Follow-up Comment #6, task #15653 (project administration): > I tried to come up with a criterion on what mistakes "slipped through fingers" are acceptable In software terminology, these things are known as bugs. Things the author did not intend to happen in the final product, but were missed due to the complexity of the whole/larger project and limited time.
As soon as someone finds/reports a bug, the author will fix it. This is the case here and has been demonstrated in previous messages on this thread. If I had not implemented your bug-reports, you had reason to complain. But demanding that any project be bug-free from day one is absurd. > The point of your package is bit-to-bit reproducing, isn't it? Not for the software (like GNU Guix), but for mostly the output data of the analysis. When built on different OSs or Kernel versions for example, the software will necessarily be different from the bit-wise perspective. But this does not necessarily mean that the output data are different (the software authors may/will have accounted for the difference in low-level architecture). The reason I put "mostly" is that for some scenarios in statistical analysis with high performance computing (parallel execution on many clusters), it is also impossible to have bit-wise identical data outputs. In such cases, users can implement statistical verification In fact, we recently published a paper on this project in IEEE's Computing in Science and Engineering, where we also discuss this point: https://doi.org/10.1109/MCSE.2021.3072860 (open-access), or you can see the version on arXiv: https://arxiv.org/abs/2006.03018 _______________________________________________________ Reply to this item at: <https://savannah.nongnu.org/task/?15653> _______________________________________________ Message sent via Savannah https://savannah.nongnu.org/