update test for dplyr 1.0.8

[romainfrancois]

We're about to release dplyr 1.0.8 (tidyverse/dplyr#6080) which would cause this failure in one test:

> checking installed package size ... NOTE
    installed size is 18.3Mb
    sub-directories of 1Mb or more:
      libs  16.5Mb

── Test failures ────────────────────────────────────────────────────────────────────────────── testthat ────

> library(testthat)
> library(bigsnpr)
Loading required package: bigstatsr
> 
> for (k in 1:9)
+   test_check("bigsnpr", filter = paste0(k, '-'))
══ Skipped tests ═══════════════════════════════════════════════════════════════
• is_cran is TRUE (1)

══ Failed tests ════════════════════════════════════════════════════════════════
── Failure (test-1-readBGEN.R:66:3): same variant infos as with QCTOOL ─────────
dplyr::mutate(test$map[-19, 1:6], chromosome = as.integer(chromosome)) not identical to dplyr::as_tibble(...).
Names: 3 string mismatches
Component 1: Modes: numeric, character
Component 1: target is numeric, current is character
Component 2: Modes: character, numeric
Component 2: target is character, current is numeric
Component 3: 198 string mismatches

[ FAIL 1 | WARN 0 | SKIP 1 | PASS 226 ]
Error: Test failures
Execution halted

This is because of this fix: tidyverse/dplyr#6035

[romainfrancois]

This pull request fixes the test.

[privefl]

Hi Romain.
Could you quickly explain what is the problem?
It seems it is not fixed yet. Maybe I can try to fix it myself.

[romainfrancois]

The test fails because dplyr has fixed this issue: tidyverse/dplyr#5967

[DavisVaughan]

A better fix might be to replace the transmute() in the next line with a rename() followed by a select() that selects the columns in the order seen here

[privefl]

I am sorry I still don't get what the error is here..
chromosome is the first variable, and should remain the first variable after the mutate(), which has always been the case, and it seems it is designed to remain like this.

[DavisVaughan]

Take transmute(df, chromosome, marker.ID = alternate_ids, rsid) as the example, which is identical to transmute(df, chromosome = chromosome, marker.ID = alternate_ids, rsid = rsid).

By accident, that was previously putting the columns in the order they are specified, ignoring whether or not any of them already existed, so you'd get a column order of (chromosome, marker.ID, rsid).

This is inconsistent with how mutate() works, where the invariants are:

• Modified columns that already exist are overwritten in place (their order can't change)
• Completely new columns are added to the end

So the correct resulting order should have been (chromosome, rsid, marker.ID) (assuming that in the original data frame, chromosome was before rsid), and that is now the case.

[privefl]

Wow.. Have you asked the community about this?
In transmute(), where you basically only keep the columns you are interested in (and forget about all the others), you kind of forget about the order as well and just want the order that you specify (at least this is what I want).
Please keep it that way. This would break so much code..

[romainfrancois]

In this case, there's no need for transmute() and it can be replaced by select() which can rename as well. , I've updated the pull request.

[privefl]

Do you still plan to change the behaviour of transmute() in the next release of {dplyr}?

[eitsupi]

@privefl Hi, as explained in tidyverse/dplyr#6086, the behavior seems to have been modified to the original behavior (the behavior of the current CRAN release) again in tidyverse/dplyr#6087 to make it consistent with the past behavior of transmute().

[privefl]

Great! Then I can close this.

[privefl]

I am trying to reproduce the issue with the GitHub version of {dplyr}.
However, when I try to install it with remotes::install_github("tidyverse/dplyr"),
I get the following error: could not find function "on_load".

[romainfrancois]

@privefl this is probably because you need to update rlang as well, e.g. pak::pak("r-lib/rlang")