writeManifest() gives confusing error if no app.R

[hadley]

i.e. create a new project with file foo.R containing library(shiny), then run writeManfiest():

> rsconnect::writeManifest()
Error in `quartoInspect()`:
! Failed to run `quarto inspect` against your content:
ERROR: /Users/hadleywickham/Desktop/madlibs2/madlibs.R is not a valid
Quarto input document

Presumably because it's using the wrong primary document, but why is it passing an R file to quarto? Should also log about the primary file.

[aronatkins]

Background:

It is assumed that if you have an R script, nothing else indicates a different content type, and it is not a well-known filename (app.R, plumber.R), then it must be a standalone R script that you want to render with Quarto.

See inferAppMode():

rsconnect/R/appMetadata.R

Lines 192 to 199 in 9999fdb

	if (hasR) {
	# We have R scripts but it was not otherwise identified as Shiny or Plumber
	# and also not accompanied by *.qmd or *.rmd files.
	#
	# Assume that this is a rendered script, as this is a better fall-back than
	# "static".
	return("quarto-static")
	}

In general, we want Shiny applications in app.R files. Renaming of foo.R to app.R occurs very narrowly, mostly to support RStudio.

rsconnect/R/bundle.R

Lines 25 to 36 in 9999fdb

	if (!is.null(appMode) && appMode == "shiny") {
	# When deploying a single-file Shiny application and we have been provided
	# appPrimaryDoc (usually by RStudio), rename that file to `app.R` so it
	# will be discovered and run by shiny::runApp(getwd()).
	#
	# Note: We do not expect to see writeManifest(appPrimaryDoc="notapp.R").
	if (is.character(appPrimaryDoc) &&
	tolower(tools::file_ext(appPrimaryDoc)) == "r" &&
	file == appPrimaryDoc) {
	to <- file.path(bundleDir, "app.R")
	}
	}

That renaming workflow doesn't work when you're trying to produce a manifest.json, since the produced manifest needs to reference the original filename.

Connect uses shiny::runApp() to ask Shiny to run the application contained within the current directory. It does not supply a specific endpoint R file, but relies on the default Shiny behavior. When you try to use shiny::runApp() in a directory with foo.R, it complains:

R -s -e 'shiny::runApp()'
#> Loading required package: shiny
#> Error in `shinyAppDir()`:
#> ! App dir must contain either app.R or server.R.
#> Backtrace:
#>     ▆
#>  1. └─shiny::runApp()
#>  2.   ├─shiny::as.shiny.appobj(appDir)
#>  3.   └─shiny:::as.shiny.appobj.character(appDir)
#>  4.     └─shiny::shinyAppDir(x)
#>  5.       └─rlang::abort(...)

[aronatkins]

We don't log the inferred app-mode nor why we made that choice.

[thomasp85]

Slightly related - the logic captured in

rsconnect/R/appMetadata.R

Lines 36 to 38 in 2fe8fde

	if (!is.null(appPrimaryDoc) &&
	tolower(tools::file_ext(appPrimaryDoc)) == "r") {
	appMode <- "shiny"

caught me quite off-guard. I was trying to be explicit about what the primary R script was but doing so makes writeManifest insist that it is a shiny app