print(::Float32) should use e instead of Julia specific f

[kshramt]

Note:

Currently, showcompact(::Float32) prints typed representation for Float32 (e.g. 1.0f10).
Although this behaviour is inconsistent with the documentation of showcompact,
I kept the behaviour as is, following the discussions of #17720.

[tkelman]

This is getting pretty confusing to keep track of when a Float32 should be printed with e and when it should be printed with f. I think it would be better to be consistent about always printing with f by default, and consumers like JSON.jl that want to convert to other formats can specifically request a format that uses e.

[JeffBezanson]

The second change here uses f in even more places than before, and introduces a further inconsistency:

julia> [1.0f0]
1-element Array{Float32,1}:
 1.0f0

julia> [1.0e0]
1-element Array{Float64,1}:
 1.0

The typed flag means to always print the type-identifying f, as opposed to only printing it in place of e.

[kshramt]

@tkelman
I am -1.0f0 for always printing f.
As documented, show should be used if a value should be printed with Julia-specific details.
print should not print Julia-specific details.

http://docs.julialang.org/en/latest/stdlib/io-network/?highlight=showcompact#Base.show

The representation used by show generally includes Julia-specific formatting and type information.

http://docs.julialang.org/en/latest/stdlib/io-network/?highlight=showcompact#Base.print

The representation used by print includes minimal formatting and tries to avoid Julia-specific details.

@JeffBezanson
#17720 requires show to print the redundant type information f for elements of a Float32 array.
I am -1 for the behavior of #17720, but tried to follow the behavior.
If I understand correctly, #17720 requires show([1f0]) to be printed as

1-element Array{Float32,1}:
 1.0f0

instead of

1-element Array{Float32,1}:                                                                                           │
 1.0

to make it clear that the element in the array is Float32.

(Travis CI failure is due to a time-limitation for OSX build.)

[slundberg]

Any update here? The current status is that JSON fails with small Float32 values...

[JeffBezanson]

Working on it.