This warning is generated by https://git.gnome.org/browse/gnome-calculator/tree/search-provider/search-provider.vala#n43 with:

private static string[] get_result_identifier (string[] terms)
ensures (result.length == 0 || result.length == 1)

Yeah, the problem is, it makes absolutely no sense that we would have more than one search result (the answer), or fewer than zero, so some debugging is required.

I can clearly see the C code, that is being generated is wrong.
In both cases, the condition is always going to fail.

(Does that mean, there's a bug in Vala??)

e.g. When the result has only one element, the code is as follwoing,

                result = _tmp5_;
                _tmp6_ = result;
                _tmp6__length1 = -1;
                _tmp7_ = result;
                _tmp7__length1 = -1;
                g_warn_if_fail ((_tmp6__length1 == 0) || (_tmp7__length1 == 1));
                return result;

and, when result is zero element string, the code is as follwoing,

                gchar** _tmp8_ = NULL;
                gchar** _tmp9_ = NULL;
                gint _tmp9__length1 = 0;
                gchar** _tmp10_ = NULL;
                gint _tmp10__length1 = 0;
                gchar** _tmp11_ = NULL;
                gint _tmp11__length1 = 0;
                _tmp8_ = g_new0 (gchar*, 0 + 1);
                _tmp9_ = _tmp8_;
                _tmp9__length1 = 0;
                if (result_length1) {
                        *result_length1 = _tmp9__length1;
                }
                result = _tmp9_;
                _tmp10_ = result;
                _tmp10__length1 = -1;
                _tmp11_ = result;
                _tmp11__length1 = -1;
                g_warn_if_fail ((_tmp10__length1 == 0) || (_tmp11__length1 == 1));

                _tmp10__length1 = -1;
                ...
                _tmp11__length1 = -1;
                g_warn_if_fail ((_tmp10__length1 == 0) || (_tmp11__length1 == 1));

OK, good catch.  So this is indeed looks like a Vala bug. I guess it should handle the zero-length array properly, or else give an error. Let's see what the Vala developers have to say....

I guess "don't do that" is probably a valid response, though I'm not sure how else to indicate no results.

Created attachment 292126 [details] [review]
Set array lengths for result variable

Fixes bug 737222

Created attachment 292127 [details] [review]
Improve pre/postcondition messages

Additional patch to make pre/postconditions more readable, same as assert()

Pre and post conditions are very broken atm for && or || checks. Planning to fix them. So when doing complex checks, be aware to check your C code. Sorry for the inconvenience.

Will review the patches asap, thanks.

Review of attachment 292126 [details] [review]:

Thanks, however it breaks code that previously worked by assuming that returned arrays always have a length variable:

[CCode (array_length = false, array_null_terminated = true)]
private static string[] test () ensures (result[0] == "foo") {
	return new string[]{ "bar" };
}

void main () {
	test();
}

Review of attachment 292127 [details] [review]:

Good one thanks!

Created attachment 292203 [details] [review]
Added check if result has array length

Ok, this should work now. I added a test for this case as well.

commit f403b05730a012563c75b56d7475d1e8240a7656
Author: Simon Werbeck <simon.werbeck@gmail.com>
Date:   Tue Nov 25 02:23:17 2014 +0100

    Set array lengths for result variable
    
    These where missing when generating postconditions
    
    Fixes bug 737222

Can be improved but it's better than nothing for now, thanks a lot.

This problem has been fixed in the development version. The fix will be available in the next major software release. Thank you for your bug report.

*** Bug 698336 has been marked as a duplicate of this bug. ***