On 15/8/23 10:16, David Gow wrote:
On Mon, 14 Aug 2023 at 21:23, Richard Fitzgerald <rf@xxxxxxxxxxxxxxxxxxxxx> wrote:Re-work string_stream so that it is not tied to a struct kunit. This is to allow using it for the log of struct kunit_suite. Instead of resource-managing individual allocations the whole string_stream object can be resource-managed as a single object: alloc_string_stream() API is unchanged and takes a pointer to a struct kunit but it now registers the returned string_stream object to be resource-managed. raw_alloc_string_stream() is a new function that allocates a bare string_stream without any association to a struct kunit. free_string_stream() is a new function that frees a resource-managed string_stream allocated by alloc_string_stream(). raw_free_string_stream() is a new function that frees a non-managed string_stream allocated by raw_alloc_string_stream(). The confusing function string_stream_destroy() has been removed. This only called string_stream_clear() but didn't free the struct string_stream. Instead string_stream_clear() has been exported, and the new functions use the more conventional naming of "free" as the opposite of "alloc". Signed-off-by: Richard Fitzgerald <rf@xxxxxxxxxxxxxxxxxxxxx> ---I'm in favour of this. Should we go further and get rid of the struct kunit member from string_stream totally?
I can do that. I was worried about some hairy-looking code in assert.c that used stream->test. But I've just looked at it again and it's really quite simple, and doesn't even need ->test. is_literal() allocates a temporary managed buffer, but it frees it before returning so it doesn't need to be managed.
Also, note that the kunit_action_t casting is causing warnings on some clang configs (and technically isn't valid C). Personally, I still like it, but expect more emails from the kernel test robot and others.
I can send a new version to fix that.
