Re: [PATCH 2/3] sco-tester: Introduce adapter features

["Frédéric DALLEAU" <frederic.dalleau@linux.intel.com>]

Hi Johan,

> First of all, you should use "false" here instead of "0" (and same for
> true vs 1 later).
>
> Secondly, whenever you want to manipulate some parameter of a test case
> it should not be done in the main function like that but instead you
> should have it as part of the test case data (struct sco_client_data)
> and create a separate instance for each test case.
 >
> I do realize that the initial sco-tester didn't get this right (having
> disable_sco in struct test_data instead of struct sco_client_data), but
> it'd be good to get this fixed before adding any new test cases.

The problem is that the adapter feature has to be defined far before the 
test is started.

Another problem is that sco_client_data is test dependant, eg. each
test can implement sco_client_data differently. In this case, 
adapter_features could simply be undefined! This is already the case in 
the first tests which do not use struct sco_client_data at all 
(test_socket).

IMHO We do not want that. adapter_features must be defined explictly 
before each test. This is mandatory.

In future I'm also thinking to implement some adapter "behavior". For 
example : one test could be : try connect, emulator must accept, and 
expected result is success. then try connect, emulator must refuse, and 
result is expected to be connection refused.

My suggestion would be really different from yours :
I would write a big table with the elements of test_sco.

struct test {
	char *name;
	struct adapter_features *features;
	struct adapter_behavior *behavior; /* If needed */
	void (*test_cb)(struct test *test, void *data);
         void *user_data; /* sco_client_data*/
	int expected_result;
} tests [] = {...};

while (test[i].name) {
     test_sco(test[i].name, test[i].test_cb, ...);
     i++;
}

Let me know what you think.

Regards,
Frédéric