The docs are a bit messed up at the moment:

pattern is not deprecated for data_addprefix() (and suffix), and the docs for select (for data_addprefix() is actually for the pattern argument. The select argument in data_addprefix() uses the normal select-helper functionality. We either have to split data_addprefix() and data_addsuffix() from data_rename() to have two different help-files, or, if we really use the classical select in data_rename(), we don't have to change much in the docs. Just pattern only refers to data_addprefix() and that's it.

I think it makes sense to split docs between data_rename() in one page and data_addprefix() and data_addsuffix() in another. We can use @seealso to redirect to the other.

Changes look good to me, I think it's just that tests need to be updated?

Yes I'm just waiting to see if we keep using column indices as valid replacements or not.

Yes I'm just waiting to see if we keep using column indices as valid replacements or not.

I'd say we can remove this. It's easy to do colnames(x) <- seq_len(ncol(x)). We also don't use this type of replacement in any easystats-packages, if I haven't missed something.

@etiennebacher I think I updated all packages in easystats that are affected by the new data_rename() changes. One thing I thought of was if we could/should keep the safe argument? Based on its value, we could set ifnotfound either to "warn" or "error", which should restore the previous behaviour. WDYT?

Right, now that we have ifnotfound we can do that, I'll open a PR

Actually it's not that easy because even if .select_nse() only warns, it doesn't include the unknown column name in the output. So for instance, if we do data_rename(iris, "FakeCol", "length") then .select_nse() returns character(0) and therefore we still error with There are more names in 'replacement' than in 'select'.

Hm, true. Or, if safe = TRUE, we just skip this check as well? Meaning that safe will "intersect" all valid columns and their replacement values, and use that for rename. Any values in replacement without a match in select will be ignored. For safe = FALSE, we set ifnotfound = "error", and include the check for unequal lengths.

Or, if safe = TRUE, we just skip this check as well? Meaning that safe will "intersect" all valid columns and their replacement values, and use that for rename.

But we don't know the position of the wrong column names since it's not returned by .select_nse(). For instance if we do this:

data_rename(iris, c("Sepal.Length", "foo"), c("rep1", "rep2"))

then .select_nse() returns "Sepal.Length", but how do we know which replacement we should use? We can't compare to the original select since it could be regex() for instance. I don't see a good way to enable this. Is the safe argument really important?

Ok, agree. Then we don't need to change anything and deprecate the safe argument, as we now do.