WIP: update for julia 0.6

[bjarthur]

tests are still failing to a macro call in Compose:

while loading /Users/arthurb/.julia/v0.6/Gadfly/test/runtests.jl, in expression starting on line 202
FAILED!
ERROR: LoadError: UndefVarError: codes not defined
Stacktrace:
 [1] macro expansion at /Users/arthurb/.julia/v0.6/Compose/src/misc.jl:125 [inlined]
 [2] jscall(::Array{String,1}, ::Array{Array{Measures.Measure,1},1}) at /Users/arthurb/.julia/v0.6/Compose/src/property.jl:587 (repeats 2 times)
 [3] (::Gadfly.Guide.##3#11{Array{Int64,1},Array{ColorTypes.Color,1},DataStructures.OrderedDict{ColorTypes.Color,AbstractString},Gadfly.Scale.#labeler#66,Compose.Context,Gadfly.Theme,Measures.Length{:mm,Float64},Measures.Length{:mm,Float64}})(::Compose.ParentDrawContext) at /Users/arthurb/.julia/v0.6/Gadfly/src/guide.jl:277

can anyone give me a hint here?

there is still also the show / display code to port.

[tlnagy]

That's not a very helpful error message...

[tlnagy]

Somehow I think that codes is not passing properly to this function and then the inner macro call. No idea why that is happening now. https://github.com/GiovineItalia/Compose.jl/blob/7ea1616cc084e6426bb5a6db422d564328cd6edd/src/property.jl#L585-L591

[codecov-io]

Codecov Report

Merging #973 into master will increase coverage by 0.04%.
The diff coverage is 78.62%.

@@            Coverage Diff             @@
##           master     #973      +/-   ##
==========================================
+ Coverage   78.65%   78.69%   +0.04%     
==========================================
  Files          33       33              
  Lines        4000     4008       +8     
==========================================
+ Hits         3146     3154       +8     
  Misses        854      854

Impacted Files	Coverage Δ
src/geometry.jl	71.42% <ø> (ø)	⬆️
src/coord.jl	83.65% <ø> (ø)	⬆️
src/mapping.jl	20.15% <0%> (-0.16%)	⬇️
src/format.jl	0% <0%> (ø)	⬆️
src/geom/rectbin.jl	70.58% <0%> (ø)	⬆️
src/geom/point.jl	97.56% <100%> (ø)	⬆️
src/geom/ribbon.jl	100% <100%> (ø)	⬆️
src/aesthetics.jl	76.69% <100%> (ø)	⬆️
src/scale.jl	85.43% <100%> (+0.39%)	⬆️
src/geom/bar.jl	91.66% <100%> (ø)	⬆️
... and 17 more

---

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 5a129fb...994ffa1. Read the comment docs.

[bjarthur]

all tests pass locally now.

there are still some deprecation warnings though, mostly from FixedSizeArrays.jl, which Contour.jl uses. we should consider switching to StaticArrays.jl at some point. difficulty there is that it has dropped support for 0.5.

Media.jl also throws a couple warnings.

i would still suggest not merging this PR until folks have actually used it. i have not visually checked the appearance of any plots yet to make sure they look okay.

[bjarthur]

wait, nevermind, the codes not defined error has reared it's head again...

[bjarthur]

the codes not defined error was a macro hygiene problem. not sure how it ever worked on julia 0.5. adding a few esc() got it working on 0.6.

now the tests are stuck on a world age problem...

[tlnagy]

Awesome! Thanks for finding that out. There seem to be a couple bugs in 0.5 that Gadfly/Compose relied on that were fixed in 0.6

[bjarthur]

the world age problem is now fixed, and all tests pass locally on julia 0.5 and 0.6!

would be great if someone could give it a try. you'll need to checkout master on Loess, Hexagons, and Showoff, and checkout bja/julia06 on Compose and Gadfly.

there are still a couple depwarns from the Media.jl dependency, which is for Juno.

and a dozen or so depwarns remain from FixedSizeArrays.jl, which Contour.jl uses. the plan is not to ever upgrade FixedSizeArrays.jl to julia 0.6, but rather to migrate everything that depends on it to StaticArrays.jl.

there are also quite a few broadcast depwarns, and a handful from cycle. the file/line# info is not specific enough for me fix them. i'm guessing they come from some dependency, perhaps FixedSizeArrays.

at any rate, as soon as my Compose PR is merged and tagged, the travis tests should pass for this PR on nightly, and i think we can merge.

[tlnagy]

Is StaticArrays a drop in replacement for FixedSizeArrays? Because then we could move over to StaticArrays for both 0.5 and 0.6. Otherwise, we might have play around with conditional dependencies or something.

[bjarthur]

the names of types have changed, but otherwise it's drop-in. whenever someone gets around to migrating Contour from FixedSizeArrays to StaticArrays, Gadfly will either have to drop support for julia 0.5 or add an upper bound on the Contour version. until then, we just have to live with a few deprecation warnings.

[tlnagy]

StaticArrays appears to have 0.5 as the lower bound so we could probably migrate Contour right now without dropping 0.5 support.

[bjarthur]

this says 0.6-. also, Contour.jl is in the process of being moved to JuliaGeometry.

[tlnagy]

Ah, I missed that. I just read the README, but I guess that was out of date. I guess the moment that 0.6 drops we can make a final release for 0.5 and then move master solely to 0.6 and switch to staticarrays.

[ararslan]

Does anything else need to happen before you guys can make a 0.6-compatible tag of Gadfly?

[bjarthur]

someone should check that it works in the various IDEs... Juno, IJulia, JuliaBox. i've only tested on the REPL. specifically, Media.jl throws a few dep warns.

[ararslan]

IIRC the version of 0.6 on JuliaBox is pretty old so whether Gadfly works there is probably not entirely representative. If I may, I'd recommend that if it works in the REPL you'd might as well tag a version since mostly works is better than not working at all, and as Tony says, "tags are cheap."

[bjarthur]

@tlnagy @Mattriks either of you have any plans to submit PRs in the imminent future? if not, we should tag a release.

[Mattriks]

Yes I hope to submit Geom.vectorfield (#608 & #791) say next week sometime.

[tlnagy]

As @tkelman says, tagging is cheap. Lets go ahead and go forward with tagging a release so we have 0.6 compat.