Previously compiled/fitted model suddenly crashes (without any changes in the specification)


I have fitted a certain model various times without any problems. After one month, I wanted to return to the analysis. However, when fitting the same model using the same code and the same setup (same R-version, same Stan version, everything without changes), I run into a compilation error: The output is massive (about 1000 lines of C+±code). It begins with:

make: *** [file784b9b28ef.o] Error 1

ERROR(s) during compilation: source code errors or compiler configuration errors!

Program source:
2: // includes from the plugin
5: // user includes
6: #define STAN__SERVICES__COMMAND_HPP// Code generated by Stan version 2.17.0
8: #include <stan/model/model_header.hpp>
10: namespace model784798bea_t0hier3_namespace {
12: using std::istream;
13: using std::string;
14: using std::stringstream;
15: using std::vector;
16: using stan::io::dump;
17: using stan::math::lgamma;
18: using stan::model::prob_grad;
19: using namespace stan::math;

and it ends with:
Error in compileCode(f, code, language = language, verbose = verbose) :
Compilation ERROR, function(s)/method(s) not created!

I would be very grateful for some help! (I use R version 3.5 with rstan version 2.17.3 and a windows 7 operating system). It seems very puzzling, because I did not change anything to the stan-setup, yet suddenly previously compiled/fitted models do not work anymore.

We need the part of the output that starts with error: (including the colon).

Thanks. The part of the output which includes “error:” is:

“error: ‘enable_both’ is not a member of ‘boost::exception_detail’
throw exception_detail::enable_both( e );”

What version of BH do you have?

Sorry I don’t know where I can access the version number?

In fact, I do not recall that I changed anything in the setup (including any C++ file, R version, stan version), yet it does not work anymore.


Thanks. It is version

And you have RStan 2.19.2? The only other person who encountered an error like that managed to overcome it by upgrading

Thank you very much for clarifying this. I’ll probably follow your suggestion. However, I guess in the case of upgrading, I won’t get the same results when I run the previously compiled model again - right?

You should not expect to get literally the same realizations from the posterior distribution, although the posterior distribution should be the same.