Reorganize 'Selecting variables' section #61
Conversation
|
@hadley: Would you like to review? I can't request a review (no permissions?) |
There was a problem hiding this comment.
I know this is only extending an existing section, but this extension now makes it clearer to me that we might be getting ahead of ourselves by writing this out in principles. I fully support the new tidyr UI, but it isn't released yet. It feels too early to me to draw general conclusions about vars() vs enquo(). I think we're still missing things about quoting vs regular eval, and about selections vs actions. In general, this discussion is missing one entire goal of the colwise functions when we designed them: they were meant to use standard eval, as they were a bridge between purrr and dplyr. Similarly facet_grid() and facet_wrap() are standard eval functions, so taking vars() was an obvious fit.
When we rewrite the colwise verbs with automatic splicing, one possibility is to have a mapping() function that enquo() a selection. That would be consistent with the new tidyr UI. Another possibility is to compose with pick() and tibble() to take selections and actions. Perhaps we'll have both, the latter being more advanced and requiring knowledge of purrr::map_df(). In any case, pick() will be a new way of interacting with the data mask (creating a tibble based on dynamically scoped variables).
I guess my point is that it might be too early to write about this in the principles book because we're still figuring things out. Should we have a place where we can develop working documents instead? Ideally these would not be fully fleshed out so that they retain a "still in progress" feel conducive to exploration.
| ggplot2::facet_grid(rows = drv, cols = c(vs, am)) | ||
| ``` | ||
|
|
||
| The `vars()` function is a simple wrapper around `quos()`, and adds no value compared to using `c()` directly, which requires quoting by the called function. |
There was a problem hiding this comment.
I wouldn't say it adds no value. It can be useful to take selections in standard eval functions.
Perhaps the motivation becomes more apparent when explained by the obsolescence of
vars()and then extending this to functions that use dots for selection.