Posted for discussion, not for approval. The idea that the way to maintain critical old code is to feed it through an AI which is supposed to convert it to a newer language without any loss of functionality, and expecting it subsequently to be readable and maintainable, strikes me as literally insane.
https://www.pcmag.com/articles/ibms-plan-to-update-cobol-with-watson
@medigoth Can’t agree more. A lot of old code (especially COBOL) is filled with poor implementations of — what are now — common things: associative mappings. It’s not clear, and each case is unique. (I know. I’m old. I wrote that stuff.)
The bulk of an app is a unique optimization to avoid excessive file reads. You can’t find enough examples to train an AI on this feature. You can’t even define the feature.
And. Some examples have bugs.