pretty_table should require_one_based_indexing

[yurivish]

Currently pretty_table prints incorrect information about arrays with offset axes.

Julia has a function Base.require_one_based_indexing that can be used to disallow such arrays as input and throw an error instead of exhibiting undefined behavior or returning incorrect results.

julia> using PrettyTables, OffsetArrays

julia> a = OffsetArray(0:10, -5:5)
0:10 with indices -5:5

julia> pretty_table(a)
┌────────┐
│ Col. 1 │
├────────┤
│      6 │
│      7 │
│      8 │
│      9 │
│     10 │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
└────────┘

[ronisbr]

Hi @yurivish,

Thanks for the tip! I will add this to avoid those problems.

However, I just had an amazing idea to support offset arrays. Inside PrettyTables, I can wrap the array inside a custom type (I already do this for Tables.jl). The only functions I need to create for this new type is getindex, size, and isassigned. Hence, I can check inside those if we have an offset array an just do the conversion so that table[1,1] returns the first element and so on.

[yurivish]

Interesting idea! Could that new type itself be an OffsetArray that cancels out the offset of the array that was passed to it (which might be an OffsetArray or any other type with offset axes)?

[ronisbr]

I do not want it to be dependent of OffsetArray. The idea is something like

struct GenericTable
    data
    offset
    size
end

Then we can do something like:

getindex(table::GenericTable, inds...) = table[inds[1] - table.offset[1], inds[2] - table.offset[2]]

This code will not work, but it illustrates my idea :)

[GlenHertz]

Is there a reason too not use firstindex(obj):lastindex(obj) for the indices of the object? I believe those could be defined with @compat for older version of Julia that don't have those functions.

[ronisbr]

Is there a reason too not use firstindex(obj):lastindex(obj) for the indices of the object? I believe those could be defined with @compat for older version of Julia that don't have those functions.

Can you please provide more details? lastindex of a Matrix returns the number of elements and I need to know the structure (number of elements in a row and how many rows we have).

[GlenHertz]

I'm trying to do something similar to:

julia> pretty_table(OffsetVector(1:11, -5:5))
┌────────┐
│ Col. 1 │
├────────┤
│      7 │
│      8 │
│      9 │
│     10 │
│     11 │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
│ #undef │
└────────┘

but I have my own struct that doesn't use 1-based indexing and it also doesn't work (same as above).

I haven't looked into the root issue (it could be from Tables.jl as I implemented the AbstractColumns interface from it to get it compatible with PrettyTables).

[ronisbr]

Hi @yurivish and @GlenHertz !

First of all, sorry for the HUGE delay. PrettyTables needed a huge rewrite in the algorithm that handles the input table internally before supporting OffsetArrays. Everything is done and we now support OffsetArrays in master! Can you please try?

julia> using PrettyTables

julia> using OffsetArrays

julia> A = OffsetArray(rand(3,3), -2:0, -3:-1);

julia> pretty_table(A, show_row_number = true)
┌─────┬──────────┬──────────┬──────────┐
│ Row │  Col. -3 │  Col. -2 │  Col. -1 │
├─────┼──────────┼──────────┼──────────┤
│  -2 │ 0.332428 │ 0.698186 │ 0.251798 │
│  -1 │ 0.285071 │ 0.434293 │ 0.987304 │
│   0 │ 0.515081 │ 0.756921 │ 0.397783 │
└─────┴──────────┴──────────┴──────────┘