rpy2 crashes upon import with R 3.0.2

Issue #150 resolved
Luca Beltrame created an issue

As described on the ML, rpy2 crashes upon import.

The initial suspicion is that the following reported change in R-3.0.2 is linked to the issue:

 \item \code{eval} and \code{applyClosure} are now protected
       against package code supplying an invalid \code{rho}.

For now reverting to R 3.0.1 is the only possible option.

Comments (6)

  1. Laurent Gautier

    The change reported in the R changelog for 3.0.2 is indeed triggering the issue. That's revision 63421 in R's SVN, file src/main/eval.c :

     63421     ripley 
     63421     ripley     if (!rho)
     63421     ripley       error("'rho' cannot be C NULL: detected in C-level eval");
     63421     ripley     if (!isEnvironment(rho)) 
     63421     ripley       error("'rho' must be an environment not %s: detected in C-level eval", 
     63421     ripley             type2char(TYPEOF(rho)));
     63421     ripley 

    R is trying to bubble up the error and print the message while Python (thinks it) is in charge. It does not end up well.

    Now this latest change is R is uncovering a rather intriguing behavior with R. In rpy2 there is the following code (newPySexpObject() in _rinterface.c):

      if (TYPEOF(sexp) == PROMSXP) {
        PROTECT(env_R = PRENV(sexp));
        PROTECT(sexp_ok = eval(sexp, env_R));
        printf("  evaluating promise %p into %p.\n", sexp, sexp_ok);

    It is checking whether an R object in a "promise", and if so evaluates (possible performance hit, but there are already enough hoops to jump through to interface with R as it is - promises are evaluated when the object must be exposed to Python).

    This call to R's eval() is creating a crash... when called a second time (example below).

    import rpy2.rinterface as ri
    # this is fine, the promise is evaluated
    # this is no longer fine (?!), `env_R` is no longer an environment


  2. Laurent Gautier

    Could people try if 2059:e9b8d9b473fe is fixing the issue for everyone (this is in the branch version_2.4.x) ?

    If it does, I can backport the fix to the 2.3.x series and release quickly an rpy2-2.3.8.

  3. Log in to comment