dry-run option for clean()

[pat-s]

Prework

• Read and abide by drake's code of conduct.
• Search for duplicates among the existing issues, both open and closed.

Description

Would it makes sense to have a dry_run argument to clean() to be able to check which targets will be removed from the cache? Especially when using tidyselect helpers it is unclear which targets will be affected.

[wlandau]

Something like it, sure. make() has outdated() and recoverable(), and unmaking targets is high-stakes too. Trying to decide between clean(dry_run = TRUE) and cleanable()...

[wlandau]

Hmm... the word "cleanable" does not actually get at the idea, and dry_run would make the return value of clean() less consistent than I would like. Some other possible names:

• dry_clean()
• mock_clean()
• predict_clean()
• list_clean()
• peek_clean()

Any sound good to you? Got a better idea for a name?

[wlandau]

Let's go with which_clean().

[pat-s]

I am used to clean(dry_run = TRUE) from other tools but I am fine with any. 👍

I even thought if this might server as a default to prevent undesirable events of people deleting important things - but I vaguely remember there is another option to undo cache cleaning, isn't it?

[wlandau]

Yes, to undo clean(garbage_collection = FALSE), run make(recover = TRUE). It's not a perfect inverse transformation (it could recover additional targets and build others) but it does the job.

What other tools support clean(dry_run = TRUE)? It would be nice to be consistent with what other developers do, but I just feel like clean(dry_run = TRUE) would return a character vector and have no side effects whereas regular clean() has large side effects and an empty return value. I would rather be internally consistent than externally consistent. The alternative to returning a character vector seems to be printing to the console, and that does not seem ideal.

[pat-s]

What other tools support clean(dry_run = TRUE)?

Some shell and python tools AFAIR. Not sure if I've seen this in R before.

I would rather be internally consistent than externally consistent.

Totally understandable! I am totally fine with an extra function for it.