Plots in for loop

[cderv]

Opening this issue to follow up on knitr report:

• Display all plots in a for loop (knitted roxygen2 documentation) yihui/knitr#2392

and as a way to keep track of this example. However, possibly related to

• plot handler called too late when evaluating a loop #67

Plots in for loop are not treated the same as plots outside of it.

• Only last is in output

res <- evaluate::evaluate(function(){ 
    library(tinyplot)
    for (thm in c("dark", "minimal")) {
        tinytheme(thm)
        tinyplot(I(Sepal.Length * 1e4) ~ Petal.Length | Species, data = iris)
    }
})
str(res, 1)
#> List of 3
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 2
#>   ..- attr(*, "engineVersion")= int 16
#>   ..- attr(*, "pid")= int 49608
#>   ..- attr(*, "Rversion")=Classes 'R_system_version', 'package_version', 'numeric_version'  hidden list of 1
#>   ..- attr(*, "load")= chr(0) 
#>   ..- attr(*, "attach")= chr(0) 
#>   ..- attr(*, "class")= chr "recordedplot"
#>  - attr(*, "class")= chr [1:2] "evaluate_evaluation" "list"

• Both plots are recorded

res <- evaluate::evaluate(function(){ 
    library(tinyplot)
    tinytheme("dark")
    tinyplot(I(Sepal.Length * 1e4) ~ Petal.Length | Species, data = iris)
    tinytheme("minimal")
    tinyplot(I(Sepal.Length * 1e4) ~ Petal.Length | Species, data = iris)
})
str(res, 1)
#> List of 7
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 2
#>   ..- attr(*, "engineVersion")= int 16
#>   ..- attr(*, "pid")= int 71848
#>   ..- attr(*, "Rversion")=Classes 'R_system_version', 'package_version', 'numeric_version'  hidden list of 1
#>   ..- attr(*, "load")= chr(0) 
#>   ..- attr(*, "attach")= chr(0) 
#>   ..- attr(*, "class")= chr "recordedplot"
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 1
#>   ..- attr(*, "class")= chr "source"
#>  $ :List of 2
#>   ..- attr(*, "engineVersion")= int 16
#>   ..- attr(*, "pid")= int 71848
#>   ..- attr(*, "Rversion")=Classes 'R_system_version', 'package_version', 'numeric_version'  hidden list of 1
#>   ..- attr(*, "load")= chr(0) 
#>   ..- attr(*, "attach")= chr(0) 
#>   ..- attr(*, "class")= chr "recordedplot"
#>  - attr(*, "class")= chr [1:2] "evaluate_evaluation" "list"

Not sure if something can be done around that, but this is now definitely a differentiating approach with how litedown::fuse() handles plot. So I would like to dig into this.

[hadley]

Simpler reprex?

res <- evaluate::evaluate(function(){
  for (i in 1:2) {
    plot(1)
  }
})
sapply(res, class)
#> [1] "source"       "recordedplot"

^{Created on 2025-02-06 with reprex v2.1.0}

[hadley]

Although if I run that directly in the console, I see:

[1] "source"       "recordedplot" "recordedplot"

But I don't see that with tinyplot, which suggests that it's doing something different to plot().

[etiennebacher]

But I don't see that with tinyplot, which suggests that it's doing something different to plot().

It's the opposite for me:

• in reprex() and in the console, using plot() returns [1] "source" "recordedplot"
• in reprex() and in the console, using tinyplot() returns [1] "source" "recordedplot" "recordedplot"

[etiennebacher]

Looks like tinytheme() might be the culprit here:

• without it:

reprex::reprex({
  library(tinyplot)
  for (thm in 1:2) {
    tinyplot(Sepal.Length ~ Petal.Length, data = iris)
  }
})

prints 2 plots.

• with it:

reprex::reprex({
  library(tinyplot)
  tinytheme("default")
  for (thm in 1:2) {
    tinyplot(Sepal.Length ~ Petal.Length, data = iris)
  }
})

prints 1 plot.

What I don't understand is that

reprex::reprex({
  for (thm in 1:2) {
    plot(Sepal.Length ~ Petal.Length, data = iris)
  }
})

also prints a single plot. I can't help further but hopefully that's useful to you.

