CompatHelper: bump compat for "Colors" to "0.11"

[github-actions]

This pull request changes the compat entry for the Colors package from 0.9 to 0.9, 0.11.

This keeps the compat entries for earlier versions.

Note: I have not tested your package with this new compat entry. It is your responsibility to make sure that your package tests pass before you merge this pull request.

[codecov-io]

Codecov Report

Merging #1369 into master will not change coverage.
The diff coverage is n/a.

@@          Coverage Diff           @@
##           master   #1369   +/-   ##
======================================
  Coverage    90.6%   90.6%           
======================================
  Files          38      38           
  Lines        3961    3961           
======================================
  Hits         3589    3589           
  Misses        372     372

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bb210ab...22d25fa. Read the comment docs.

[kimikage]

@bjarthur, how did you confirm the compatibility with Colors.jl v0.11? If you didn't confirm it, this change (or pressing the merge button carelessly) is very dangerous.

[Mattriks]

I note that this PR was opened and tested before Compose.jl's Colors 0.11 compat was triggered (GiovineItalia/Compose.jl#375), I'm guessing that the above Travis test used Colors 0.9.6 not 0.11. Compare with the more recent PRs #1370 and #1371 which both fail on the testscript "discrete_color_hue.jl" for Julia 1.0.5

[kimikage]

I am pending submitting an issue because I have not yet identified the cause. (I think this is not a problem of Gadfly's code base.)

However, I know some possible causes. The significant difference between Colors v0.9.6 and v0.10 which may cause the segmentation fault, is the precompilation (cf. JuliaGraphics/Colors.jl#370)

Gadfly.jl/test/testscripts/discrete_color_hue.jl

Lines 3 to 4 in 7c14b91

	plot(ones(10,3).*[1 2 3], x=Row.index, y=Col.value, color=Col.index,
	Geom.line, Scale.color_discrete_hue(x->range(RGB(1,0,0), stop=RGB(0,0,1), length=x)))

RGB(1,0,0) generates RGB{N0f8}(1,0,0). N0f8 is defined in FixedPointNumbers.
If we do not use RGB{N0f8} (e.g. using RGB{Float32}), we will get different results.

Edit:

FixedPointNumbers	ColorTypes	Test
v0.6.1 (w/o pc)	v0.8.0 (w/o pc)	✔️
v0.6.1 (w/o pc)	v0.8.1 (w/ pc)	❌ segfault
v0.7.0 (w/ pc)	v0.8.0 (w/o pc)	❌ segfault
v0.7.0 (w/ pc)	v0.8.1 (w/ pc)	❌ segfault

cf. JuliaLang/julia#34121

[tlnagy]

I'm guessing that the above Travis test used Colors 0.9.6 not 0.11.

I can confirm that this is true. I merged a PR to revert this change over at #1372, despite tests failing since it's simply returning to the state prior to this merge.

[tlnagy]

I rebased #1371 on #1372 and I'm letting the tests run (everything should pass) and we need to merge it and it should get us some time to work out the issues with Colors v0.10+. That PR will pin Colors back to the v0.9 branch in Gadfly's dependency list until we resolve this.