Ubuntu and windows doppelgangers differ

[willgearty]

I'm currently working on a revamp of an R package that can be used to add vectorized images (specifially phylopics) to ggplot and base R plots (https://github.com/palaeoverse-community/rphylopic/pull/37). Everything seems to work fine, except for SOME of the expect_doppelganger tests on Ubuntu. Some of the snapshot svgs that are generated on Ubuntu appear to be VERY slightly different from the respective snapshot svgs that I've generated on my Windows machine.

We've tried everything we can think of and have been unable to make the tests pass. Can anyone think of any reason this might occur on Ubuntu but not the other OSes?

The weird details (https://github.com/palaeoverse-community/rphylopic/actions/runs/4306035830):

• The MacOS snapshots are identical to the Windows snapshots
• The R version doesn't appear to matter
• When a raster image is plotted (imported from png), it doesn't break the doppelganger tests on Ubuntu
• Only the added vectorized image (stored as a grImport2::Picture object) is different on Ubuntu, the rest of the plot is identical
• Only the width of the picture is different, not the height (even though the width of the svg and the widths of all other elements are the same)
• The problem persists for both ggplot and base R

Part of the svg on windows:

<defs>
  <clipPath id='cpMjYxLjUzfDQ4OC4yM3wyMi43OHw1NDUuMTE='>
    <rect x='261.53' y='22.78' width='226.71' height='522.33' />
  </clipPath>
</defs>
<g clip-path='url(#cpMjYxLjUzfDQ4OC4yM3wyMi43OHw1NDUuMTE=)'>
<path d='M 380.82 47.62 L 380.82 47.62 L 380.64 48.10 L 380.43...

Part of the svg on ubuntu (via github actions):

<defs>
  <clipPath id='cpMjYxLjY5fDQ4OC4wN3wyMi43OHw1NDUuMTE='>
    <rect x='261.69' y='22.78' width='226.39' height='522.33' />
  </clipPath>
</defs>
<g clip-path='url(#cpMjYxLjY5fDQ4OC4wN3wyMi43OHw1NDUuMTE=)'>
<path d='M 380.92 47.62 L 380.92 47.62 L 380.73 48.10 L 380.53...

[willgearty]

I've done some more digging and it looks like this is an rsvg issue: https://github.com/ropensci/rsvg/issues/37

[willgearty]

For anyone else that comes across this issue, I was able to resolve this by switching to using Ubuntu 20.04 for my Github Actions tests. It looks like the most recent version of librsvg (which rsvg depends on) on Ubuntu 20.04 is only 2.48.9, which seems close enough to 2.48.8 (the version bundled with the Windows/OSX versions of rsvg) to not cause the issue anymore.