bind_rows should have col_type argument

[hadley]

So you could prespecify the types of the columns. For performance and safety.

[krlmlr]

I'm not sure, we can already do:

tibble(
  x = character(), y = integer(), z = list()
) %>%
  bind_rows(a, b)

to presprecify column types. Do we need to implement a safer coercion logic?

[hadley]

I think so - this is about providing a common interface across read_csv(), convert_types(), map_c() etc.

[RichardJActon]

As I just recently mentioned in #1485 I think it would be useful to have an option in bind_rows that instead of coercing factors to character vectors lets you specify that you want a new factor with levels from both of the original dataframes/tbls so you know don't have to manually make new factors. However, letting you specify types would likely also solve this, depending on how it handled factors with different levels in the source dfs.

Although I suppose you would probably want warning if you are altering factor levels as a result of a bind, so maybe an explicit relevel-factors=TRUE / refactor=TRUE like argument should be required to suppress a warning about releveling factors.

Perhaps a short-form like inherit-col-types that lets you specify that you want the same datatypes as the original dfs (assuming all the columns have matching types) would be handy?

[mkoohafkan]

Could this a different approach to tibble's as_tibble_row? If we could specify certain column types as "list" in bind_rows it would probably override the need for as_tibble_row for most use cases.

[romainfrancois]

@hadley is this vctrs::vec_rbind ?

[hadley]

Yup

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/