Should snapshot_accept() succeed from within tests/?

[MichaelChirico]

I sometimes run tests like

setwd('tests')
source('testthat.R')

However, this conflicts with using snapshot tests; after the output updated, I was instructed to run snapshot_accept(...), but this failed:

snapshot_accept(...)
No snapshots to update

However, it did create the temp (.new) file, and succeeded like so:

setwd('..')
snapshot_accept(...)
Updating snapshots:
* ...

[hadley]

This feels like a sufficiently unusual workflow that we're unlikely to work on it, but I'd review a PR for it.

[MichaelChirico]

ran into this again 😅

[MichaelChirico]

Now noticed snapshot_accept() has a path argument that I could change to testthat from the default tests/testthat.

I think what's most confusing here is the misleading message:

* Run `testthat::snapshot_accept('$DIR')` to accept the change.
* Run `testthat::snapshot_review('$DIR')` to interactively review the change.

Would adding "Run snapshot*() from the package directory to ..." be too distracting for the default use case?

[hadley]

Yeah, I think so.

Are you changing the path with code? (e.g. with withr::local_dir())?

[MichaelChirico]

no. workflow is

• start R in package dir
• setwd("tests") (is that considered "with code"?)
• source("testthat.R")

I find that's the easiest way to run basically as R CMD check does without needing to memorize anything like whether it's test_dir()/test_check()/check_package()/test_package().

[hadley]

Ah yeah, in that case I'd just case you have an unusual workflow and you're on your own.

You could make life a little easier on yourself by running source("tests/testhat.R", chdir = TRUE)