rowwiseDT() could support setting key #7905

data.table() constructor accepts key= to determine the key() of the resulting table.

I think rowwiseDT() could do the same. Some other considerations:

• There's some back-compatibility risk if users are already creating tables with columns named key
  • I see exactly one place this happens on GitHub (0 on CRAN): https://github.com/search?q=lang%3Ar+%2FrowwiseDT%5C%28%2F+%2Fkey%5Cs*%3D%2F&type=code
• It's no longer possible to create a table with a column named key directly
  • We could in principle allow missing(key) to mean "column named key" and !missing(key) to be the key() setting
• It could be abused to create really confusing code:

rowwiseDT(
  a = , b = , key = 'a',
  2, 1,
  1, 2
)

rowwiseDT(
  a = , b = ,
  2, key = 'a', 1,
  1, 2
)

rowwiseDT(
  a = , key = 'a', b = ,
  2, 1,
  1, 2
)

I would suggest a linter to discourage that -- key= should either be the first or last argument:

rowwiseDT(key = 'a',
  a = , b = ,
  2, 1,
  1, 2
)

rowwiseDT(
  a = , b = ,
  2, 1,
  1, 2,
   key = 'a'
)