Super-fy the Super JSON format doc

[philrz]

In the same spirit as #5368.

[philrz]

Turns out case is important here.

$ super -version
Version: v1.18.0-69-g5ab9615c

$ echo '1 NaN 2' | super -z -c 'yield typeof(this)' -
<int64>
<float64>
<int64>

The following causes a syntax error, which makes sense to me:

$ echo '1 Nan 2' | super -z -c 'yield typeof(this)' -
stdio:stdin: Super JSON syntax error

Not sure how to explain this though. A bug?

$ echo 'Nan' | super -z -c 'yield typeof(this)' -
[no output]

Also related, it looks like Inf requires the explicit + or - at the moment.

$ echo '1 +Inf -Inf 2' | super -z -c 'yield typeof(this)' -
<int64>
<float64>
<float64>
<int64>

$ echo '1 Inf 2' | super -z -c 'yield typeof(this)' -
stdio:stdin: Super JSON syntax error

$ echo 'Inf' | super -z -c 'yield typeof(this)' -
[no output]

Whereas NaN seemed like an easy docs fix since that's the way it's shown in Excel/etc., it's not clear to me if we'd rather add support for plain Inf to the grammar or drop it from the spec.

[mccanne]

This is a bug in the jsup parser for any input that is a single identifier then EOF. It shouldn't be too hard to fix.

[mccanne]

Regarding Inf, in IEEE floating point, there's a +Inf and a -Inf. I think we should have one way to represent +Inf in Super JSON and we could change +Inf to Inf. I think we went with +Inf because the golang fmt library prints it that way.

[philrz]

Thanks @mccanne. I've opened #5391 to track the JSUP parser bug and #5392 to track possibly changing +Inf to just Inf.