dd..e49f3a1e45 100644
> --- a/tests/qtest/libqtest.c
> +++ b/tests/qtest/libqtest.c
> @@ -545,6 +545,7 @@ redo:
if (expected_args) {
for (i = 0; i < expected_args; i++) {
g_assert(words[i] != NULL);
> }
> } else {
> g_strfreev(words);
> + words = NULL;
> }
>
> return words;
The patch is okay, so
Reviewed-by: Markus Armbruster <arm...@redhat.com>
Odd: we require at least @expected_args, but more is fine. Not what the
name @expected_args made me expect.
We treat expected_args == 0 as special case to save callers the trouble
to free @words. Okay, but I'd try a more regular interface:
* one function that takes @expected_args and returns @words, always.
* a simple wrapper around it that passes zero @expected_args and frees
@words.
