unload namespace before reinstalling, #4403

[jangorecki]

closes #4403

#R version 4.0.0 (2020-04-24) -- "Arbor Day"
#Copyright (C) 2020 The R Foundation for Statistical Computing
#Platform: x86_64-w64-mingw32/x64 (64-bit)
#...
data.table::update.dev.pkg()
#trying URL 'https://Rdatatable.gitlab.io/data.table/bin/windows/contrib/4.0/data.table_1.12.9.zip'
#Content type 'application/zip' length 2675878 bytes (2.6 MB)
#downloaded 2.6 MB
#
#package 'data.table' successfully unpacked and MD5 sums checked
#
#The downloaded binary packages are in
#        C:\Users\jan\AppData\Local\Temp\Rtmps5UOqq\downloaded_packages
#R data.table package has been updated to dd7609e83132e19c9be80d71c73e8a8f95e19b27 (1.12.9)

[codecov]

Codecov Report

Merging #4404 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #4404   +/-   ##
=======================================
  Coverage   99.60%   99.60%           
=======================================
  Files          72       72           
  Lines       13970    13970           
=======================================
  Hits        13915    13915           
  Misses         55       55           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 8a57236...8346e79. Read the comment docs.

[jangorecki]

codecov fail is false positive

[jangorecki]

We might not even need on.exit() but just unloadNamespace alone (see code chunk below), but better to leave it.
If PR would not always fix the issue, then we could add Sys.sleep(2) after unloadNamespace and before install.packages. More testing on windows R 4.0.0 would be appreciated.

R -q
unload = function() {unloadNamespace("pkg"); !"pkg"%in%loadedNamespaces()}
utils::package.skeleton("pkg", list="unload")
q("no")
rm pkg/man/unload.Rd
R CMD build pkg
R CMD INSTALL pkg_1.0.tar.gz 
R -q
library(pkg)
unload()
#[1] TRUE

[mattdowle]

Maybe wrap this with a tryCatch()? Just so that a more helpful error message can be produced instead; e.g. "Please close all R sessions to release the dll lock on Windows, and try again. If the dll lock is still open, try a reboot which should release the lock on Windows. Also, if you have a .Rpofile which loads data.table on startup, turn that off temporarily in order to upgrade the dll on Windows."

[jangorecki]

.Rprofile should not be an issue, because DT is already loaded at this point.

[jangorecki]

I would lean towards keeping on.exit chunk as simple as possible