covr causing failed TravisCI builds?

[billdenney]

In the past couple of days, I've had failed builds on TravisCI through what appears to be it waiting on covr() to return:

https://travis-ci.org/billdenney/pknca/jobs/431929216

The main package is here: https://github.com/billdenney/pknca

Is there a way to fire-and-forget the covr() call after a successful TravisCI build so that it becomes only the responsibility of codecov.io at that point?

[billdenney]

A subsequent build yielded this message during the codecov() run:

Error: Failure in `/tmp/RtmpXtetLQ/R_LIBS30cf80d11a9/PKNCA/PKNCA-tests/testthat.Rout.fail`
Execution halted
Warning message:
system call failed: Cannot allocate memory 

https://travis-ci.org/billdenney/pknca/jobs/431938785

[jimhester]

You will have to try and debug this locally first with package_coverage()

[billdenney]

package_coverage() locally (Windows 10) gives no issues:

> library(covr)
> foo <- package_coverage()
> print(foo)
PKNCA Coverage: 94.19%
R/class-general.R: 73.81%
R/class-PKNCAresults.R: 79.63%
R/tss.stepwise.linear.R: 87.72%
R/class-PKNCAconc.R: 88.24%
R/tss.R: 88.68%
R/class-PKNCAdata.R: 89.15%
R/tss.monoexponential.R: 89.47%
R/pk.calc.all.R: 90.88%
R/exclude_nca.R: 92.41%
R/class-PKNCAdose.R: 93.80%
R/check.intervals.R: 95.88%
R/interpolate.conc.R: 96.55%
R/001-add.interval.col.R: 96.92%
R/superposition.R: 97.48%
R/parseFormula.R: 97.70%
R/pk.calc.simple.R: 97.93%
R/PKNCA.options.R: 98.84%
R/general.functions.R: 99.12%
R/002-pk.business.rules.R: 100.00%
R/AIC.list.R: 100.00%
R/auc.R: 100.00%
R/aucint.R: 100.00%
R/choose.intervals.R: 100.00%
R/cleaners.R: 100.00%
R/exclude.R: 100.00%
R/half.life.R: 100.00%
R/merge.R: 100.00%
R/pk.calc.c0.R: 100.00%
R/pk.calc.dn.R: 100.00%
R/pk.calc.urine.R: 100.00%
R/provenance.R: 100.00%
> 

[jimhester]

From your error messages it seems to be caused by using too much memory, I am not sure there is a lot that can be done about it. You could try running package_coverage() on travis first, then pass the result to codecov to see if it is failing when calculating coverage (which is likely) or when uploading the reports to codecov.

Rscript -e 'cov <- covr::package_coverage()' -e 'print(cov)' -e 'covr::codecov(coverage = cov)'

[billdenney]

I'm still having trouble: https://travis-ci.org/billdenney/pknca/jobs/432495895

When running locally, package_coverage() returns in about a minute.

[wlandau]

So am I: https://travis-ci.org/ropensci/drake/builds/433793953. I am working on better reprexes: https://travis-ci.org/ropensci/drake/builds/434014128 (without covr) https://travis-ci.org/ropensci/drake/builds/434014394 (with covr).

[wlandau]

Whatever the issue may be, it seems like these builds started failing right after the latest CRAN release of covr. Could something have changed in how covr uses memory resources? Could there be conflicts between processes forked by covr and those forked by the tests?

[jimhester]

Please try with the current devel version of covr, it may be #335 which is now fixed.

[wlandau]

Works for drake, thanks!

[billdenney]

Works for PKNCA, too. Thanks!

[wlandau]

In that case, shall we close?

[billdenney]

I think yes.