Leakage in contourf filled contours?

[rafael-guerra-www]

There seems to be some leakage in GMT's contourf filled contours, as per code below and attached picture with result.
I have GMT "6.2.0_6f6d002-dirty_2020.10.19" installed in Windows 10 and I am running Julia 1.5.2 using VS Code.

using GMT
G = GMT.peaks(N=100)
zmin, zmax = floor(minimum(G)), ceil(maximum(G))
colorstep = 0.1; contstep = 1; annotstep = 2;
C = makecpt(T=(zmin,zmax,colorstep), cmap=:geo);
contourf(G, C, cont=contstep, annot=annotstep, conf=(MAP_GRID_PEN_PRIMARY=0.1,),
        frame=(axes=(:left_full, :bot_full),),
        xaxis=(grid=1, annot=1, ticks=:auto, label="X [km]"),
        yaxis=(grid=1, annot=1, ticks=:auto, label="Y [km]") )
colorbar!(pos=(paper=true,anchor=(13,1),size=(10,0.5), vertical=true),
        frame=(annot=1, ticks=1, xlabel="Elevation [m]"), fmt=:png, show=true)

[joa-quim]

Hmm, don't know what is going on but it's on the pure GMT side. Saving the peaks grid and

makecpt -Cgeo -T-7.0/9.0/0.1 > lixo.cpt
grdview peaks.grd -JX12c/0 -Bpa1fg1 -BWS -Qs -R-3/3/-3/3 -Clixo.cpt -P  > lixo.ps

shows the same effect
@PaulWessel

[PaulWessel]

Just letting you know I am looking at this. Not visible with -Qidpi.

[PaulWessel]

grdview has lots of comments regarding past problems (near line 1660), which is very helpful. We are making sure small (the noise we add to prevent a contour from going exactly through a node) is not smaller then floating point EPS, but it can be larger. Here it was 0.000364651 based on

small = GMT_CONV4_LIMIT * (Z->header->z_max - Z->header->z_min);

I wonder why it cannot just (always) be just slightly larger than EPS (e.g., 1.0e-7). When I set small to that value the problem went away. So I understand the problem and will work on a solution.

[joa-quim]

This is now solved in GMT.

[rafael-guerra-www]

Dear Joaquim and Paul, just a word of recognition for your follow up and for the very fast code fix. Thank you.