Tobias Schieferdecker

I agree

also maybe: - [ ] reset button (probably not necessary, since the same can be achieved by repeatedly pressing undo) - [ ] info box that informs if a cycle...

Sounds alright in general. Some caveats: - for bigger datasets, determining how many columns are needed as PK could take a while longer - for some tables we might not...

I see, that is indeed a common use case. There will of course be the naming rules that the datasets have to follow, but ensuring that could be an acceptable...

> Can we handle anything related to additional metadata with code that we generate? I don't understand. Are we talking about a different scenario here than a folder full of...

Sounds good to me. For the inverse operation of writing a dm to an Excel file or csv-files, we could then at least warn/inform the user, in case the reading...

already got a rough implementation when user provides ptype; since it's descoped for now, I will not file a PR for now though

To be used like: ``` dm_recycle(dm_disentangle(entangled_dm(), a), dm_ptype(entangled_dm())) ``` BTW: name is completely preliminary: comes from the fact that we're reintroducing cycles.

FWIW, `count()` seems to behave properly when using a `dm_zoomed`: ``` r suppressPackageStartupMessages({ library(dm) library(dplyr) }) dm_nycflights13() %>% dm_zoom_to(airlines) %>% count(carrier) %>% dm_update_zoomed() %>% dm_get_all_pks() #> # A tibble: 4...

Would you expect the columns used for `count()` to constitute a new PK? I would think that in your example no PK was actually dropped, but a potential one wasn't...