Par restore

[grantmcdermott]

Fixes #6 by adding a par_restore argument.

Defaults to par_restore = FALSE, in which case any changes to par during a plot2() call (e.g., from a legend repositioning) will persist afterwards. I initially resisted persisting changes as a default, but in hindsight I think this is more consistent with the normal base plot workflow. (And, in the case of #6, means that we can add elements after a plot is drawn in the regular manner.) Ofc users can opt out of this behaviour by switching to par_restore=TRUE.

Examples:

library(plot2)

op = par(no.readonly = TRUE)

plot2(Sepal.Width ~ Sepal.Length | Species, iris, grid = grid())
points(6,3, pch = 17, col = "red", cex = 1.5)

plot2(
  mpg ~ wt | cyl, mtcars,
  pch = 19,
  grid = grid(),
  legend.position = "right!",
  legend.args = list(title = "How many cylnders do you have?")
  )
lines(lowess(mtcars[["wt"]], mtcars[["mpg"]]), col = "red")

# Caution: Note that adjust margin persists b/c we didn't specify 
plot(1:10)

par(op)

# Try again with par_restore = TRUE
plot2(
  mpg ~ wt | cyl, mtcars,
  pch = 19,
  grid = grid(),
  legend.position = "right!",
  legend.args = list(title = "How many cylnders do you have?"),
  par_restore = TRUE
)

# Adding elements to the plot won't work properly now...
#lines(lowess(mtcars[["wt"]], mtcars[["mpg"]]), col = "red")

# But our normal defaults have been preserved
plot(1:10)

^{Created on 2023-04-09 with reprex v2.0.2}

[grantmcdermott]

@vincentarelbundock I can't get these tinysnapshot tests to pass on GitHub's CI even though I generated them locally using Ubuntu 22.04 on a Rocker image. (As mentioned before, my default Arch system generates slight differences in fonts that also triggers errors.)

I can switch to MacOS, which I have on my work machine. But I'm wondering what the best long-term solution, is given that you and (I think?) @zeileis are on Ubuntu. Any thoughts?

[zeileis]

Just fyi: I'm on Debian testing. Not much I can contribute off the top of my head otherwise. Just maybe that there is the gdiff package from Paul Murrell. I've never worked with it, though.

[vincentarelbundock]

@zeileis tinysnaphot uses a very similar approach to gdiff, based on magick.

@grantmcdermott I added a new argument to tinysnapshot which allows us to set the font. Maybe that will help? I pushed a commit to this PR. Can you try this:

1. Make sure Liberation font is installed on your Rocker. Maybe: sudo apt install fonts-liberation
2. Install the dev version of tinysnapshot: remotes::install_github("vincentarelbundock/tinysnapshot")
3. Pull from this updated PR.
4. Run the test suite locally. tinytest::run_test_dir("inst/tinytest")

The snapshots were re-generated on my computer, and they match the Ubuntu from Github actions. Now I need to figure out if this will work on your local install.

[grantmcdermott]

Great, thanks for taking a look. I'm a bit pushed for time this morning. So let me merge this in a bid for expediency and then I'll test with Liberation when I get a chance later.

[grantmcdermott]

Tested quickly and good news: explicitly requiring Liberation fonts fixes the problem (and on my native Arch system too).

Huzzah!

[zeileis]

Thanks, Vincent @vincentarelbundock, I didn't realize that tinysnapshot was yours!

Will try to use it for our topmodels package as well, this might resolve some problems we had with our current hand-crafted solution.

[vincentarelbundock]

@zeileis Yeah, this is a very very new package. So fair warning: there may (will?) be some rough corners. But I'm always eager to get feedback from early adopters!

Just make sure you read this section of the README on "deterministic" plots, and maybe wait a few days for version 0.0.3 to come out.