Using withr::with_seed does not propagate to ggrepel::geom_label_repel seed by default

[radlinsky]

This is more of an FYI and question than an issue because I solved it.. it confused me for a solid few hours.. similarly to #97 when I snapshot_review, my plots looked identical by eye, but the log showed very subtle differences. Spot the difference?

< before                                                                       
> after                                                                        
@@ 40,5 / 40,5 @@                                                              
  <circle cx='261.35' cy='114.31' r='8.71' style='stroke-width: 0.35;' />      
  <line x1='261.35' y1='93.46' x2='261.35' y2='113.23' style='stroke-width: 1.07
  ; stroke: #868686; stroke-opacity: 0.70; stroke-dasharray: 1.42,4.27;' />    
< <polygon points='246.86,93.46 275.84,93.46 275.76,93.46 276.10,93.45 276.44,93
: .38 276.77,93.25 277.07,93.08 277.34,92.86 277.57,92.60 277.75,92.31 277.89,91
: .99 277.97,91.65 278.00,91.30 278.00,91.30 278.00,82.55 278.00,82.55 277.97,82
: .20 277.89,81.86 277.75,81.55 277.57,81.25 277.34,80.99 277.07,80.77 276.77,80
: .60 276.44,80.47 276.10,80.40 275.84,80.39 246.86,80.39 247.12,80.40 246.77,80
: .39 246.43,80.43 246.09,80.53 245.78,80.68 245.49,80.88 245.24,81.12 245.03,81
: .39 244.87,81.70 244.76,82.03 244.71,82.38 244.70,82.55 244.70,91.30 244.71,91
: .13 244.71,91.48 244.76,91.82 244.87,92.15 245.03,92.46 245.24,92.74 245.49,92
: .98 245.78,93.17 246.09,93.32 246.43,93.42 246.77,93.46 ' style='stroke-width:
:  0.53; stroke: #000000; stroke-opacity: 0.70; fill: #FF8000; fill-opacity: 0.7
: 0;' />                                                                       
> <polygon points='246.86,93.46 275.84,93.46 275.75,93.46 276.10,93.45 276.44,93
: .38 276.77,93.25 277.07,93.08 277.34,92.86 277.57,92.60 277.75,92.31 277.89,91
: .99 277.97,91.65 278.00,91.30 278.00,91.30 278.00,82.55 278.00,82.55 277.97,82
: .20 277.89,81.86 277.75,81.55 277.57,81.25 277.34,80.99 277.07,80.77 276.77,80
: .60 276.44,80.47 276.10,80.40 275.84,80.39 246.86,80.39 247.12,80.40 246.77,80
: .39 246.42,80.43 246.09,80.53 245.78,80.68 245.49,80.88 245.24,81.12 245.03,81
: .39 244.87,81.70 244.76,82.03 244.70,82.38 244.70,82.55 244.70,91.30 244.70,91
: .13 244.70,91.48 244.76,91.82 244.87,92.15 245.03,92.46 245.24,92.74 245.49,92
: .98 245.78,93.17 246.09,93.32 246.42,93.42 246.77,93.46 ' style='stroke-width:
:  0.53; stroke: #000000; stroke-opacity: 0.70; fill: #FF8000; fill-opacity: 0.7
: 0;' />                                                                       
  <text x='261.35' y='89.86' text-anchor='middle' style='font-size: 8.54px; fill
  : #000000; fill-opacity: 0.70; font-family: sans;' textLength='26.10px' length
  Adjust='spacingAndGlyphs'>chr1#4</text>                                      
  </g>                                                                         

Third set of points 275.76,93.46 vs 275.75,93.46

In my unit test, I was wrapping plot generation with withr::with_seed so I expected identical plots. However, the plot generation included ggrepel::geom_label_repel which ignores seeds by default. After I set geom_text_repel seed manually, then the plots do generate identically, and vdiffr passes.

@lionel- what about adding an option to do some kind of rounding to allow for very slight differences like this?

[radlinsky]

Actually, never mind, devtools::test() passes, but I still get the failure when I run devtools::check()...

I think I've narrowed it down to something to do with gtable. The plot looks every so slightly squished on the x-axis in the devtools::check version, compared to the devtools::test version (and the fonts all look like the same font, albeit maybe slightly different kerneling)