Set Printing

[samuelcolvin]

Currently if you display a Set the entire thing is slammed to STDOUT, if it's very long it can take ages/fill up the shell.

Shouldn't Sets have a pretty show function that chops out some elements same as arrays?

[stevengj]

Yes. Although it shouldn't be show: the pretty/truncated output should be for writemime(io, ::MIME"text/plain", x::Set)

[stevengj]

I'm tempted to write something very similar to the matrix output:

import Base: writemime, with_output_limit, print_matrix

function writemime(io::IO, ::MIME"text/plain", S::Set)
    print(io, summary(S))
    if !isempty(S)
        println(io, ":")
        with_output_limit(()->print_matrix(io, sort!([x for x in S])))
    end
end

This produces nice-looking, sorted, truncated output. It is Θ(length(S) * log(length(S))) complexity, but since the current show method is Θ(length(S)) it isn't much of a regression.

[stevengj]

If we are worried about the overhead of making a copy of S etc. for really huge sets (gigabytes), there could be a threshold size over which it just makes an array of the first 1000 elements of S and prints those, unsorted. For example, modify the above to:

import Base: writemime, with_output_limit, print_matrix

function writemime{T}(io::IO, ::MIME"text/plain", S::Set{T})
    print(io, summary(S))
    if !isempty(S)
        println(io, ":")
        if length(S) > 10^7
            A = copy!(Array(T, 1000), 1, S, 1, 1000)
        else
            A = sort!(T[x for x in S])
        end
        with_output_limit(()->print_matrix(io, A))
    end
end

[stevengj]

Though I guess you would only want to sort! if method_exists(isless, (T, T)).

[samuelcolvin]

Surely it's not canonical to sort the set prior to printing it?

It might confuse people thinking the set is actually sorted. Best and easiest just to pump it out as it comes.

[stevengj]

The set elements are unordered, so it is equally valid to print them in any order we want. So why not print it in an order where it is much easier to see what is in the set.

Quiz: with which ordering is it easier to tell what is in the following set:

 30  32  2  16  26  10  8  22  6  24  4  28  20  14  12  18

or

 2  4  6  8  10  12  14  16  18  20  22  24  26  28  30  32

?

Especially if we are going to print only a subset of the set, it is much more useful to print the largest and smallest elements (as in my code above) than it is to print some random selection of elements.

[samuelcolvin]

Fail point it would be easier, but you will get people thinking the set is naturally sorted. I suppose that's just the way it is.

I suppose it just needs to be clear in the docs what a set is and isn't.

[samuelcolvin]

How do we make this happen? Shall I submit a pull request?

[stevengj]

Sure.

[pkofod]

This seems fixed? Or am I misunderstanding what the issue is?

Set(map(x->x^2,1:4:356))
Set([8649,31329,47089,1369,2401,108241,289,27225,43681,625  …  50625,66049,11881,83521,119025,48841,11025,93025,841,34225])

[samuelcolvin]

The output still doesn't appear to be ordered correctly.

[pkofod]

I figured they weren't supposed to be ordered, as you argued for yourself. In that case, more work is needed. Is there an easy way to see which writemime is called?

[samuelcolvin]

you're right I argued order doesn't matter but as @stevengj said order makes checking for the presence of a value visually much much easier.

[pkofod]

Okay. The question remains, how will you sort

 Set(Any[linspace(1,4,240), 3, rand(20,30), rand(3), "a", 'a'])

? :)

(which by the way looks a lot better than Any[linspace(1,4,240), 3, rand(20,30), rand(3), "a", 'a'] on my computer)

[stevengj]

@pkofod, since there is no isless method for Any, you wouldn't sort in that case.

[pkofod]

Okay, so

if method_exists(isless, (T, T))
show a sorted version
else
show unsorted version
end

That makes sense. What is the easiest way to find the writemime method for a given type @stevengj ? I'm finding it a bit hard to see what takes care of Set{T}. Is there a tool like @which for functions, but which shows the writemime being used?

[vtjnash]

cross-link with #14052

[zslie]

Didn't mean to spam this issue - I jacked up the commit message and did not know it would link before creating the pull request.

[JeffBezanson]

I think the problem with sorting is that a set has to iterate its elements in some order, so when we print it we should show what that order is.

[vtjnash]

Shouldn't Sets have a pretty show function that chops out some elements same as arrays?

As noted above by #7153 (comment), this is fixed:

julia> Set(1:1000)
Set([306, 29, 74, 905, 176, 892, 285, 318, 873, 975  …  632, 618, 341, 186, 321, 420, 6, 856, 322, 218])

Also, consensus (by bureaucratic delay) seems to be against adding ordering to the printing.