assert.throws can succeed if actual value is not a function and no matchers are used

[smcclure15]

I noticed that passing something like "1" into the throws assertion actually passes:

assert.throws(1);
// or
assert.throws(1, 'succeeds');

What's actually going on is shown if you use a matcher:

assert.throws(1, /somematch/);

  message: fails
  severity: failed
  actual  : TypeError: block.call is not a function
  expected: "/somematch/"

Which is what is used to invoke the actual function inside a try/catch block. So this is true for really anything that isn't a function:

assert.throws(undefined, 'succeeds');
assert.throws(undefined, /somematch/, 'fails');

assert.throws('notafunction', 'succeeds');
assert.throws('notafunction', /somematch/, 'fails');

assert.throws([], 'succeeds');
assert.throws([], /somematch/, 'fails');

assert.throws({}, 'succeeds');
assert.throws({}, /somematch/, 'fails1');

On one hand, sure, when the actual value is invoked, an exception is thrown. But I think this has the potential of false-positives; passing tests but it's really hiding something there. Even something like the following now has me second-guessing and desiring more validation (maybe assertingtypeof ahead of the throws check?):

let myfcn = utils.myFunctionGenerator();
assert.throws(myfcn)

let myfcn2 = utils.myFunctionGenerator(config);
assert.throws(myfcn2)

I'm proposing something more explicit for this case...

As a summary spec, throws shall:

• pass if the actual value is a function, which throws, and the expected value is undefined
• pass if the actual value is a function, which throws, and the expected value "matches"
• fail if the actual value is a function and does not throw
• (NEW) fail if the actual value is not a function
• error if the expected value is invalid (Assert: Improve rejects/throws validation handling #1635)

The new item would more align with the rejects validation that the actual value must be a promise (or thenable).

[Krinkle]

Agreed fully. I propose we deprecate this along with the invalid expected values, possibly in the same patch if you like.