Thanks all for the help. The reward (as usual) is another dumb
question.
If I add up the elapsed times listed in
https://win-builder.r-project.org/incoming_pretest/lme4_1.1-26_20201122_184744/Windows/examples_and_tests/examples_x64/lme4-Ex.timings
I get 21.43 seconds. But the output in
https://win-builder.r-project.org/incoming_pretest/lme4_1.1-26_20201122_184744/Windows/00check.log
says
** running examples for arch 'x64' ... [41s] OK
The values for i386 have a similar discrepancy (20.6 vs 36s). I can
appreciate that would be a bit of overhead (and rounding error), but ...
? (there are a total of 61 examples)
I wouldn't normally worry about the discrepancy, but I'm doing my
best to shave seconds where I can ... can anyone see any obvious thinkos?
I might as well add (in my partial defense, in response to Uwe's
point about not needing real-world/big examples for tests); many of
lme4's tests are regression tests of edge cases that came up in the real
world with moderately sized data sets. *In principle* it might be
possible to find a way to reduce those medium-sized problems to small,
fast problems that still demonstrated the same numerical problems, but
it's hard and time-consuming (at least for me) ...
cheers
Ben Bolker
On 11/22/20 4:06 PM, Uwe Ligges wrote:
Thanks Dirk. Yes, for lme4 the tests for each archiecture take longer
than 5 min, so the overall check time exceeds 10 min.
So one can follow Dirk's advise.
As a general remark for others who will read this in the future:
tests should test the software, but it is generally not important to
have real world examples. Small data and few iterations are typically
sufficient for tests.
It is also possible to run less important tests only conditionally if
some environment variable is set that you only define on your machine.
Best,
Uwe Ligges
On 22.11.2020 20:36, Dirk Eddelbuettel wrote:
On 22 November 2020 at 13:44, Ben Bolker wrote:
| My current guess is that the problem is with the too-long check
time on Windows (NOTE: "Overall checktime 18 min > 10 min")
Yes.
| I guess I have to get busy setting more tests and examples to
skip-on-CRAN (kind of a pain as there's no low-hanging fruit - other
than the 'testthat' tests, none of the individual test files take
longer than 15sec, although this is doubled because they have to be
run on 386 and x64 ...)
It's under your control. You can detect 'are we on Windows' and branch
or, as
I do with test runner I use, exit_file("...") based on such conditions.
| An alternative is that this is a confusingly worded message
indicating that there are strong rev dependencies so the package needs
to be further checked? (That seems unlikely as it explicitly asks me
to resubmit)
No. If there were any (even false positive ones) they'd be listed there.
Hth, Dirk
______________________________________________
R-package-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel
