[BUG] Uneven pixel spacing with heatmap in GR backend

[jusack]

Details

When plotting a matrix of RGB colors as a heatmap for certain matrix sizes the pixel spacing of the heatmap are uneven. The severity of the unevenness depends on the size of the matrix that I am plotting. However, I could not recognize any pattern, of when the unevenness is worse.

Below is a MWE of an extreme case (the axis limits are there to see it better, they do not change the unevenness):

using Plots
gr()
heatmap(rand(RGB,300,300), xlims=(0,50), ylims=(0,50))

produces:

Changing the matrix size to 100x100 looks uniform. This error also does not appear when a normal Float value is used instead of a RGB value.

In the plotly backend the RGB heatmap also looks fine:

Backends

This bug occurs on ( insert x below )

Backend	yes	no	untested
gr (default)	x
pythonplot			x
plotlyjs		x
pgfplotsx			x
unicodeplots			x
inspectdr			x
gaston			x

Versions

Plots.jl version: v1.39.0
Backend version: GR v0.72.7
Output of versioninfo():

Commit bed2cd540a (2023-08-24 14:43 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: 8 × Intel(R) Core(TM) i7-10510U CPU @ 1.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, skylake)
  Threads: 4 on 8 virtual cores

[jusack]

Might be related to #4158

[jusack]

I identified the yflip parameter which is set in the following recipe as the cause of this weird scaling:

Plots.jl/src/recipes.jl

Lines 1339 to 1354 in 10ee593

	@recipe function f(mat::AMat{T}) where {T<:Colorant} # COV_EXCL_LINE
	n, m = map(a -> range(first(a) - 0.5, stop = last(a) + 0.5), axes(mat))
	if is_seriestype_supported(:image)
	seriestype := :image
	yflip --> true
	SliceIt, m, n, Surface(mat)
	else
	seriestype := :heatmap
	yflip --> true
	colorbar --> false
	aspect_ratio --> :equal
	z, plotattributes[:fillcolor] = replace_image_with_heatmap(mat)
	SliceIt, m, n, Surface(z)
	end
	end

Another problem is, that the recipe overrides the heatmap series type to image if it is available by the backend and manually defines the x and y coordinates, circumventing the adjustment of the x and y coordinates for centered pixel bins, which is applied only to the heatmap series type in the gr backend by the function heatmap_edges.

I will look into updating the recipes for color matrices and open a PR.

[jusack]

The underlying issue is fixed in gr now sciapp/gr@b9eafa6 Until the fix is released in GR.jl and Plots uses the updated GR version the workaround is to explicitly set yflip=false in the call to heatmap, as it will not be overwritten by the recipe.