merge_headers, as some of the
headers that will be detected here are going to be multi-row and
therefore need collapsing.
split_headers(x, data_allow_merged = FALSE)
split_headers_find(x, use_frames = TRUE, data_allow_merged = FALSE)
split_headers_apply(x, n)TRUE, horizontal merged cells
are allowed in data regions, otherwise a region is considered
not to be a data region if it has any merged cells that extend
horizontally. Vertical merged cells, so long as they do not
span horizontally, are allowed.split_headers and split_headers_apply,
a worksheet view; in this view the data$headers element
will be a worksheet view of the headers. For
split_headers_find, a single integer representing the
number of rows of headers found (with zero indicating no
headers).
split_metadata, we need to employ some
heuristics to get this right. These will probably evolve as we
throw them at more spreadsheets.This seems pretty hard to generalise, but at the same time it's easy enough to tell when looking at a spreadsheet which bits are headers, but it seems hard to teach the package what these rules are.
Things that are going to be common include varying the background colour; to detect that we're going to need some colour distance metrics (plain colour shifts aren't enough because multi-colour headings are going to be common). Deciding what the relevant colour differences is tricky because that will depend on the sheet so I should look into perhaps something about edge detection here?
Another common way of marking out header rows will be boldness. That often will carry through to the row labels too though, but we'd be OK looking for rows where essentially everything is bold vs ones where a few things are bold.
If we're lucky enough to have split frames and we have a full sheet (not a view) then we can probably treat that as the point where the headers start.
Otherwise we can filter based on things like