critical failure -- need CRAN release in a week or so

[dankelley]

Below is a new error on CRAN package check. The error appears on several test machines. (I thank BR for pointing this out.)

Version: 1.7-10 
Check: tests 
Result: ERROR 
     Running ‘testthat.R’ [36s/36s]
    Running the tests in ‘tests/testthat.R’ failed.
    Complete output:
     > library(testthat)
     Loading required namespace: cli
     > library(oce)
     Loading required package: gsw
     > 
     > test_check("oce")
     Loading required namespace: sf
     Loading required namespace: dplyr
     Loading required namespace: sp
     Loading required namespace: RSQLite
     OMP: Warning #96: Cannot form a team with 24 threads, using 2 instead.
     OMP: Hint Consider unsetting KMP_DEVICE_THREAD_LIMIT (KMP_ALL_THREADS), KMP_TEAMS_THREAD_LIMIT, and OMP_THREAD_LIMIT (if any are set).
     [ FAIL 1 | WARN 0 | SKIP 9 | PASS 2687 ]
     
     ══ Skipped tests ═══════════════════════════════════════════════════════════════
     • On CRAN (3)
     • empty test (6)
     
     ══ Failed tests ════════════════════════════════════════════════════════════════
     ── Failure ('test_argo.R:15:9'): plot works on indexed subsets ─────────────────
     `plot(argo[["profile", 1]], which = which)` produced messages.
     
     [ FAIL 1 | WARN 0 | SKIP 9 | PASS 2687 ]
     Error: Test failures
     Execution halted 
Flavor: [r-devel-linux-x86_64-fedora-clang](https://www.r-project.org/nosvn/R.check/r-devel-linux-x86_64-fedora-clang/oce-00check.html)

[dankelley]

Hm. I am not 100% sure what the error is. I see this "OMP" problem, which is not something I understand. Below that, I see a problem with a plot() call. Note that the call is as below: apparently it was not silent until issue #1603. I'll go back in the commits to see when that was, to see if the old CRAN was before that.

But first I'll try build-test the CRAN source.

[dankelley]

FYI devtools::check_mac_release() on "develop" has no problem. I'll now work on the CRAN source to see. It would be great if I could reproduce it quickly (and this remote test is faster than a build on my own machine).

[dankelley]

Given comments from a CRAN member, and after discussion with @richardsc, I have decided to comment-out all plot() calls in the test suite. I think users will tell us about these anyway. Perhaps, in doing this, the test code will be faster to run, reducing the package build time a little. (I still don't think it will be under the 10-minute limit...)

We have only 37 cases.

$git grep 'plot('|grep -v ': *#'|wc
      37     269    3127

[dankelley]

Actually, I am going to remove the lines, instead of commenting them out. That ought to be clearer (which is relevant given a comment from a CRAN member about obfuscation). We can rely on a good git commit to tell us about what has changed.

[dankelley]

After some discussion with CR, my present plan is to move the plot() tests from the CRAN test suite to a local test suite that is not packaged in the tarball. That way, things can be tested deeply by developers or anyone building from source, while still cleaning up the CRAN tests. I'll do this tomorrow.

[dankelley]

I've moved the tests involving plot() to files named tests/test_local*R, which are not included in the tarball and which therefore will not appear on CRAN, so I'll close this now. (I did the movement yesterday but forgot to close this.)