update docs following CSV.jl 0.9 release

[bkamins]

Fixes #2864

[bkamins]

waiting for a decision in JuliaData/WeakRefStrings.jl#85

[bkamins]

@nalimilan - I have updated the docs. I left String[X] types and added a paragraph explaining how they work.

[quinnj]

I restarted the jobs now that WeakRefStrings has released

[nalimilan]

Looks much better now!

[nalimilan]

Though doctests fail due to that String15… weirdness:

│ Evaluated output:
│ 
│ 3×5 DataFrame
│  Row │ Species          SepalLength_mean  SepalWidth_mean  PetalLength_mean  P ⋯
│      │ String15…        Float64           Float64          Float64           F ⋯
│ ─────┼──────────────────────────────────────────────────────────────────────────
│    1 │ Iris-setosa                 5.006            3.418             1.464    ⋯
│    2 │ Iris-versicolor             5.936            2.77              4.26
│    3 │ Iris-virginica              6.588            2.974             5.552
│                                                                 1 column omitted
│ 
│ Expected output:
│ 
│ 3×5 DataFrame
│  Row │ Species          SepalLength_mean  SepalWidth_mean  PetalLength_mean  P ⋯
│      │ String15         Float64           Float64          Float64           F ⋯
│ ─────┼──────────────────────────────────────────────────────────────────────────
│    1 │ Iris-setosa                 5.006            3.418             1.464    ⋯
│    2 │ Iris-versicolor             5.936            2.77              4.26
│    3 │ Iris-virginica              6.588            2.974             5.552
│                                                                 1 column omitted

Cc: @ronisbr

[bkamins]

Though doctests fail due to that String15… weirdness:

I will check what is the reason of failure and fix it - it is most likely on DataFrames.jl side. The strange thing is that what I put into this PR passes on my local machine.

[bkamins]

The problem was the following:

julia> using DataFrames

julia> import Dates

julia> DataFrame(d=Dates.Date(1))
1×1 DataFrame
 Row │ d
     │ Date…
─────┼────────────
   1 │ 0001-01-01

julia> using Dates

julia> DataFrame(d=Dates.Date(1))
1×1 DataFrame
 Row │ d
     │ Date
─────┼────────────
   1 │ 0001-01-01

Now I have fixed it to never try printing type module name as a prefix (so the latter output is presented consistently).

In this way data frame is displayed in the same way no matter if the module is loaded with using or with import or only indirectly by some other module.

[nalimilan]

For full consistency we should recursively call nameof on each part of the union, as otherwise the module name is printed.

[bkamins]

It is even more complicated 😞. I will try to propose something reasonable. The problem is that nameof drops parameters from parametric types.

[bkamins]

The patches lead me to uncover that we printed MD type incorrectly previously which got uncaught and that we inconsistently printed type information in text vs HTML/LaTeX. Now I made it more consistent (use 9 text width characters by default - @ronisbr - this might affect your work on HTML and LaTeX output, but these are minor changes)

[bkamins]

Just as a comment why 9 if someone would ask. The String[X] types have up to 9 characters of length (and also this is the number we used in text/plain already, I just synced HTML and LaTeX with this reference)

[bkamins]

Thank you!

[nalimilan]

Just curious: why don't we print Dates.DateTime?

Anyway it would make sense to never print the module name to avoid this kind of variation (which is hard to understand for users).

[bkamins]

Just curious: why don't we print Dates.DateTime?

Because we match type-width only to column name, not shown contents (which can vary). I guess @ronisbr could change this if we asked for this. E.g.:

julia> DataFrame(x=Dates.DateTime(1))
1×1 DataFrame
 Row │ x
     │ DateTime…
─────┼─────────────────────
   1 │ 0001-01-01T00:00:00

julia> DataFrame(xxxxxxxxxxxxxxxx=Dates.DateTime(1))
1×1 DataFrame
 Row │ xxxxxxxxxxxxxxxx
     │ Dates.DateTime
─────┼─────────────────────
   1 │ 0001-01-01T00:00:00

Anyway it would make sense to never print the module name to avoid this kind of variation (which is hard to understand for users).

We could go for finding the last dot (i.e. .) where the sequence XXX.YYYY.ZZZZ consists of valid Julia identifiers and strip the front of the string, but it is quite tricky to do in 100% correct way (of course it is doable).

[nalimilan]

If that's doable, I'd say that taking into account the final size width of the column (including contents) would be better. Otherwise there's some wasted space which in some cases could be used to print relevant information.

[bkamins]

@ronisbr - is this EASILY doable? (if it is hard it is probably better to focus on HTML/LaTeX support first). Thank you!