as_tibble inconsistencies

[hadley]

library(tibble)

# rows vs columns
as_tibble(c(x = 1, y = 2))
#> # A tibble: 2 x 1
#>   value
#> * <dbl>
#> 1     1
#> 2     2
as_tibble(list(x = 1, y = 2))
#> # A tibble: 1 x 2
#>       x     y
#>   <dbl> <dbl>
#> 1     1     2

# fills in names
as_tibble(c(1, 2))
#> # A tibble: 2 x 1
#>   value
#>   <dbl>
#> 1     1
#> 2     2
as_tibble(list(1, 2))
#> Error: Columns 1, 2 must be named.

# different scheme for missing names
as_tibble(t(1:2))
#> # A tibble: 1 x 2
#>      V1    V2
#>   <int> <int>
#> 1     1     2

I'd recommend that as_tibble() always treats inputs as columns as if you want rows, you can use enframe(). Related to #205

[hadley]

Alternatively, we could soft-deprecate as_tibble() and enframe() in favour of something with symmetric names, e.g. as_tibble_row(), as_tibble_col()

[hadley]

cc @jennybc

[krlmlr]

Current as_tibble.default() forwards to as.data.frame(stringsAsFactors = FALSE), to handle inputs we didn't anticipate. It's also the primary conversion function for data frames -> tibbles, I'd rather avoid renaming it.

We could add a special check for vectors/matrices and e.g. show a message that points to a method that won't print that message. For vectors where we remove the names we could use enframe(name = NULL) (#449). Auto-naming could be made consistent with repair_names() .

[hadley]

I think the invariant of as_tibble() is that ncol(tibble(x)) == length(x) (assuming the object doesn't have dimensions)

[hadley]

For names, we need to apply the same treatment to #445.

[krlmlr]

Do we need to distinguish between plain and classed atomic vectors (e.g. from hms)?

[krlmlr]

In master now, let's discuss improvements in new issues.

[krlmlr]

Suggesting to deprecate as_tibble.list() (in favor of new_tibble()) and not add new behaviors.

[hadley]

Both should be soft-deprecations

[krlmlr]

Does new_tibble() need a .name_repair or name_repair argument? We have tons of tests that test as_tibble(list(...)) .