On Wed, Feb 27, 2019 at 15:05:18 +0100, Erik Skultety wrote:
On Tue, Feb 26, 2019 at 04:48:26PM +0100, Peter Krempa wrote:
> We'd free only the first element of the vector leaking the rest.
>
> Signed-off-by: Peter Krempa <pkrempa(a)redhat.com>
> ---
> src/util/viralloc.h | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/src/util/viralloc.h b/src/util/viralloc.h
> index 15451d4673..572b7d1c1c 100644
> --- a/src/util/viralloc.h
> +++ b/src/util/viralloc.h
> @@ -650,6 +650,9 @@ void virAllocTestHook(void (*func)(int, void*), void *data);
> * the variable declared with it by calling the function
> * defined by VIR_DEFINE_AUTOPTR_FUNC when the variable
> * goes out of scope.
> + *
> + * Note that this macro must NOT be used with vectors! The cleaning function
> + * will not free any elements beyond the first.
s/cleaning/freeing/
I understand, but if you have happen to have a dedicated list type, then you'd
have a dedicated destructor, so both of these would be okay with vectors. On
Note that the function registered via __attribute(cleanup ... gets only
the pointer to the stack'd variable as an argument. This means that you
can do only 'value-terminated' (NULL, -1, ... ) lists.
Anything requiring count of elements will need to be encapsulated in a
struct which makes it a container. Thus the comment does not apply.
the other hand I'm not sure whether we have such a thing at the
moment, so I
guess it's meaningful to document in the meantime.
I don't think we'll get much value terminated lists because the usage is
quite cumbersome and error-prone (e.g. if you forget your terminator).
In such case we can always do a rather simple macro for them.