devcontainer?

[grantmcdermott]

Recently, I'm having to do more of my dev work on a Mac because of work... which means that I can't run our test suite (which requires Linux) without a running container (Docker/finch/etc.)

I've got a simple .devcontainer/devcontainer.json file that I use locally for this, but I was thinking it might be useful to add it to the repo for others to use too?

{
  "name": "tinyplot devcontainer",
  "image": "rocker/r2u:24.04",
  
  "containerEnv": {
    "NOT_CRAN": "true"
  },
  
  "postCreateCommand": "R -e \"install.packages(c('devtools', 'tinytest', 'tinysnapshot'))\" && R -e \"install.packages('httpgd', repos = c('https://community.r-multiverse.org', 'https://cloud.r-project.org'))\" && Rscript -e 'read.dcf(\"DESCRIPTION\", c(\"Imports\", \"Suggests\")) |> tools:::.split_dependencies() |> names() |> setdiff(tools:::.get_standard_package_names()$base) |> install.packages()'",
  
  "customizations": {
    "vscode": {
      "extensions": [
        "REditorSupport.r",
        "quarto.quarto"
      ],
      "settings": {
        "r.plot.useHttpgd": true
      }
    }
  },
  
  "remoteUser": "root"
}

@vincentarelbundock Do you use WSL or something similar?

[vincentarelbundock]

My guess is that you are hitting this breaking change in tinysnapshot:

0.2.0

Breaking change:

Tests are automatically skipped by default unless the test suite is run interactively or the NOT_CRAN environment variable is set to true. This means that the skip argument in expect_snapshot_print() and expect_snapshot_plot() is no longer needed for skipping tests on CRAN.

I have a mac laptop and a linux desktop. On the mac, I do something like:

export NOT_CRAN=true
make testall

That gives me about 3 failures, with only a few pixel differences. Before pushing complicated things, I'll sometimes run it on linux before, but mostly I just recognize the three failures and count them as "likely pass".

[grantmcdermott]

On the mac, I do something like:

export NOT_CRAN=true
make testall

That gives me about 3 failures, with only a few pixel differences. Before pushing complicated things, I'll sometimes run it on linux before, but mostly I just recognize the three failures and count them as "likely pass".

Huh, I somehow didn't realise the tests would run on Mac 🤦

And, yes, I get the same minor three errors as you. I'll close this, but we can maybe revisit in the future if the need for a .devcontainer becomes clear.

[grantmcdermott]

Changed my mind and decided to add one. It's a cheap, single file addition that offers some nice functionality for remote testing.

[vincentarelbundock]

I've never used this but will try. Thanks!

[grantmcdermott]

My devcontainer.json file is a little more convoluted than the standard case because I have to use finch on my system (as opposed to Docker). But the end result is the same.

To use from VS Code:

• Install the Remote Development extensions pack.
• Open your command palette and search for "Dev Containers: Rebuild and reopen in container". (TBH, just typing "reopen" should be enough to bring it up.) The first time-build will take a minute or so, just let it finish. You can ignore the prompts to install the R extension during the build; it will be installed for you.
• Open up an R file per normal and run some code to trigger the R console.
• Install the current dev version of the package with devtools::load_all(). (We don't build it as part of the devcontainer, since we don't want an old version to be cached.)
• Test away!
• To revert back to a regular folder, open command palette and search for "Reopen folder locally"

You can also test it out directly on GitHub using Codespaces. (It's dead simple, see here for an equivalent example.) But I just ask that you do it on your own fork of tinyplot, since I'm liable for billing on this fork and don't have GitHub scholar credits to burn anymore ;-)