Gadfly, PyPlot, Winston all don't work with 0.5 dev

[dioptre]

Getting lots of BoundsErrors

ERROR: BoundsError: attempt to access (1,2,false)

Ie.

julia> using Winston

WARNING: deprecated syntax "AbstractVecOrMat{T} (" at /home/a/.julia/v0.5/Winston/src/Winston.jl:96.
Use "AbstractVecOrMat{T}(" instead.
WARNING: requiring "Dates" in module "Winston" did not define a corresponding module.
WARNING: module Winston should explicitly import * from Base
WARNING: super(T::DataType) is deprecated, use supertype(T) instead.
in depwarn at ./deprecated.jl:73
in super at ./deprecated.jl:50
in iniattr at /home/a/.julia/v0.5/Winston/src/Winston.jl:2677
in call at /home/a/.julia/v0.5/Winston/src/Winston.jl:2676
in call at /home/a/.julia/v0.5/Winston/src/Winston.jl:1038
in include at ./boot.jl:262
in include_from_node1 at ./loading.jl:401
in eval at ./boot.jl:265
[inlined code] from ./sysimg.jl:14
in require at ./loading.jl:332
in eval at ./boot.jl:265
while loading /home/a/.julia/v0.5/Winston/src/Winston.jl, in expression starting on line 2814
WARNING: super(T::DataType) is deprecated, use supertype(T) instead.
in depwarn at ./deprecated.jl:73
in super at ./deprecated.jl:50
in iniattr at /home/a/.julia/v0.5/Winston/src/Winston.jl:2678
in call at /home/a/.julia/v0.5/Winston/src/Winston.jl:2676
in call at /home/a/.julia/v0.5/Winston/src/Winston.jl:1038
in include at ./boot.jl:262
in include_from_node1 at ./loading.jl:401
in eval at ./boot.jl:265
[inlined code] from ./sysimg.jl:14
in require at ./loading.jl:332
in eval at ./boot.jl:265
while loading /home/a/.julia/v0.5/Winston/src/Winston.jl, in expression starting on line 2814

julia> plot([sin(a) for a in 0.0:0.1:2 * pi])
error during Tk callback:
ERROR: BoundsError: attempt to access (1,2,false)
at index [4]
Error showing value of type Winston.FramedPlot:
ERROR: BoundsError: attempt to access (1,2,false)
at index [4]

[tkelman]

At the moment, unless you can reduce these (see e.g. #14704) to standalone code it's unclear whether the underlying issue is a bug in Base or a bug in these packages. See #14623 (comment) for the underlying cause of the Gadfly issue.

[JeffBezanson]

I believe this is due to calling Base.Grisu.grisu, which now returns 3 values instead of 4.

[JeffBezanson]

I didn't realize packages called grisu directly. Hopefully the authors of that code can suggest a number formatting API that would be more useful.

[tkelman]

Yep: https://github.com/nolta/Winston.jl/blob/808a3182f4622dd7f75925799ee2080f76a4603f/src/Winston.jl#L462

Something about grisu should probably be done in Compat.jl, after we can come up with a suitable API. Both Winston and Showoff look like they're making use of the buffer output that was removed, so not sure what the best approach will be. People had also been taking a look at https://github.com/marcandrysco/Errol recently so it's not unlikely that we might move away from a Grisu-specific implementation at some point.

And JuliaPy/PyPlot.jl#190 should cover PyPlot unless that gets determined to be a base issue.

[JeffBezanson]

It's easy to patch up by using Base.Grisu.DIGITS as the 4th return value. AFAICT it always returned that object in these use cases.

[alsam]

Gadfly works ok for me with the patch:

~/.julia/v0.5/Showoff/src$ git diff -u
diff --git a/src/Showoff.jl b/src/Showoff.jl
index 425ca9a..bd95fe1 100644
--- a/src/Showoff.jl
+++ b/src/Showoff.jl
@@ -26,7 +26,9 @@ function grisu(v::AbstractFloat, mode, requested_digits)
         @grisu_ccall v mode requested_digits
         return Base.Grisu.LEN[1], Base.Grisu.POINT[1], Base.Grisu.NEG, Base.Grisu.DIGITS
     else
-        return Base.Grisu.grisu(v, mode, requested_digits)
+        #return Base.Grisu.grisu(v, mode, requested_digits)
+        len, point, neg = Base.Grisu.grisu(v, mode, requested_digits)
+        return len, point, neg, Base.Grisu.DIGITS
     end
 end

[tkelman]

would need to make that conditional on the julia version number of jeff's commit, otherwise neg would be a tuple on 0.4.

[RaulDurand]

Any update on PyPlot?