Get other.Rraw running again in r-devel

[MichaelChirico]

A few changes here:

• We need to mess with the attach order. Originally tried attach.required=FALSE but that's not present on R 3.5.0.
• {ggplot2} internals have moved around again. I think the intent of those tests is just to ensure print(<ggplot-obj>) works without error, so I refactored the tests to do that more directly.
• A legit bugfix for example() (described below).
• Robustness improvement for a {bit64} test when lacking proper long double. I'm of half a mind to just delete that test because it's only testing {bit64} functionality anyway...
• Since RFC: fail test.data.table() if non-test code produces warnings #7210, warning() outside of test() is caught, so change an intentional warning on Windows to cat().

Without the update to cedta() we get the telltale cedta() issues from here:

data.table/inst/tests/other.Rraw

Lines 227 to 229 in fae95de

	# example(":=", local=TRUE) triggered cedta==FALSE and then error, #2972
	test(14.1, {example(':=', package='data.table', local=TRUE, echo=FALSE); TRUE})
	test(14.2, {example('CJ', package='data.table', local=TRUE, echo=FALSE); TRUE})

Expected: 
Observed: [ was called on a data.table in an environment that is not data.table-aware (i.e. cedta()), but ':=' was used, implying the owner of this call really intended for data.table methods to be called. See vignette('datatable-importing') for details on properly importing data.table.
Test 14.2 produced 1 errors but expected 0
Expected: 
Observed: could not find function "J"

AIUI this is because of #7162 adding one more item on the call stack during execution of example().

Split off from #7832 as the actual debugging of package code should be done artisinally.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.01%. Comparing base (46d0934) to head (20963c5).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #7833   +/-   ##
=======================================
  Coverage   99.01%   99.01%           
=======================================
  Files          88       88           
  Lines       17234    17234           
=======================================
  Hits        17065    17065           
  Misses        169      169           

☔ 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.

[github-actions]

• HEAD=other-attach-order much faster for DT[by,verbose=TRUE] improved in #6296
  

Generated via commit 20963c5

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

Task	Duration
R setup and installing dependencies	2 minutes and 16 seconds
Installing different package versions	33 seconds
Running and plotting the test cases	5 minutes and 57 seconds

[ben-schwen]

Could we maybe add a comment above why we enforce this ordering? seems out of the blue without reading later comment

[MichaelChirico]

per the line below:

.gitlab-ci.yml uses parse(,n=1L) to read one expression from this file and installs pkgs.

(OTOH, that's not true, anymore at least)

[MichaelChirico]

It is used here though:

data.table/.github/workflows/R-CMD-check-occasional.yaml

Line 96 in 895067d

other_deps_expr = parse('inst/tests/other.Rraw', n=1L)

[ben-schwen]

Do we really need this here? Previously we/you have used the test(, {print(DT); TRUE}) idiom e.g. in test 2187

[MichaelChirico]

Yea I was just finding it a bit cluttered here given how big the first expression before ; TRUE is.

Another option is to add a new test() argument to signify "not testing the output, just checking for errors/output". Maybe test(ignore_value = TRUE), WDYT?

[ben-schwen]

Yeah ignore_value=FALSe seems like the right way to got, while I personally find positive checks easier (personal taste) like check_value=TRUE