[R-package] [ci] added echoing of R CMD check logs to CI

[jameslamb]

No that we use Rscript build_r.R --skip--install, the only build of the R package is from R CMD check. R CMD check suppresses output from the build and writes it to a file. Those logs are useful for us to check for compiler warnings and debug issues.

See https://github.com/microsoft/LightGBM/pull/2936/files#r411607175

[StrikerRUS]

@jameslamb Thanks!
Seems that it is really usefull!
There are a lot of warnings like this one produced by gcc-9 compiler (macOS job):

/Users/travis/build/microsoft/LightGBM/lightgbm.Rcheck/00_pkg_src/lightgbm/src/src/lightgbm_R.cpp:664:52: warning: cast between incompatible function types from 'LGBM_SER* (*)(LGBM_SE, LGBM_SE, LGBM_SE)' {aka 'LGBM_SER* (*)(LGBM_SER*, LGBM_SER*, LGBM_SER*)'} to 'DL_FUNC' {aka 'void* (*)()'} [-Wcast-function-type]
  664 |   {"LGBM_GetLastError_R"              , (DL_FUNC) &LGBM_GetLastError_R              , 3},
      |                                                    ^~~~~~~~~~~~~~~~~~~

[jameslamb]

@jameslamb Thanks!
Seems that it is really usefull!
There are a lot of warnings like this one produced by gcc-9 compiler (macOS job):

/Users/travis/build/microsoft/LightGBM/lightgbm.Rcheck/00_pkg_src/lightgbm/src/src/lightgbm_R.cpp:664:52: warning: cast between incompatible function types from 'LGBM_SER* (*)(LGBM_SE, LGBM_SE, LGBM_SE)' {aka 'LGBM_SER* (*)(LGBM_SER*, LGBM_SER*, LGBM_SER*)'} to 'DL_FUNC' {aka 'void* (*)()'} [-Wcast-function-type]
  664 |   {"LGBM_GetLastError_R"              , (DL_FUNC) &LGBM_GetLastError_R              , 3},
      |                                                    ^~~~~~~~~~~~~~~~~~~

Thanks! Yeah I believe these can safely be ignored right now, but I also think I know how to fix them eventually.

These are all coming from our call to R_RegisterRoutines (https://github.com/microsoft/LightGBM/blob/master/src/lightgbm_R.cpp#L709). I believe I set that up correctly per CRAN's "Writing R Extensions", but could be wrong. Recall that you'll get an R CMD check warning if you don't call R_RegisterRoutines (#1910). My guess is that we're getting these compiler warnings because we're using our custom R-to-C interface instead of SEXP from R.h.

It would be good to eventually just use the R headers instead of our own, but I don't think that it will be a blocker to getting to CRAN, so I haven't prioritized it yet. I'll create an issue and put it in #2302

[jameslamb]

Added #3016 to capture the work to suppress that compiler warning from #3009 (review)