Bring back CI and upgrade to Julia 1.6

[IanButterworth]

Progress:

• Switched over to github actions CI
• Created new test environment
• Updated reference tests (needs careful review!! I just blanket updated them without understanding whether it was correct to do so)
• Added a force_raw_txt kwarg to ReferenceTests.@reference_test in Allow forcing reading the reference file as raw txt/string JuliaTesting/ReferenceTests.jl#91, as a few of the reference test .txt files were being accidentally detected by FileIO to be Gadget2 format

Locally with the above PR, tests get to the barplot tests and fail on the @inferred because Plot is inferred, but Plot{BarplotGraphics{Real}} is returned

Fix #138
Fix #134
Close #119

[t-bltg]

Updated reference tests (needs careful review!! I just blanket updated them without understanding whether it was correct to do so)

I believe there is a change in the random sequences provided by Random (we observed this in Plots.jl), so let's use a stable rng from StableRNGs.

@IanButterworth, the corresponding PR on your branch is here: IanButterworth#1.

[IanButterworth]

Great. This means the reference images need updating again? Feel free to PR that update also

[t-bltg]

Great. This means the reference images need updating again? Feel free to PR that update also

Thanks, yes, I can generate the refs images again.

Can you maybe add me as a collaborator to your fork ? That would allow me to push directly into this branch, and not annoy you with more PRs.

[IanButterworth]

Can you maybe add me as a collaborator to your fork

done 👍🏻

[t-bltg]

@IanButterworth , I think this PR is ready. Do you mind editing your first post by adding:

Fix https://github.com/Evizero/UnicodePlots.jl/issues/138.
Fix https://github.com/Evizero/UnicodePlots.jl/issues/134.
Close https://github.com/Evizero/UnicodePlots.jl/pull/119.

for Github to link fixed issues ?

@Evizero, can you review this PR, so that we get a running CI ?

[johnnychen94]

I didn't check the reference files. Hopefully, that's all because of the change of random generator. As long as the CI passes, I think it's good to merge.

[t-bltg]

Interesting: on 1.2 and 1.6, v = c.grid[x,row] + 1 is correct but lookup_decode[v] has different behaviors across versions.
Shouldn't be too hard to fix using the following mwe:

MWE

Copy lines 1-97 of https://github.com/Evizero/UnicodePlots.jl/blob/master/src/canvas/asciicanvas.jl in a julia repl:

1.0.5 or 1.2.0

julia> VERSION
v"1.0.5"
julia> ascii_decode[429]
'|': ASCII/Unicode U+007c (category Sm: Symbol, math)

1.6.0 or up

julia> VERSION
v"1.6.2"
julia> ascii_decode[429]
'^': ASCII/Unicode U+005E (category Sk: Symbol, modifier)

EDIT:
As I suspected, it is a Dict looping order issue, the solution is to sort before looping:

ascii_decode[0b1] = ' '
for i in 1:511
    min_dist = typemax(Int)
    min_char = ' '
    for (k, v) in sort(collect(pairs(ascii_lookup)), by=x -> x[1])
        cur_dist = count_ones(UInt16(i) ⊻ k)
        if cur_dist < min_dist
            min_dist = cur_dist
            min_char = v
        end
    end
    ascii_decode[i + 1] = min_char
end

[t-bltg]

I'm getting close, now test/tst_spy.jl fails only on 1.0:

[t-bltg]

Finally, all tests pass 🎉 ! Blaming the non-stable sprand across julia minor versions 🤯.

@IanButterworth and @johnnychen94 I'll let you review this while I'm taking a ☕.

This will need another extra check of the reference images.

[t-bltg]

I've carefully visually checked all the references images using:

root="$HOME/Downloads/UnicodePlots.jl"  # fork

for f in $(find $root/test/references -name '*.txt'); do
 echo; echo
 echo "== local($f) =="
 cat $f
 echo
 url="https://raw.githubusercontent.com/Evizero/UnicodePlots.jl/master/${f#$root/}"
 echo "== remote($url) =="
 wget -q -O - $url
done

, and everything looks good (no changes apart from random numbers, as expected).

[johnnychen94]

@t-bltg This looks great! Thank you for the detective work and bringing back CI!

Just two small comments and with that resolved I think we can merge it.