undef does not have parseable repr

[nickrobinson251]

julia> VERSION
v"1.2.0"

julia> repr(undef)
"array initializer with undefined values"

It seems needless, to me, not to have a "parseable" representation here (#30757) (in the sense x == eval(Meta.parse(repr(x)))).

We sort of workaround this in a few places, by having Base.undef_ref_str

julia/base/show.jl

Line 1848 in 90d481e

const undef_ref_str = "#undef"

As far as I can tell, we could just change the current show method from

julia/base/show.jl

Line 3 in 90d481e

show(io::IO, ::UndefInitializer) = print(io, "array initializer with undefined values")

to only be for the text/plain MIME type,
e.g.

show(io::IO, ::MIME"text/plain", ::UndefInitializer) =  print(io, "array initializer with undefined 
values")

Since we have

julia/base/boot.jl

Lines 401 to 402 in 4111609

	struct UndefInitializer end
	const undef = UndefInitializer()

perhaps if we wanted more compact printing we could also define

show(io::IO, ::UndefInitializer) = print(io, "undef")

[nickrobinson251]

Slightly separate from the issue of a parseable repr: we only sometimes get the more compact Base.undef_ref_str printing in cases we probably want it:
e.g.

julia> fill(undef)
0-dimensional Array{UndefInitializer,0}:
array initializer with undefined values

julia> Array{Any, 0}(undef)
0-dimensional Array{Any,0}:
#undef

It seem to me that something like

julia> [undef undef undef undef]
1×4 Array{UndefInitializer,2}:
 array initializer with undefined values  …  array initializer with undefined values

would be much clearer as

julia> [undef undef undef undef]
1×4 Array{UndefInitializer,2}:
 undef  undef  undef  undef

[KristofferC]

would be much clearer as

1×4 Array{UndefInitializer,2}:
 undef  undef  undef  undef```

I disagree. undef is only to initalize arrays with undefined values. It is not an undef value itelf. It is not the same as #undef (which if not a first class Julia object). You should probably never store it inside anything.

[nickrobinson251]

undef is only to initalize arrays with undefined values. It is not an undef value itelf. It is not the same as #undef

That seems fair enough, and we do get a "good" display in sensible use cases, e.g.

julia> Matrix(undef, 1, 4)
1×4 Array{Any,2}:
 #undef  #undef  #undef  #undef

Anyway, this was an aside :)

The main issue is the second case here:

# this seems fine
julia> UndefInitializer()  
array initializer with undefined values

# expected `undef` or `UndefInitializer()`
julia> repr(UndefInitializer())
"array initializer with undefined values" 

[KristofferC]

What is the use case here, i.e. what are you trying to do where this is a problem?

[nickrobinson251]

I noticed this in #33206, trying to fix #31481

[nickrobinson251]

my main use-case of repr in general is to answer "how do I create a value like this?"

[KristofferC]

c = fill(undef)

seems like you tried to fill an array with undefined values and mistakenly used the undef array initializer. To me, the current printing prevented a bug.

[nickrobinson251]

certainly open to suggestions on the 0-dim undef case - i.e. how to print Array{Any, 0}(undef) -- but I think PR #33206 may be a better place to discuss that case :)

[stevengj]

Should be easy to fix: just change base/show.jl from show(io::IO, ::UndefInitializer) = print(io, "array initializer with undefined values") to:

show(io::IO, ::UndefInitializer) = print(io, "undef")
show(io::IO, ::MIME"text/plain", ::UndefInitializer) = print(io, "undef: array initializer with undefined values")

This would keep the verbose interactive display output, but would change the output in containers, print, repr(undef), etc.

I would be in favor of this change.

[stevengj]

I would also be in favor of printing UndefInitializer() instead of undef, since it seems more informative to make it clear that it is an instance of UndefInitializer.