[R-package] RFC: require R >= 3.5.0

[jameslamb]

I'm opening this request for comment as part of the changes we're considering for LightGBM 3.0.0 (#3071 ).

I'd like to propose the following:

as of LightGBM 3.0.0, the R package cannot be used with R versions older than 3.5.0.

How this makes LightGBM better

Today, we maintain a couple places in the code for the R package which are if-else blocks checking whether you are on a version of R earlier than R 3.5.0

• LightGBM/CMakeLists.txt

  Line 14 in 1e2013a

  OPTION(USE_R35 "Set to ON if your R version is not earlier than 3.5" OFF)

• LightGBM/R-package/src/R_object_helper.h

  Lines 21 to 46 in 1e2013a

  	#ifdef R_VER_ABOVE_35
  	#define NAMED_BITS 16
  	struct lgbm_sxpinfo {
  	unsigned int type : 5;
  	unsigned int scalar : 1;
  	unsigned int obj : 1;
  	unsigned int alt : 1;
  	unsigned int gp : 16;
  	unsigned int mark : 1;
  	unsigned int debug : 1;
  	unsigned int trace : 1;
  	unsigned int spare : 1;
  	unsigned int gcgen : 1;
  	unsigned int gccls : 3;
  	unsigned int named : NAMED_BITS;
  	unsigned int extra : 32 - NAMED_BITS;
  	};
  	// 64bit pointer
  	#if INTPTR_MAX == INT64_MAX
  	typedef int64_t xlen_t;
  	#else
  	typedef int xlen_t;
  	#endif
  	#else

• LightGBM/R-package/src/install.libs.R

  Lines 173 to 175 in 1e2013a

  	if (R_ver >= 3.5) {
  	cmake_args <- c(cmake_args, "-DUSE_R35=ON")
  	}

Removing this support would simplify the R package's code, reducing the risk of mistakes and the maintenance burden. This simplification would also make implementing #3016 easier.

Why I think the impact to users will be minimal

R 3.5.0 was released more than 2 years ago, in April 2018 (version history).

In all of the R-package bug reports submitted by our users in the last 6 months, people have been using R version 3.5 or later

• prediction failure with integer data #3139: 3.6.3
• Error in cv_booster$record_evals[["valid"]][[first_metric]] #3112: 4.0.0
• [R-package] MAPE metric not being minimized in R #3099: 4.0.0
• [R-package] Installation with R4.0.0 on Windows is broken #3064: 3.6.3 -> 4.0.0
• [R-package] I never could make the library run any simple model. The session always exploded #2970: 3.6.3
• lgb.cv data.table error - R package #2715: 3.5.x

Note for reviewers

In addition to maintainers, I'd love comments from some of the people who have helped the R package by opening issues and pull requests: @mayer79 @rgranvil

[mayer79]

I remember there where some troubles when R 3.5 was being released.

I think in this case it is fully justified to require R >= 3.5, even if one or the other company might still rely on older R versions (with old versions of LightGBM, probably, so no real issue).

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.