Memory issues on CRAN

[osorensen]

Seem to be caused by this in the case where newind = -2

galamm/src/update_funs.h

Lines 24 to 46 in 7cd848a

	template <typename T>
	void update_X(Mdual<T>& X, const Vdual<T>& lambda,
	const std::vector<std::vector<int>>& lambda_mapping_X){
	if(lambda_mapping_X.size() == 0) return;
	for(int i = 0; i < X.size(); i++){
	std::vector<int> newinds = lambda_mapping_X[i];
	T loading{0};
	bool update{false};
	int j{0};
	for(int newind : newinds){
	if(newind != -1){
	loading += lambda(newind);
	update = true;
	}
	j++;
	}
	if(update){
	*(X.data() + i) *= loading;
	}
	}
	};

I was able to reproduce a problem using R -d lldb. There may be more problems, of course. This change fixed it:

galamm/src/update_funs.h

Lines 24 to 53 in d4e2ab5

	template <typename T>
	void update_X(Mdual<T>& X, const Vdual<T>& lambda,
	const std::vector<std::vector<int>>& lambda_mapping_X){
	if(lambda_mapping_X.size() == 0) return;
	if(lambda_mapping_X.size() != X.size()) Rcpp::stop("Mismatch in lambda_mapping_X size.");
	for(size_t i{}; i < lambda_mapping_X.size(); i++){
	std::vector<int> newinds = lambda_mapping_X[i];
	T loading{0};
	bool update{false};
	int j{0};
	for(int newind : newinds){
	if(newind == -2){
	loading = 0;
	update = true;
	} else if(newind != -1){
	loading += lambda(newind);
	update = true;
	}
	j++;
	}
	if(update){
	*(X.data() + i) *= loading;
	}
	}
	};

[osorensen]

On rocker/r-devel-san, R CMD check now returns 1 NOTE, 0 WARNINGs, 0 ERRORs

* checking installed package size ... NOTE
  installed size is 35.1Mb
  sub-directories of 1Mb or more:
    doc    1.9Mb
    libs  32.0Mb

[osorensen]

I have also reproduced the CRAN issue on rocker/r-devel-san, using the current CRAN version. The same issue goes away when using the fix described above.

[osorensen]

Furthermore, on rocker/r-devel-san with the new update (the current CRAN version crashed it):

> devtools::test()
ℹ Testing galamm
✔ | F W  S  OK | Context
✔ |         12 | galamm-glmm [125.8s]                                                                                                         
✔ |         10 | galamm-heteroscedastic [28.0s]                                                                                               
✔ |          8 | galamm-latent-covariates-interaction [690.7s]                                                                                
✔ |         36 | galamm-lmm [505.8s]                                                                                                          
✔ |         13 | galamm-mixed-resp [708.5s]                                                                                                   
✔ |         68 | galamm-semiparametric [832.4s]                                                                                               
✔ |         40 | galamm-setup [485.7s]                                                                                                        

══ Results ═══════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════
Duration: 3377.1 s

[ FAIL 0 | WARN 0 | SKIP 0 | PASS 187 ]

devtools::run_examples() also succeeded.

[osorensen]

A note here. As an additional safety measure, I now explicitly make sure the program never asks for vector elements at negative indices:

galamm/src/update_funs.h

Lines 30 to 50 in 724f799

	for(size_t i{}; i < lambda_mapping_X.size(); i++){
	std::vector<int> newinds = lambda_mapping_X[i];
	T loading{0};
	bool update{false};
	int j{0};
	for(int newind : newinds){
	if(newind == -2){
	loading = 0;
	update = true;
	} else if(newind >= 0){
	loading += lambda(newind);
	update = true;
	}
	j++;
	}
	if(update){
	*(X.data() + i) *= loading;
	}

[osorensen]

Keeping the issue open until the CRAN checks are ok.

[osorensen]

CRAN checks ok