Argument to handle "unspecified" fields

[mgirlich]

Given an object to tibblify and a specification for this there might be some fields that are not specified. Now, this could be because:

• you don't care about these additional fields
• an optional field has changed the name or there are new fields
• some fields simply vary

The specification should help to clarify what's the case (a bit similar to readr::cols() vs. readr::cols_only()). Therefore, the spec should get these arguments:

• unspecified_fields = c("inform", "warn", "error", "ignore", "collect")
  • "ignore" implies that you know there are (or might be) other fields but you deliberately ignore them
  • "error" implies that you don't expect other fields, e.g. useful when getting data from an API and storing it
  • "inform"/"warn" not sure if both are needed
• unspecified_fields_to is a string specifying the column name where unspecified fields should end up in (similar to tib_multi() How to handle fields whose specification cannot be guessed? #38).
• known_fields to specify the keys of other fields you know exist but you decide to ignore

Things to consider

• a known field is always optional
• a field can be in known_fields and have a specification
• allow known_fields to be a list so that they can be used for nested keys as well

[mattwarkentin]

Is there no way currently to silence the message emitted by tibblify::tibblify() when there are unspecified columns other than wrapping it with suppressMessages()?

[jonthegeek]

Much of this already exists via the unspecified argument in tibblify(), but we silently ignore fields that are present in x and completely absent in a provided spec. I'm keeping this ticket open to clarify that difference.

[jonthegeek]

To clarify:

We do all of our unspecified processing without ever looking at x (other than to potentially use it to guess the spec if spec isn't explicitly provided). We should respect unspecified for "leftovers" in x that aren't in spec at all. I can see two ways to deal with this:

1. guess_tspec(x) even when spec is provided to detect fields that are in our guessed spec but not in the given spec. We'd probably need to suppress the inform_unspecified. This feels wasteful.
2. Do something "inside" ffi_tibblify to "count" such fields, and use unspecified to determine what to do about them if there are any. This might end up related to something like Replace .mark_empty_list_argument() option with environment #293; we'd put this information in an environment that we consult before returning.