Unified recordGraphics

[grantmcdermott]

Fixes #313
Fixes posit-dev/positron#7316

This PR wraps all of the tinyplot drawing logic in a single, unified recordGraphics({...}) call and thereby resolves the layering + resizing issues that we have been running into for faceted plots, as well as some related problems specific to the Positron IDE. The code is a little (tiny?) bit uglier, but that's a trade-off I'm happy to make if we can avoid unexpected plotting errors when adding layers.

(Previously, we had separate recordGraphics() logic for the draw_legend and draw_facet_window stages, but something was obviously falling through the cracks that was leading to misalignment of added layers.)

My screen (video) recorder isn't working for some reason, but here's a simple screenshot of a correctly aligned layers from a Positron session.

Please give this a quick test drive when you get a sec @zeileis, @vincentarelbundock and/or @jhpantel. Thanks!

[grantmcdermott]

Just to add: I know the diff makes it look like I changed all of tinyplot.R, but a lot of that is just formatting, i.e. we gain extra indentation because of wrapping lines within

recordGraphics(
   {
      ...
   },
   ...
)

There's very little changed in terms of the actual code logic or lines of code. We just wrap more steps in the recording brace.

[zeileis]

Thanks, this is great! I've played around with a couple of examples which all work nicely now.

[grantmcdermott]

Ugh Everything was looking great, but I've just tested a local website build and it looks like this reintroduces a similar problem to the one that was solved by @vincentarelbundock in #425. Specifically, we only get one rendered plot for every example page on the website.

It doesn't appear to be theme specific, since every help documentation page is affected (here: using type_histogram as an example.)

Running these same Examples manually one at a time seems to work fine... But copying the full examples code chunk and running en masse leads to a similar problem... we get multiple copies of the same, final graph.

I'd have to do more digging to figure this out. But please feel free to test and jump in with suggestions if you any.

[grantmcdermott]

Update: This appears to be a recurrence of #388 (resolved in #394).

What's really odd (annoying) is that the problem is not consistent across IDEs or runtimes.

• If I run example(tinyplot) or source('test.R') from Positron, then all of the plots render correctly.
• If I try the exact same commands from RStudio, then I run into the same problems as Not all doc examples are being displayed on the website #388.

Obviously, calling altdoc::render_docs() suffers from the same issue as RStudio...

[zeileis]

I played around with your test.R script:

library("tinyplot")
tinyplot(Sepal.Length ~ Petal.Length | Species, data = iris)
# Sys.sleep(1e-1)
tinyplot(Sepal.Length ~ Petal.Length | Species, data = iris, pch = 16, cex = 2)

Instead of doing the Sys.sleep(1e-1) in between the plots, it also works to do plot.new() in between the plots. This also has the effect that both plots show up in my RStudio plot history. Maybe it helps to put the plot.new() call in a different position within the recordGraphics() - or maybe to put it outside the recordGraphics()? But I'm really just guessing here...

[grantmcdermott]

Thanks Achim. Fingers crossed I've found a simpler workaround. Will test after dinner.

[grantmcdermott]

Closing in favour of #438.

[jhpantel]

Thanks for this fix, and all of your hard work. My original example, and some of the ones I tried from these threads, all worked perfectly, great!