Incorrect conversion for very small numbers

[dpsanders]

julia> @interval "1e-400"
[0, 0]

julia> I"1e-400"
[0, 0]

[lbenet]

Note that:

julia> nextfloat(0.0)
5.0e-324

[dpsanders]

To fix this we would have to do all parsing via BigFloat, which may be very slow.

[dpsanders]

julia> x = parse(Interval{BigFloat}, "1e-400")
[0, 0]₂₅₆

There is also an error here!

[dpsanders]

This now gives

parse(Float64, "1e-400", RoundDown)
ERROR: ArgumentError: cannot parse "1e-400" as Float64

[dpsanders]

We now have

julia> parse(Interval{BigFloat}, "1e-400")
Interval(9.9999999999999993e-401, 1.0000000000000001e-400)

We should probably use the new parse with rounding that was implemented for Windows also on the other systems to avoid these problems.

[dpsanders]

We have

julia> convert(Interval{Float64}, parse(Interval{BigFloat}, "1e-400"))
Interval(0.0, 5.0e-324)

This is correct.

[lucaferranti]

this is fixed in #522

julia> @interval "1e-400"
[0, 4.94066e-324]

julia> I"1e-400"
[0, 4.94066e-324]

[lucaferranti]

@lbenet how should we deal with issues like these? Should we close them when the PR is merged to 1.0-dev or when 1.0-dev is merged to master?

[lbenet]

We should probably keep this open to remember to port back that back from 1.0-dev... or if we don't, close them once 1.0-dev becomes 1.0 and is merged to master 😄

[Kolaru]

Fixed in #567