[R-package] R memory leaks

[david-cortes]

There are some situations in which the R version can leak memory by not destructing C++ objects that become unreachable:

• Functions that return an R external pointer are wrapped into R6 classes, with the finalizer for the external pointer being under the R6 finalizer. If an error occurs during initialization (when calling Class$new(...), like it is done when creating a dataset), the finalizer will not be called, thus leaking the C++ object (e.g. a dataset or a model). This could be solved by setting the finalizer in the R external pointer object itself.
• Some of the C++ functions allocate R vectors, such as these:
  

  LightGBM/R-package/src/lightgbm_R.cpp

  Line 203 in 8937022

  feature_names = PROTECT(Rf_allocVector(STRSXP, len));

  
  

  LightGBM/R-package/src/lightgbm_R.cpp

  Line 507 in 8937022

  eval_names = PROTECT(Rf_allocVector(STRSXP, len));

  
  

  LightGBM/R-package/src/lightgbm_R.cpp

  Line 683 in 8937022

  model_str = PROTECT(Rf_allocVector(STRSXP, 1));

  
  

  LightGBM/R-package/src/lightgbm_R.cpp

  Line 706 in 8937022

  model_str = PROTECT(Rf_allocVector(STRSXP, 1));

  
  If the allocation of the R vectors fail, it will trigger a long jump which would leave the C++ vectors that were allocated before undestructed and unreachable. This could be solved by using unwind protection as described in the R extensions manual.
• Same as per the other issues, calling Rf_error also triggers long jumps. Could either change it to something else while throwing the error on the R side, or add unwind protection for those too.

[jameslamb]

Thanks for writing this up!

unwind protection as described in the R extensions manual

Are you referring to https://cran.r-project.org/doc/manuals/r-release/R-exts.html#Condition-handling-and-cleanup-code? In the future, we'd very much appreciate if you included relevant links or supporting evidence when opening issues like this or leaving comments like #4273 (comment).

[david-cortes]

Yes, that's what I was referring to.

[jameslamb]

Ok perfect, thanks very much!

[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.