nextfloat(x), prevfloat(x), eps(x) respect precision of BigFloat

[dpsanders]

nextfloat(x), prevfloat(x) and eps(x) now respect the precision of the BigFloat x, rather than using the global BigFloat precision.

This includes a rewritten generic eps function.

Part of #13155

[dpsanders]

cc @simonbyrne

[dpsanders]

Is this considered a feature? I.e. do I need to polish it off soonish to get into 0.6?

[tkelman]

a new keyword argument to an exported constructor looks like a feature to me

[dpsanders]

Is there any chance of getting some of these changes into 1.0?

[simonbyrne]

I'm in favour. The main I would like to see changes are:

1. eps(x::BigFloat) should have the same precision as x
2. check that eps(Float32)/eps(Float64) still get reduced to constants at compile time
3. add documentation noting that these functions ignore global precision.

[dpsanders]

OK thanks - I agree.
I think that changing nextfloat and prevfloat to also respect the precision of x goes along with that?

[simonbyrne]

Yes, sorry I didn't mention that because you had already made that change!

[laborg]

@dpsanders mind to close this? As of now nextfloat and prevfloat respect the precision of BigFloat.

[dpsanders]

Thanks @laborg!