revdep shapr new test failures due to warnings

[tdhock]

revdep shapr has the following new test failures if data.table from github master is installed:

* checking tests ...
  Running 'testthat.R'
 ERROR
Running the tests in 'tests/testthat.R' failed.
Last 13 lines of output:
  `shapr(train_df_used_numeric, l_numeric[[i]])` produced warnings.
  -- Failure ('test-a-shapley.R:87'): Testing data input to shapr in shapley.R ---
  `shapr(train_df_used_numeric, l_numeric[[i]])` produced warnings.
  -- Failure ('test-a-shapley.R:92'): Testing data input to shapr in shapley.R ---
  `shapr(train_df_used_factor, l_factor[[i]])` produced warnings.
  -- Failure ('test-a-shapley.R:92'): Testing data input to shapr in shapley.R ---
  `shapr(train_df_used_factor, l_factor[[i]])` produced warnings.
  -- Failure ('test-a-shapley.R:92'): Testing data input to shapr in shapley.R ---
  `shapr(train_df_used_factor, l_factor[[i]])` produced warnings.
  -- Failure ('test-a-shapley.R:92'): Testing data input to shapr in shapley.R ---
  `shapr(train_df_used_factor, l_factor[[i]])` produced warnings.
  
  [ FAIL 7 | WARN 45 | SKIP 0 | PASS 637 ]
  Error: Test failures
  Execution halted

git bisect says this started happening after merging #5233
@ben-schwen can you please have a look since you were the author of that PR?

[MichaelChirico]

Same warning as #5532, this time because of the nomatch= argument being passed to merge.data.table.

Here's the invocation in question:

https://github.com/NorskRegnesentral/shapr/blob/3f4f61e90056839baec90c301015bc730d7cd8bd/R/models.R#L467

Should we add nomatch= to the signature of merge.data.table?

(in this case, I'm not sure the nomatch= argument is doing anything because all=TRUE, we might just change that anyway, but I think nomatch= is in general useful -- in my mental model, merge.data.table() is a convenient wrapper around [, is that no longer the case?)

[jangorecki]

IMO should be kept baseR-ish, and to me a good reason to add nomatch there is to catch it and raise a warning. merge.data.table is suboptimal in the sense that it calls [ join. Coming mergelist will be good substitute for merge.data.table.

[tdhock]

I would also suggest to keep merge as simple as possible (do not add nomatch argument), instead fix this issue by asking revdep maintainer to use [ instead of merge if he needs to specify nomatch arg.

[tdhock]

this is now fixed on CRAN.