[R-package] changes to support read/write performance in CMake builds

[jameslamb]

Opening this PR to aid in the discussion on microsoft#3405, about whether bringing dependencies fmt and fast_double_parser into LightGBM will cause issues for getting {lightgbm} (LightGBM's R package) onto CRAN.

With the changes below, it's possible to build a {lightgbm} that passes almost all tests.

CMake-based installation

git submodule --init --recursive
Rscript build_r.R
R CMD check lightgbm_*.tar.gz --as-cran

CRAN installation with configure created by Autoconf

git submodule --init --recursive
sh build-cran-package.sh
R CMD check lightgbm_*.tar.gz --as-cran

R CMD check results

The CMake-based installation did not have any issues on my Mac.

The CRAN installation has this NOTE that would prevent the R package from getting to CRAN:

* checking compiled code ... NOTE
File ‘lightgbm/libs/lightgbm.so’:
  Found ‘___stderrp’, possibly from ‘stderr’ (C)
    Object: ‘external_libs/fmt/format.o’
  Found ‘___stdoutp’, possibly from ‘stdout’ (C)
    Object: ‘external_libs/fmt/format.o’

Compiled code should not call entry points which might terminate R nor
write to stdout/stderr instead of to the console, nor use Fortran I/O
nor system RNGs.

For a lot of background on this issue, see microsoft#2901.

Notes for reviewers

Let's talk about this on microsoft#3405 instead of here, to avoid splitting the conversation.

[AlbertoEAF]

Had to rebase+squash manually. Merged to my MR through a fast-forward. Don't know why it didn't detect the rebase&squash&merge.

[AlbertoEAF]

@jameslamb given that you're adding more changes to master in other PR's related to the same issue, now I'm not sure I was supposed to have merged this to my MR, can you confirm?

Thank you

[jameslamb]

given that you're adding more changes to master in other PRs related to the same issue

@AlbertoEAF what does this refer to? Can you give me an example?

now I'm not sure I was supposed to have merged this to my MR, can you confirm?

It's fine that you merged it, we can move all of the discussion back to microsoft#3405 now.

[AlbertoEAF]

Great ;)

@AlbertoEAF what does this refer to? Can you give me an example?

I was talking of the other PR's you have mentioned in the main PR microsoft#3405:

• [R-package] [ci] Make Windows CI fail if R CMD check fails microsoft/LightGBM#3435
• [R-package] add 'cleanup' script to handle left-behind Makevars microsoft/LightGBM#3434

[jameslamb]

Oh I see. I wouldn't call those "related to the same issue" and they don't conflict with the ongoing work in microsoft#3405