feat: Add support for running Shiny for R apps

[gadenbuie]

Closes #29

• Clones shiny.python.runApp into a new shiny.r.runApp command.
• Does not include app debugging with shiny.r.debugApp (yet)
• Does not support codespaces
• Adds Shiny for R devmode and port options
• Renames the configuration section "Shiny"
• Calls shiny::devmode() before running the app when shiny.r.devmode is true or options(shiny.autoreload = TRUE) when false (to maintain the reload feature).

Todo

• Should Shiny for R and Shiny for Python apps use the same Shiny terminal?
  • Yes, since they'd use the same Simple browser window as well.
• Auto-reload port - is this required in R?
  • I don't think so. The extension either calls shiny::devmode() or options(shiny.autoreload = TRUE) before running the app and both appear to be working as expected.
• Separate setting for port for Shiny for R apps?
  • Yes, otherwise it's confusing that the Shiny > Python > Port option is for both.
  • I looked into changing the setting to shiny.port but I couldn't find a good mechanism for migrating users from shiny.python.port to shiny.port.
• Add setting for enabling devmode in R, defaults to true
• Call Rscript on Mac/Linux or Rscript.exe if on Windows
• How to get the path to Rscript.exe on Windows?

[jcheng5]

In package.json we take a dependency on the ms-python.python package. Maybe we should make that optional and testing for its existence when actually running a Shiny for Python app, as it would be a shame if R users got that extension dragged along just by installing the extension.

[jcheng5]

${path} needs escaping here. Not only for edge cases but on Windows the path will contain backslashes (I assume).

[gadenbuie]

IIUC escapeCommandForTerminal() handles this, It's pretty aggressive and has worked in my testing. In fact, the R code in runApp isn't inside quotes because it's handled correctly by escapeCommandForTerminal() (on my system at least, I may be missing something about how that function works).

[gadenbuie]

There are actually two issues happening here:

1. I've just used Rscript as the command which likely won't work on Windows, where we might need the full path to the Rscript binary, e.g. "C:\Program Files\R\R-3.6.0\bin\Rscript.exe".

2. Typically, I'd use Rscript -e 'shiny::runApp("app.R")' in my terminal, but when I used this formulation escapeCommandForTerminal() broke the command by escaping " within the '' quotes.

Given the above, I've refactored the rRunApp() command to use a packaged runShinyApp.R script that takes commandline arguments for path, port and --devmode.

That means that the escaping now works as expected, but leaves us with the problem of figuring out the full path to Rscript.exe on Windows. Do you have any ideas about how to approach that problem?

[gadenbuie]

This might not be as bad as I was thinking. I paired with @andrie this morning, who uses rig to install R on Windows, and in Powershell 7 and cmd.exe Rscript can be called directly. I'm pretty sure this is because rig adds .bat files to create an alias to run Rscript.

[jcheng5]

Did you refactor to use runShinyApp.R only because of the escapeCommandForTerminal() behavior? We can definitely fix that.

[jcheng5]

I don't think we can count on rig, this is my first time even hearing of it. You can see how the R extension does it here (and RStudio does something similar):

https://github.com/REditorSupport/vscode-R/blob/d582b9184838a32352f624e451d4c1856f347b9c/src/util.ts#L98
https://github.com/REditorSupport/vscode-R/blob/d582b9184838a32352f624e451d4c1856f347b9c/src/util.ts#L63

[gadenbuie]

@jcheng5 I've implemented this using the vscode-R extension as inspiration. I also tested locally on a Windows VM

[gadenbuie]

Did you refactor to use runShinyApp.R only because of the escapeCommandForTerminal() behavior? We can definitely fix that.

It might have started out this way, but we also run shiny::devmode() or set options(shiny.autoreload) before running the app. With that and the port-setting logic, I found it a lot easier to use a separate R script than to try to build the R code in strings.

I did add some logic to echo the shiny::runApp() command so users could copy and paste the command into an R console.

[gadenbuie]

In package.json we take a dependency on the ms-python.python package. Maybe we should make that optional and testing for its existence when actually running a Shiny for Python app, as it would be a shame if R users got that extension dragged along just by installing the extension.

Are these two entries in package.json jointly declaring the dependency on the Python Extension?

  "extensionDependencies": [
    "ms-python.python"
  ],
  "dependencies": {
    "@vscode/python-extension": "^1.0.5"
  }

I generally agree with you, but I don't think we need to do that work yet. I think it'd be okay for us to require the Python extension for now (it's a very popular extension and I'd imagine the number of R users who are using VSCode but not that extension is currently rather small).

[jcheng5]

extensionDependencies is what makes VS Code install the Python extension. dependencies is bringing an npm package into our own project--in this case, it's JavaScript stubs for calling into the Python extension.

[jcheng5]

I implemented making the Python extension a soft dependency, but I don't feel that strongly about it either way.