tibble() is dropping inner names

[jennybc]

Noticed because an incidental update of my local tibble caused pretty catastrophic test failure in tidyr, whose release / revdeps I'm working on. I'm at current HEAD of tibble here.

devtools::load_all("~/rrr/tibble")
#> Loading tibble

df <- tibble(x = 1:2, y = list(a = 1, b = 1:2))
df$y
#> [[1]]
#> [1] 1
#> 
#> [[2]]
#> [1] 1 2
names(df$y)
#> NULL

^{Created on 2019-08-07 by the reprex package (v0.3.0.9000)}

[jennybc]

This is the guilty commit: defe1ee

[jennybc]

From the just-previous commit 96cbc7a, I see:

devtools::load_all("~/rrr/tibble")
#> Loading tibble

df <- tibble(x = 1:2, y = list(a = 1, b = 1:2))
df$y
#> $a
#> [1] 1
#> 
#> $b
#> [1] 1 2
names(df$y)
#> [1] "a" "b"

^{Created on 2019-08-07 by the reprex package (v0.3.0.9000)}

[krlmlr]

Thanks. I submitted a PR to fix the tests in tidyr.

Are there legitimate use cases for inner names apart from the (now deprecated) use of .id in unnest()?

This is what I see:

# Names are dropped for atomic vectors
df <- data.frame(a = c(x = 1, y = 2), b = c(z = 3, w = 4))
df$a
#> [1] 1 2
df$b
#> [1] 3 4

# Names are retained for lists
df <- data.frame(a = I(list(x = 1, y = 2)), b = I(list(z = 3, w = 4)))
df$a
#> $x
#> [1] 1
#> 
#> $y
#> [1] 2
df$b
#> $z
#> [1] 3
#> 
#> $w
#> [1] 4

library(tidyverse)

# Names are dropped after dplyr operations
df %>%
  slice(1) %>%
  pull()
#> [[1]]
#> [1] 3

^{Created on 2019-08-08 by the reprex package (v0.3.0)}

It seems that inner names aren't well supported in the tidyverse anyway (nor in base). I'm in favor of removing them early on.

[hadley]

I don't understand why we are stripping names from all columns. I think you should remove

x[] <- map(x, strip_names)

all together.

[krlmlr]

I don't understand why we should allow names in columns. Are there legitimate use cases?

We never allowed names for columns of atomic types. Now we also drop names from bare lists, via vctrs:::set_names2() (which needs a better solution).

[krlmlr]

• We'll keep supporting names for list columns, because removing this is likely to cause downstream pain

• It's not tibble's responsibility to police user input

• Why do we strip names from atomic vectors? <-- review history

[krlmlr]

tibble 1.0 keeps internal names also for atomic types, 1.1 doesn't. Need to bisect.

[krlmlr]

09a07b5 is the first commit that strips internal names from atomic types. It looks like the names were stripped accidentally.

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.