Preventing stale ggplots

[llrs]

Sometimes a target is run but when it is actually loaded it fails. I am not sure if this is only restricted to ggplot2 or related to non standard programming. Probably it is the former as most of the calculations happen when the print.gg method is invoked.

See this targets from the example plus a test target:

list(
  tar_target(
    name = data,
    command = tibble(x = rnorm(100), y = rnorm(100))
#   format = "feather" # efficient storage of large data frames # nolint
  ),
  tar_target(
    name = model,
    command = coefficients(lm(y ~ x, data = data))
  ),
  tar_target(
    name = test,
    command = ggplot(data) + geom_point(aes(x, y, col = a)),
    packages = "ggplot2"
  )
)

The test target uses ggplot2 to create a plot, but uses a column named a that doesn't exists in data.

tar_make(test)
## ✔ skip target data
## ✔ skip target test
## ✔ skip pipeline: 0.17 seconds
tar_load(test)
test
## Error in FUN(X[[i]], ...) : object 'a' not found

Traceback

traceback()
11: FUN(X[[i]], ...)
10: lapply(aesthetics[new_aesthetics], eval_tidy, data = data)
9: scales_add_defaults(plot$scales, data, aesthetics, plot$plot_env)
8: f(..., self = self)
7: l$compute_aesthetics(d, plot)
6: f(l = layers[[i]], d = data[[i]])
5: by_layer(function(l, d) l$compute_aesthetics(d, plot))
4: ggplot_build.ggplot(x)
3: ggplot_build(x)
2: print.ggplot(x)
1: (function (x, ...) 
   UseMethod("print"))(x)

Solution

The easiest solution is to print the ggplot2 object in the target:

list(
  tar_target(
    name = data,
    command = tibble(x = rnorm(100), y = rnorm(100))
#   format = "feather" # efficient storage of large data frames # nolint
  ),
  tar_target(
    name = model,
    command = coefficients(lm(y ~ x, data = data))
  ),
  tar_target(
    name = test,
    command = print(ggplot(data) + geom_point(aes(x, y, col = a))),
    packages = "ggplot2"
  )
)

Then tar_make(test) fails directly:

tar_make(test)
## ✔ skip target data
## • start target test
## ✖ error target test
## • end pipeline: 0.553 seconds
## Error : object 'a' not found
## ✖ Problem with the pipeline.
## Error:
## ! problem with the pipeline.

[wlandau]

Nice workaround. This is an edge case specific to how ggplot2 works, so targets isn't in a position to take specific action.

[llrs]

Yes, initially I intended to ask for a solution, but realize this might be unique to ggplot2 (or other visulization tools) due to how it works, so I thought it would worth to document it somewhere.