Adding roundtrip for nested processing

[lorenzwalthert]

Might be implemented similar to depreciated flat roundtrip

#' Verify parse data modifications
#'
#' @description Check whether serializing the parse data results in the same
#' number of lines as the initial data that should be styled.
#' @param pd_flat A parse table.
#' @param text A character vector with the initial text to compare against.
#' @return If the verification is successful, `pd` is returned, with empty
#'   lines at the end of `text` stripped. \cr
#'   Otherwise, an error is thrown.
#' @seealso [serialize_parse_data_flat()]
verify_roundtrip <- function(pd_flat, text) {
  roundtrip <- serialize_parse_data_flat(pd_flat)

  if (length(roundtrip) < length(text)) {
    stopifnot(text[-seq_along(roundtrip)] == "")
    text <- text[seq_along(roundtrip)]
  }

  stopifnot(identical(text, roundtrip))
  text
}

[lorenzwalthert]

Reference: #85.

[lorenzwalthert]

Related: #216.

[lorenzwalthert]

I wonder how that should be implemented. Because with reference to #85, I think it would make sense to not just test parsing and serialising (because that almost certainly works now), but also whether the code is still valid after tokens were inserted, compare #259. So a simple parse statement of the output could check that.

[krlmlr]

I think there are multiple layers of roundtrip testing:

• At the text level: parse + nest + serialize should give exactly the same output (minus tabs to spaces conversion)
• At the code level: styling with any scope below "token" should lead to identical results (language objects or call trees) after parsing
• At the semantic level: styling with scope "token" should not change semantics of the code

The first two can be tested automatically, for the third we need tests with good coverage and rely on the user. We still could test that the resulting code is parseable, but this doesn't buy us much.

[lorenzwalthert]

Can we label this with "Status: Postponed" and close it for now?

[krlmlr]

Sure.

[lorenzwalthert]

Reference: #280.

[krlmlr]

Have we implemented roundtrip checking by now?

[lorenzwalthert]

In #287, we implemented verifying whenever possible (can_verify_roundtrip), which seems to be the case whenever we don't have any token manipulators.

[lorenzwalthert]

Reference: #368.