Re: [PATCH] kunit: Fix a NULL vs IS_ERR() bug

[Dan Carpenter]

On Fri, Jan 12, 2024 at 07:39:14AM +0800, David Gow wrote:
> On Thu, 11 Jan 2024 at 02:55, Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> >
> > The kunit_device_register() function doesn't return NULL, it returns
> > error pointers.  Change the KUNIT_ASSERT_NOT_NULL() to check for
> > ERR_OR_NULL().
> >
> > Fixes: d03c720e03bd ("kunit: Add APIs for managing devices")
> > Signed-off-by: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> > ---
> 
> Nice catch, thanks!
> 
> Reviewed-by: David Gow <davidgow@xxxxxxxxxx>
> 
> > It's a pity that there isn't a KUNIT_ASSERT_NOT_ERR_PTR() macro...
> 
> I think we'll add one, but I'm not yet totally convinced that it would
> be better than using ASSERT_NOT_ERR_OR_NULL() in cases like this,
> where we're:
> 1. In a test; and,
> 2. using the pointer afterwards, expecting it to be valid
> (dereferencing it and/or passing it to functions which will)
> 
> This is largely because it'd be nicer, if the pointer is NULL (due to
> a bug), to get a more explicit assertion failure, rather than a crash.
> It does make the test code less indicative of how the APIs are meant
> to be used elsewhere, though, and annoys the static analysis, though.
> 
> Thoughts?

It doesn't annoy any static checkers because nothing looks for it.

Expecting that this test code might be buggier than normal code probably
isn't unreasonable so I guess that makes sense.

regards,
dan carpenter