Target factory to rerun after a given delay

[ropensci]

Target factory to rerun after a given delay

[wlandau]

I am open to name suggestions. My thoughts so far:

• tar_interval()
• tar_periodically()
• tar_age()
• tar_schedule()
• tar_delay()
• tar_timer()
• tar_frequently()

[noamross]

I like tar_interval(). Another possibility is tar_expire()

[petrbouchal]

tar_periodically() and tar_expire() seem intuitively closest to what is in the tin. I think it might be nice to stick to the tar_[noun] pattern for target-generating functions, but nothing appropriate comes to mind.

tar_after(), tar_timeout(), tar_every(), tar_timed() come to mind otherwise.

[wlandau]

Thanks for the suggestions, they definitely help. After thinking it over, I like tar_difftime(). Since I plan to have users pass a difftime object to mark the interval, I find this to be clear and specific.

[wlandau]

And actually, there are at least two conceivable versions of tar_difftime(): one to force a target to rerun when it gets old enough, another to suppress a target until it gets old enough. I think we just want the former.

[wlandau]

Let's first build some custom cues:

• tar_cue_force(condition): equivalent to tar_cue(mode = ifelse(any(condition), "always", "thorough"))
• tar_cue_skip(condition): equivalent to tar_cue(mode = ifelse(any(condition), "never", "thorough"))
• tar_cue_age(name, age): equivalent to tar_cue(mode = ifelse(any(Sys.time() - tar_timestamp_raw(name) > age), "always", "thorough"))

Then build tar_age() for this issue on top of tar_cue_age(). (I think I like that one more than tar_difftime().)

[petrbouchal]

Yes, my intuition initially was that I needed a special-purpose cue, so this makes good sense. For simplicity, age could be input as just an integer denoting the number of seconds - though not sure what implications that would have for the internals doing the time comparison.

[wlandau]

tar_age() and those cues I mentioned are now implemented.

For simplicity, age could be input as just an integer denoting the number of seconds - though not sure what implications that would have for the internals doing the time comparison.

I prefer difftime because we don't know what scale of time the user is looking for. If it's seconds and most people turn out to want days, most users will end up supplying a high number for seconds and doing repetitive mental arithmetic.

[wlandau]

Related: turns out tar_age() etc. can work with dynamic branching if we take the max timestamp (so all the branches rerun if the youngest branch's data becomes old enough).