[cderv]

If the exact same plot is drawn, then it won't be recorded.
plot capture logic is here

evaluate/R/watchout.R

Lines 46 to 74 in 9b41223

	capture_plot <- function(incomplete = FALSE) {
	# no plots open; par("page") will open a device
	if (is.null(dev.list())) {
	return()
	}
	# only record plots for our graphics device
	if (!identical(dev.cur(), dev)) {
	return()
	}
	# current page is incomplete
	if (!par("page") && !incomplete) {
	return()
	}
	plot <- recordPlot()
	if (!makes_visual_change(plot[[1]])) {
	return()
	}
	if (!looks_different(last_plot[[1]], plot[[1]])) {
	return()
	}
	last_plot <<- plot
	push(plot)
	invisible()
	}

So what does tinytheme do exactly that would make the plot different ? Maybe plot change are not correctly detected. If you can give details on this, it would help undertand this better

I tried changing par() in the loop, and got two outputs

res <- evaluate::evaluate(function() {
    for (bg in c("red", "blue")) {
        par(bg = bg)
        plot(Sepal.Length ~ Petal.Length, data = iris)
    }
})
sapply(res, class)
#> [1] "source"       "recordedplot" "recordedplot"

While changing theme for tinyplot does return one

res <- evaluate::evaluate(function(){ 
    library(tinyplot)
    for (thm in c("dark", "minimal")) {
        tinytheme(thm)
        tinyplot(I(Sepal.Length * 1e4) ~ Petal.Length | Species, data = iris)
    }
})
sapply(res, class)
#> [1] "source"       "source"       "recordedplot"

[etiennebacher]

So what does tinytheme do exactly that would make the plot different ? Maybe plot change are not correctly detected. If you can give details on this, it would help undertand this better

I don't know enough about graphics in base R (I'm not the author of tinyplot) but I narrowed it down to the init_tpar() and tpar() calls in tinytheme(). If those two calls are commented out then the two plots are printed.

Also, the tinyplot() call isn't even needed in the reprex.

This gives two plots:

reprex::reprex({
  for (i in 1:2) {
      plot(i)
  }
})

This gives one plot:

reprex::reprex({
  tinyplot::tinytheme("basic")
  for (i in 1:2) {
      plot(i)
  }
})

[grantmcdermott]

Thanks for diving into this.

I don't know enough about graphics in base R (I'm not the author of tinyplot) but I narrowed it down to the init_tpar() and tpar() calls in tinytheme(). If those two calls are commented out then the two plots are printed.

Also, the tinyplot() call isn't even needed in the reprex.

Hmmm, very interesting. What tinytheme() does behind the scenes is set a hook before a new plot is drawn. See: https://github.com/grantmcdermott/tinyplot/blob/cfd5059e4e8e3213ed2425831a00fb26bb0c4c39/R/tinytheme.R#L156-L168

Is this perhaps interfering with a hook that knitr sets (and needs)?

[cderv]

@grantmcdermott - quick summary:

What I tried to look at in #237 (comment) is to reproduce only with evaluate to get the behavior right without anything interfering from knitr.

From some tries, it seems using tinytheme() does not end up in recording two plots from evaluate()

> res <- evaluate::evaluate(function(){ 
    library(tinyplot)
    for (thm in c("dark", "minimal")) {
        tinytheme(thm)
        tinyplot(I(Sepal.Length * 1e4) ~ Petal.Length | Species, data = iris)
    }
})
> sapply(res, class)
[1] "source"       "source"       "recordedplot"

I would have expected two plots. So I am trying to understand what happens here, and possibly make this more minimal based on what tinytheme() is doing

At first I though this was related to an old closed as won't fix issue

• plot handler called too late when evaluating a loop #67

But then I tried plot() with background change and I get two plots

> res <- evaluate::evaluate(function() {
    for (bg in c("red", "blue")) {
        par(bg = bg)
        plot(Sepal.Length ~ Petal.Length, data = iris)
    }
})
> sapply(res, class)
[1] "source"       "recordedplot" "recordedplot"

[hadley]

That looks like a bug in tinyplot to me — it is removing the plot hook that evaluate sets, which obviously breaks evaluate.

[grantmcdermott]

That looks like a bug in tinyplot to me — it is removing the plot hook that evaluate sets, which obviously breaks evaluate.

@vincentarelbundock is that what we do? Any pointers for how to combine these hooks, rather than removing would be much appreciated.

[hadley]

You could look at the source code for evaluate::set_hooks()/evaluate::remove_hooks() to see how to edit hooks safely/correctly,