Fix broadcasting of messages with headers with NULL or all-whitespace values

[habbbe]

Fix parsing of headers with NULL or all-whitespace values when lifting

Add tests for NULL and all-whitespace header values

[kozlovic]

I was going to say that we should simply fix the fact that we allow NULL or empty values, but I checked the Go client, and those are not rejected. Even worse, on the Go client, there is no error returned when a key is not present, we just return "", so user will not really know if the key was present but empty or the value is empty.
So I am willing to accept those changes. I have however made some comments on the tests.

[kozlovic]

Let's add some test() so that when running the test in verbose, we see what we are testing and the PASSED result. The name of the keys are wrong. They need to be of the form This-Is-Correct-Name. The last part I suggest adding is to make sure that your changes still work when adding more than one of those "bad" values to the same key, and later retrieve them with Values() API:

    test("Add NULL header: ");
    IFOK(s, natsMsgHeader_Add(msg, "Null-Header", NULL))
    testCond(s == NATS_OK);

    test("Add whitespace header: ");
    IFOK(s, natsMsgHeader_Add(msg, "Whitespace-Header", "  "))
    testCond(s == NATS_OK);

    test("Add NULL and whitespace values under same header: ");
    IFOK(s, natsMsgHeader_Add(msg, "Special-Headers", NULL))
    IFOK(s, natsMsgHeader_Add(msg, "Special-Headers", "  "))
    testCond(s == NATS_OK);

[kozlovic]

I would change with this:

    test("Check null header: ");
    s = natsMsgHeader_Get(rmsg, "Null-Header", &val);
    testCond((s == NATS_OK)
                && (val != NULL) && *val == '\0');

    test("Check whitespace header: ");
    s = natsMsgHeader_Get(rmsg, "Whitespace-Header", &val);
    testCond((s == NATS_OK)
                && (val != NULL) && *val == '\0');

    test("Check getting values from special header: ");
    s = natsMsgHeader_Values(rmsg, "Special-Headers", &values, &count);
    testCond((s == NATS_OK) && (count == 2)
                && (values != NULL) && (*values[0] == '\0') && (*values[1] == '\0'));
    free(values);
    values = NULL;

[habbbe]

I was going to say that we should simply fix the fact that we allow NULL or empty values, but I checked the Go client, and those are not rejected. Even worse, on the Go client, there is no error returned when a key is not present, we just return "", so user will not really know if the key was present but empty or the value is empty. So I am willing to accept those changes. I have however made some comments on the tests.

Thank you for the comments. I will fix these ASAP.

The reason for making the PR was that all functions for adding headers do say that the value can be NULL or empty, and empty (or all-whitespace) headers could already be broadcast (but not retrieved, at least not with the C client).

The fact that there were tests explicitly testing for non-support for NULL/empty though made me slightly worried that the comments were a typo, but in the name of not breaking existing API, I opted for adding support rather than removing it.

[habbbe]

Added suggestions. I also added an additional test for completely empty value.

[kozlovic]

LGTM - Thank you for your contribution!