Segmentation fault in CVAdataStore

[lcontento]

For some parameter values I get a segmentation fault while doing adjoint sensitivity analysis.
The problematic point is line 2033 in the following snippet:

sundials/src/cvodes/cvodea.c

Lines 2027 to 2037 in 0c83e0b

	i = 1;
	do {
	flag = CVode(cv_mem, ck_mem->ck_t1, ca_mem->ca_ytmp, &t, CV_ONE_STEP);
	if (flag < 0) return(CV_FWD_FAIL);
	dt_mem[i]->t = t;
	ca_mem->ca_IMstore(cv_mem, dt_mem[i]);
	i++;
	} while ( sign*(ck_mem->ck_t1 - t) > ZERO );

It seems that the index i is not checked for being inside the bounds of the dt_mem array.
If I add the check if (i > ca_mem->ca_nsteps) return(CV_FWD_FAIL); at the beginning of the loop body the segmentation fault is solved.
However, I am not sure if this is the correct thing to do or if index i should always be inside the bounds and the actual problem lies elsewhere.

[lcontento]

I investigated the matter a bit more. It seems that the software package I use (https://github.com/AMICI-dev/AMICI) is calling functions not in the public API. So maybe when only the public API is used the index i would be guaranteed to be inside the valid bounds (not sure).

[aseyboldt]

This might be the same issue as #31?