"Richard O'Keefe" <rao...@gmail.com> writes: > Thus I hypothesise that there is room for Smalltalk as a tool for > *generating* and configuring HPC code.
Yes. But it will be hard to convince people that Smalltalk is a better choice than Python (well established in HPC as you say) for this use case. > My main worry is that when it comes to "bet your whole economy" > software and worse still, "bet the entire global economy and human > happiness for centuries" software like climate models, it's > appropriate to use the very highest quality assurance tools practical, > including *serious* verification. That's not common practice. Indeed, and that's one of my main worries in computational science as well. There is a culture of scientific validation, but not of technical verification of artifacts such as code. My own project in this space (implemented in Pharo) aims at supporting *human* verification for the aspects that no automatic tool can possibly verify. In technical terms, that's verifying the formal specification rather than the implementation of some method. The first step, which I have made good progress on, is a specification language for scientific models and methods that human readers would be happy to proofread. Details: https://github.com/khinsen/leibniz-pharo/ https://science-in-the-digital-era.khinsen.net/#Leibniz Konrad
