[rikivillalba]

closes #7905

• key specification can occur in front or last position, or between headers and data blocks, but not within them, to prevent confusing code.
• missing "key" is a column name, for backwards compatibility.
• key = "" means missing key (again bc, as A = "" is interpreted as a column named A).
  So

rowwiseDT(
  a=, b=,
  key = "a",
  1, 2,
  3, 4
)

is valid but rowwiseDT(a=, key="a", b=, ...) isn't

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.08%. Comparing base (7b7f63b) to head (d8864d9).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #7940   +/-   ##
=======================================
  Coverage   99.08%   99.08%           
=======================================
  Files          88       88           
  Lines       17420    17432   +12     
=======================================
+ Hits        17261    17273   +12     
  Misses        159      159           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[MichaelChirico]

why length(header_pos)? do we lose a lot by making this an error as well?

(it can be a bit hard to map the test case to how it might look in "real" code, so I might be missing something, although this was something mentioned by me as 'bad' in #7905)

test(2282.11, rowwiseDT(A=, B=, key="B", 1, 2), DT)

[rikivillalba]

It is because I imagined data.table(a=.., b=.., key = "a") plus a data block, but I agree the extra flexibility is not very worthy here. I'll remove the middle option.

[tdhock]

Is there a strong reason why you can’t just use setkey(rowwiseDT(…), colname) ?
(if no strong reason, i would suggest closing this issue/pr without adding the new arg, which could be confusing… after all the whole point of rowwiseDT is readable code?)

[MichaelChirico]

The same can be said of data.table(); to me it's more awkward to have one data.table constructor that can set key but not the other.

I.e., I think this issue is about consistency with data.table().

[github-actions]

• HEAD=rowwiseDT_allow_key much slower for transform improved in #5493

(Plot image upload to picrd.com failed; download the artifact below to view plots)

Generated via commit d8864d9

Download link for the artifact containing the test results: ↓ atime-results.zip

Task	Duration
R setup and installing dependencies	1 minutes and 20 seconds
Installing different package versions	12 minutes and 36 seconds
Running and plotting the test cases	5 minutes and 33 seconds