bump R version

[StrikerRUS]

Refer to https://github.com/RGF-team/rgf/pull/300#issuecomment-520357097.

[fukatani]

Coverage too small is direct reason why appveyor failed. But I don't have no idea why coverage is small...

[fukatani]

Removing

  - if [[ $TASK == "R_PACKAGE" ]] && [[ $TRAVIS_OS_NAME == "linux" ]]; then

branch worked well.

[fukatani]

@mlampros Do you have any idea?

[StrikerRUS]

@fukatani

Coverage too small is direct reason why appveyor failed. But I don't have no idea why coverage is small...

Yeah, now R installation process is OK. Coverage is small means that R-package fails and doesn't work, only stub file zzz.R works.

Speaking to setuptools error, it seems to me that it's conda activation issue. I'll take a look.

[StrikerRUS]

@fukatani setuptools issue is fixed by pinning Python to 3.7.3. Possibly related: https://github.com/ContinuumIO/anaconda-issues/issues/11195.

[StrikerRUS]

I want to test appveyor with previous version of reticulate package, with which one we had no errors at master 2 months ago. When I write

Rscript -e "if(!'reticulate' %in% rownames(installed.packages())) { devtools::install_version('reticulate', version = '1.12', dependencies = TRUE) }"

it still forces downloading of new version:

The downloaded binary packages are in
    C:\Users\appveyor\AppData\Local\Temp\1\RtmpojOLQ5\downloaded_packages
reticulate (1.12 -> 1.13) [CRAN]
Installing 1 packages: reticulate
trying URL 'https://cran.rstudio.com/bin/windows/contrib/3.6/reticulate_1.13.zip'
Content type 'application/zip' length 1638901 bytes (1.6 MB)
==================================================
downloaded 1.6 MB
package 'reticulate' successfully unpacked and MD5 sums checked
The downloaded binary packages are in
    C:\Users\appveyor\AppData\Local\Temp\1\RtmpsNLSdF\downloaded_packages

Does anyone know how to install 1.12 version?

UPD: Got it work by removing dependencies = TRUE. However, it didn't help. 😢

[StrikerRUS]

Unpinning Python version, since upstream issue has been fixed.

[mlampros]

I'm sorry to both of you I just saw the issue,

• an older version of an R package can be installed using either the 'devtools' package or from source as explained here. I think that using devtools is simpler,

require(devtools)
install_version("reticulate", version = "1.12.0", repos = "https://cran.rstudio.com")

• the low coverage indicates that RGF or / and FastRGF can't be detected from the R-package during testing (the tests are skipped because the python modules are not detectable). The 'zzz.R' file has a percentage greater than 0 because it does prints only a message to the console (independently of the python package)

It probably has to do with the installation file. I can't give an advice for the appveyor ci because as you already know there is an open issue with a 'help-wanted' badge for the windows installation.

[StrikerRUS]

Nothing had been changed from our side since the last successful CI run, so I guess we need to investigate which package update leaded to current fail.

[mlampros]

@StrikerRUS,

I might be in place to make the execution of the tests and loading of the R package more verbose, by that I mean to modify the package.R file and the skip_test_if_no_module by adding if-else statements, so that messages are printed to the console if either Python (3.7 in this case) or any of the loaded Python modules is not detectable. That way we will be at least in place to understand if it has to do with the Python version or with the RGF modules. But may I ask you something, why there aren't any artifacts for the appveyor build. For instance for one of my R packages that is tested with appveyor the artifacts can show me if tests run successfully. Especially the tests_i386\testthat.Rout and tests_x64\testthat.Rout files which show printed messages during testing of the R package.

[StrikerRUS]

@mlampros I think both of your ideas are good! Feel free to contribute!

[mlampros]

@StrikerRUS,

I think that I understood what the problem of the failed build is. The default Miniconda installation is,

 default {$env:MINICONDA = "C:\Miniconda37"}

however the R-package is tested with Miniconda36

      PYTHON_VERSION: 3.6
      TASK: R_PACKAGE

May I ask why the R package is tested on Python 3.6 when the default Python version is 3.7? This is a quite sensitive case for the reticulate package and there are numerous issues related with the default version when using Python from within R. Probably here the

reticulate::py_discover_config()

could give us the answer to the issue.

[StrikerRUS]

@mlampros

May I ask why the R package is tested on Python 3.6 when the default Python version is 3.7?

I remember it was done due to issues with early Python 3.7 version on Windows.

I think we can update the version now.

The default Miniconda installation is, ...

It's default only for us. There is no any Miniconda paths in PATH variable initially. So I don't think it's the real root cause.

I want to understand what changed since our last successful run 2 months ago. I'm going to use RDP connection to Appveyor to investigate this when have more free time.

[StrikerRUS]

Closing in favor of #305.