Data.table: Fail on win-builder for R 3.5.3

[mattdowle]

Passes OK on R 3.6.1

But fails with this on R 3.5.3 :
```
Running test id 1658.07 Test 1658.07 produced 1 warnings but expected 0
Expected:
Observed: partial argument match of 'along' to 'along.with'

Running test id 1658.08
Running test id 1658.09 Test 1658.09 produced 1 warnings but expected 0
Expected:
Observed: partial argument match of 'along' to 'along.with'
```

[MichaelChirico]

doesn't seem to be any signature change on seq.default:

https://github.com/wch/r-source/blob/tags/R-3-5-3/src/library/base/R/seq.R#L23

https://github.com/wch/r-source/blob/trunk/src/library/base/R/seq.R#L23

[MichaelChirico]

also 1658.07 is using only base functionality:

test(1658.07, write.table(DT, sep=";", qmethod="double", row.names=FALSE), output=ans)

anyway I'm reproducing the error on rocker/r-ver:3.5.3:

docker pull rocker/r-ver:3.5.3
docker run -it rocker/r-ver:3.5.3
system('apt-get update && apt-get install zlib1g')
install.packages('data.table', type = 'source', repos = 'http://Rdatatable.github.io/data.table')
library(data.table)
test.data.table()

[MichaelChirico]

The error is coming from base::order:

https://github.com/wch/r-source/blob/tags/R-3-5-3/src/library/base/R/sort.R#L189

I also see it in 3-5-0, 3-5-1, 3-5-2.

Current trunk (as well as 3-6-0 and 3-6-1) uses seq_along instead:

https://github.com/wch/r-source/blob/trunk/src/library/base/R/sort.R#L187

Safe to ignore?

[mattdowle]

Awesome. Thanks for getting to the bottom of this.
Can't ignore because it'll pop up as a fail if and when users run test.data.table() in those R versions. How about turning off warnPartialMatchArgs just for those two tests just in versions of R before that change in R? i.e. a minimum relaxation to pass.

[MichaelChirico]

It's not just those two tests -- there were also 50+ warnings from actually running in console. But the fix is as simple as testing order(1:10) for a warning & then we can just shut off that bit (with a message).

It would be nice to have a bugzilla item to refer to, perhaps @mmaechler could weigh in?

[mmaechler]

IIUC, this is a bug that has been fixed in R 3.6.0. We do detect and fix bugs not so rarely ourselves and fix them without the hassle of creating a bug report and then closing it etc. Did look in our NEWS list to see if the change / bug fix was mentioned there?

[MichaelChirico]

Not seeing anything in NEWS about seq or order so must have been an internal fix. Just goes to show how few people are using warnPartialMatchArgs in the wild that it survived a full minor R version on such a core function 😅

[mattdowle]

@MichaelChirico I don't think you should have flagged @mmaechler on this. We need to save his time for when we really need it.

there were also 50+ warnings from actually running in console

Can you add more color on this please? I don't see those and they aren't shown on win-builder output. Just 2 tests failed. If there are many more warnings than 2, why didn't more than 2 tests fail? test() fails if a warning occurs which is not expected. Did the 50+ warnings occur outside all test() calls (in preparing data for each tests) and only within test() in 2 cases?

[MichaelChirico]

yes that's my take on what happened. ran tests & got those errors and was prompted to run warnings() to see more. my experience w test.data.table() is that happens for issues on code outside test() calls.

have run the tests again on the PR branch and confirmed no more issues anyway so didn't dive any deeper.