assert: handle NaN properly in deep assertions

[thefourtheye]

As it is, when two container objects which have NaNs are asserted
to be equal, it will fail.

> process.version
'v3.0.0'
> assert.deepEqual([NaN], [NaN])
...
AssertionError: [ NaN ] deepEqual [ NaN ]
> assert.deepEqual({key: NaN}, {key: NaN})
...

This patch explicitly compares the NaN values with Number.isNaN.

PR-URL: #2440

[targos]

I'm not sure about this... NaN usually appears because of a programmer mistake. Do you have a use case where it should really be expected in a result ?

[vkurchatkin]

NaN is really not equal to NaN any language, it is a part of ieee 754 standard. Changing that would be confusing.

[thefourtheye]

@targos I don't have a specific use case. But it is one of the special cases in the language and as it is we don't handle it well, IMHO.

@vkurchatkin Agreed, but I believe assertion module is not necessarily just a comparison module. When I am writing a test, lets say, I am expecting a function to return an array of NaNs. And when I use assert.deepEqual family of functions to check if my expectation is correct, it would say "NO" as it is.

[brendanashworth]

I'm -1 on this, if NaN == NaN is false, then deepEqual([NaN], [NaN]) should also be false.

[evanlucas]

I agree with @brendanashworth . If you need to verify an array, then you could loop and assert(isNaN(value)) for each item

[thefourtheye]

@evanlucas What if only few of the items are NaN?

[thefourtheye]

I am still not convinced, but closing as this is -1ed.

[cjihrig]

@thefourtheye I'd encourage you to release this on npm.

[thefourtheye]

@cjihrig Even chai had the same problem. I am trying to fix it. chaijs/chai#508

[evanlucas]

@thefourtheye then you end up doing something like iterating over all items and checking isNaN on it