On Thu, 2023-11-02 at 09:04 +0100, Florian Westphal wrote:
> Thomas Haller <thaller@xxxxxxxxxx> wrote:
> > This message purely depends on the internal callbacks and at the
> > program
> > code. This is not useful. What is the user going to do with this
> > warning?
>
> Report a bug.
The way the message is worded ("warning:"), doesn't sound as if a user
should feel invited to report a bug.
> > Maybe there is a bug here, but then we shouldn't print a warning
> > but fix
> > the bug.
>
> How on earth do we do that, fix a bug before we know its there?
Writing tests that perform basic consistency checks of things that must
hold.
A unit test should check whether ops structs are consistent. This
patchset also enables a test that hits this problem (for "chain" ops).
If there is a bug, then an assertion (or just a plain crash) would be
preferable over printing a "warning" to stderr.
>
> > For example, calling `nft -j list ruleset` after test
> > "tests/shell/testcases/chains/0041chain_binding_0"
> > will trigger messages like:
> >
> > warning: stmt ops chain have no json callback
> > warning: stmt ops chain have no json callback
>
> That means that chain ops needs a bug fix as they do not support
> json output?
I don't know. The message might just trying to be helpful, and there is
no actual problem. The condition is attempted to be handled by the
following code.
I don't know which it is. Me sending the patch is basically asking that
question.
Thomas
