Beta testing results | Fabrizio Sandri

Introduction

Beta testing is the process of evaluating a software prototype right before its release in order to detect flaws. This procedure can be appropriately conducted by a third party who is using the program for the first time, the beta tester. It’s important to note that the most important part of this phase is observing the user’s interactions with the program without any input from the developers in order to identify the most critical sections that need to be handled.

Thanks to my mentor, I was able to take part in several live beta testing sessions where I was the observer and my mentor was the beta tester. The main goal of this session was to uncover weaknesses and difficulties in the technique for configuring RcppDeepState-action. As an observer, I couldn’t participate; all I could do was take notes on potential difficulties. This session showed me that no matter how much time I spent looking for bugs in the code I wrote, I would never find them in a short period of time; beta testing, on the other hand, allowed me to rapidly detect flaws.

A new strategy

Live sessions are extremely beneficial, but they are not the only strategy to beta test the code. Another possibility for speeding up the process on a large scale is to use automated techniques to test the code and compare the results to others that we already have access to. This is exactly what I talked about in my earlier blog post about beta testing RcppDeepState-action on GitHub-hosted CRAN packages¹.

To recap the previous blog posts, I wrote two script:

one that automatically initializes the RcppDeepState-action on a list of preset GitHub repositories that included problems according to Akhila’s article². This test were ran within the forked version of the repository to ensure that there were no permission issues with the GitHub tokens¹;
the other one that automatically replicates the RcppDeepState reports as a comment on the remote repositories³.

Using this method, I was able to identify a number of new issues with the RcppDeepState library. The most problematic one I discovered was that RcppDeepState would not build the shared object file in some circumstances due to missing dependencies in the tested packages. The incorrect usage of R CMD INSTALL inside the package to build the shared object file without first installing the missing package dependencies caused this issue. This problem was also reported to me by Dr.Martin R. Smith, author of numerous R packages accessible on CRAN, like Quartet and TreeSearch.

Furthermore, while examining the TreeSearch package, Dr. Smith identified another issue: one of its packages was not analyzed because some datatypes fell outside of the list of datatypes supported by RcppDeepState. He asked if this was expected behavior, and I said yes, as well as providing him with a detailed description of the error’s motivation, as well as a link to the RcppDeepState wiki article on how to add support for a new datatype. Dr. Smith then made a pull request⁴ to the RcppDeepState repository, asking for the IntegerMatrix datatype to be added to RcppDeepState. Because of this, I was able to determine whether or not the method required to add support for a new datatype is properly documented; based on the results, I can claim that Dr.Smith did an excellent job by following all of the wiki steps one by one.

Conclusion

Finally, I can say with certainty that beta testing really helped me in determining what the issues are with RcpDeepStare. Specifically, because to this new technique, I was able to collect feedback from beta testers, like in the case of Dr.Smith.