[R] Errors in R package installation

[David Stevens]

Ivan - thanks for looking into this. My answers to your comments are below.

Cheers

David

On 12/13/2019 1:46 AM, Ivan Krylov wrote:

On Thu, 12 Dec 2019 15:48:13 +0000
David Stevens <david.stevens using usu.edu><mailto:david.stevens using usu.edu> wrote:



Certain R packages will not install properly on my Windows 10
computer.



Certain, but not all? Which packages are you able to install on that
computer?


Too many to name - one is nlstools. Another is tinytex.  I updated many package after bookdown wouldn't install for the same reasons. Out of a list of ~20 packages to install, ~5 failed with a similar error message.  That sent me down the rabbit hole that brought me here.




Warning: invalid package 'C:\Users\David'
Warning: invalid package
'Stevens\AppData\Local\Temp\Rtmpk5NqrI/downloaded_packages/callr_3.4.0.tar.gz'



This looks like a command line argument quoting issue.

I wonder why doesn't install.packages use shQuote to quote the file
name when calling system2(c("R", "CMD", "INSTALL", file)) to install a
downloaded source package:

https://github.com/wch/r-source/blob/e554f7f12b22868bdae51aadaeea4d56c9f87a32/src/library/utils/R/packages2.R#L833

Maybe there should be shQuote(fil) instead, especially since R does
quote the file name when installing from local source tarballs:

https://github.com/wch/r-source/blob/e554f7f12b22868bdae51aadaeea4d56c9f87a32/src/library/utils/R/packages2.R#L676



This happens on only a subset of packages I install or update.



I have a hypothesis: perhaps this only happens for packages with no
pre-built Windows binary available yet, since "win.binary" packages may
be installed by unpacking a zip file, without calling a command line
with potential space quoting issues:

I read a thread elsewhere that said a work around is to run options(pkgType='binary') before installation and the problem went away. Does this help.



https://github.com/wch/r-source/blob/e554f7f12b22868bdae51aadaeea4d56c9f87a32/src/library/utils/R/packages2.R#L491
->
https://github.com/wch/r-source/blob/e554f7f12b22868bdae51aadaeea4d56c9f87a32/src/library/utils/R/windows/install.packages.R#L263

Calling install.packages(..., verbose = TRUE) for both a succeeding and
a failing package might help to verify whether this is the case.



This is a relatively recent issue (i.e. I never saw it before
November - I've used R for ~15 years).



Just to confirm it: you had no problems on the same Windows 10 computer
with the same user name and %USERPROFILE% path as before?

Yes, this is the case. I do the regular Windows 10 updates and update R and RStudio as soon as I am aware there's a new version out. I haven't explicitly change %USERPROFILE%.

Apparently,
tempdir() used to return a 8.3 directory path on your computer, but now
doesn't - but that should not happen, since R_reInitTempDir()
explicitly asks for a 8.3 path:

tempdir() gives

tempdir()
[1] "C:\\Users\\David Stevens\\AppData\\Local\\Temp\\RtmpQpqh0t"


>

https://github.com/wch/r-source/blob/e554f7f12b22868bdae51aadaeea4d56c9f87a32/src/main/sysutils.c#L1810

Microsoft Docs page for GetShortPathName() says:



If you call GetShortPathName on a path that doesn't have any short
names on-disk, the call will succeed, but will return the long-name
path instead. This outcome is also possible with NTFS volumes
because there's no guarantee that a short name will exist for a
given long name.



Some newer Windows 10 installations may have
NtfsDisable8dot3NameCreation enabled, thus preventing R from getting a
8.3 path to the temp directory.

I am taking the liberty of Cc-ing R-devel because this might warrant
adding a shQuote() call to install.packages().


Thanks for looking into this.




--
David K Stevens, PhD, PE
Environmental Engineering Division
Civil and Environmental Engineering
Utah State University
8200 Old Main Hill
Logan, UT 83200-8200
(435) 797-3229
david.stevens using usu.edu<mailto:david.stevens using usu.edu>

	[[alternative HTML version deleted]]