Perform tests in different operative systems

[PabRod]

No description provided.

[PabRod]

@mdingemanse, your hypotheses was 100% right. Looking into png documentation, I see they even have a section about Differences between OSes. I'll look into this.

[PabRod]

The OS dependency of saved plots seems to be a serious issue. See for instance here. This is a big problem regarding snapshots.

[PabRod]

Next step: take a look at ragg.

[mdingemanse]

Just to note that while ragg may offer a solution for this micro issue, we risk getting rabbitholed quite far from the design goals of the package and the aims of the tests. For the functionality of talkr it is not important if there happens to be slight pixel level variation in images created in tests. And for tests, the main use case is a single coder iterating over some bit of code and using the tests on their own system (keeping OS constant) as a sanity check that they're not breaking anything.

I can live with the few times where we'd see a very subtle difference that is not a difference because we ran a test on different OSs. The main issue to solve when it comes to different Issues is the stray Rplots.pdf image being created on Windows, not *nix.

[PabRod]

Take a look, @mdingemanse. I hot-fixed it by running the snapshot tests conditionally, i.e.:, only in Linux.

Surprisingly, there doesn't seem to be a more elegant way of guaranteeing identical outputs across OSs.

[mdingemanse]

okay for now, though I think the more elegant solution will be to break the plot saving code out of the function fully and expect users to find their own OS-specific ways of doing it