Of all arguments this one have no sense. Some LLMs are so “smart” that they are trying to use square brackets instead of do … end block in Elixir. The language is really well documented, so any tool that at least want to pretend it’s smart should generate a valid code depending on the latest stable version that matches mix.exs version requirement. Any kind of algorithm/tool that does not do that is not good enough to even mention it.
Also keep in mind that often many chatbots are learned based on very old data like few years ago which means few stable versions back and bringing new feature every few years is terrible idea. Some sites correctly warns about false positives when using chatbots which is nothing surprising having in mind that chatbots are prepared for conversation purposes and therefore does not need to be as strict as we expect from for example LSP.
Existing hex packages for now would just warn a deprecation warning. Nobody should complain about that. Community and especially business rarely depends on abandoned projects. Teams behind LSP implementations and editor plugins are doing amazing work already, so I consider it a problem at all.
Why we should postpone warning message that would not affect the code at all? Fixing it in every repository is trivial and there is no sense of forcing “warnings as errors” for dependency compilation and even if many projects are already updated or even didn’t had such usage.
