Don't download Rtools

[krlmlr]

if there is no src/ directory, to save on build time and bandwidth.

[HenrikBengtsson]

I'm just trying upgrade to a modern version of r-appveyor and things seem to work out of the box. However, my package does not have a src/ directory but r-appveyor is still downloading and installing Rtools;

[...]
[1] stats     graphics  grDevices utils     datasets  base     
== 03/08/2016 21:52:59: Determining Rtools version 
== 03/08/2016 21:53:01: Downloading Rtools from: https://cran.rstudio.com/bin/windows/Rtools/Rtools33.exe 
== 03/08/2016 21:53:14: Running Rtools installer 
Rtools is now available on drive C:
== 03/08/2016 21:56:14: Setting PATH 
== 03/08/2016 21:56:14: Downloading and installing travis-tool.sh 

Could there be something wrong with https://github.com/krlmlr/r-appveyor/blob/master/scripts/appveyor-tool.ps1#L150-L155:

  if ( Test-Path "/**/src" ) {
    InstallRtools
  }
  Else {
    Progress "Skipping download of Rtools because src/ directory is missing."
  }

[PolMine]

Dear colleagues,
maybe making the installation of Rtools optional causes other appveyor problems. When checking my package cwbtools (github.com/PolMine/cwbtools), I get an error that g++ is missing:

* installing *source* package 'stringi' ...
** package 'stringi' successfully unpacked and MD5 sums checked
** libs
*** arch - i386
c:/Rtools/mingw_32/bin/g++  -std=gnu++11 -I"c:/R/include" -DNDEBUG -I. -Iicu61/ -Iicu61/unicode -Iicu61/common -Iicu61/i18n -DU_STATIC_IMPLEMENTATION -DU_COMMON_IMPLEMENTATION -DU_I18N_IMPLEMENTATION -DUCONFIG_USE_LOCAL -DU_TOOLUTIL_IMPLEMENTATION -DNDEBUG -DWINVER=0x0600 -D_WIN32_WINNT=0x0600 -DU_USE_STRTOD_L=0         -O2 -Wall  -mtune=generic -c stri_ICU_settings.cpp -o stri_ICU_settings.o
/bin/sh: c:/Rtools/mingw_32/bin/g++: No such file or directory
make: *** [stri_ICU_settings.o] Error 127
ERROR: compilation failed for package 'stringi'
* removing 'c:/RLibrary/stringi'
In R CMD INSTALL
Error in i.p(...) : 
  (converted from warning) installation of package 'stringi' had non-zero exit status
Calls: <Anonymous> ... with_rprofile_user -> with_envvar -> force -> force -> i.p
Execution halted
Command exited with code 1
7z a failure.zip *.Rcheck\*

My interpretation is that appveyor recognizes that Windows binaries at CRAN are much older than the latest Package source. This is what I see at CRAN (https://cran.r-project.org/web/packages/stringi/index.html).

Reference manual: | stringi.pdf
Package source: | stringi_1.2.4.tar.gz
Windows binaries: | r-devel: stringi_1.1.7.zip, r-release: stringi_1.1.7.zip, r-oldrel: stringi_1.1.7.zip
OS X binaries: | r-release: stringi_1.2.4.tgz, r-oldrel: stringi_1.2.4.tgz
Old sources: | stringi archive

So this is why Appveyor tries to compile stringi, which results in an error?

If this is the case, then making the installation of Rtools optional on the presence of a src directory of the package that is tested might frequently result in errors of the kind I see.

I am not an expert on this - I just wanted to share my interpretation.

Kind regards
Andreas (Blätte)

[krlmlr]

You can always force installation of Rtools, see README.

It's very strange that stringi binaries lag that much.

In the future, could you please open a new issue? Comments on an old issue are easy to miss.