The camcorder package helps. Check this out from @rfortherestofus.com :

rfortherestofus.com/2024/09/prev...

So it’s not my pipeline that is burly, it is my expertise 💪
bsky.app/profile/morg...

This is {targets} week for me bc I need to manage a burly pipeline.

Monday: started small & able to make it work

Wednesday: realized there are so many support functions it makes more sense to write a package for them 🤦🏽‍♀️

Thursday: easily used my 📦; leveled up with {tarchetypes} to iterate steps

Oh thanks, yeah tar_older is new to me, that's handy!

tar_invalidate + tar_older is working for me! 👌

actually it does work how I expected, just not within tar_assign, so I'll go back a bit

really appreciate your feedback and example!!

Not sure tbh — the last pipeline I wrote (github.com/360-info/cli...) mostly used file targets!

I might be overthinking, I think a regular task that invalidates the target is what I want, it's a separate process

so, I think I understand - you must use this check and tar_invalidate() *outside of _targets.R*, so it becomes part of a routine script that runs (not encapsulated within a tar_make() step) ? I think :)

I can't see how to use it, because the initial record can't invalidate itself, so never updates ??

Make n high enough and you’ll get the famed ‘stack overflow’. I think its about 10000. That’s why you don’t see it in the wild.

However there is s technique to avoid the overflow, see Tailcall().

I was amazed to find this just works in C and Python too, I had thought it was somehow restricted to "lispy" langs