GitHub Copilot Programming Languages — assessment

• Copilot runs a type-checker internally before suggesting.
• The type system rejects a whole class of plausible-looking wrong suggestions before they ever run.
• Copilot was trained mainly on statically typed languages.
• Static languages have simpler syntax for a model to produce.

• It refuses to answer without a schema.
• Syntactically valid SQL referencing columns or tables that do not exist in your schema.
• A query that silently deletes data.
• A syntax error the database rejects immediately.

Select all that apply.

• Keeping the relevant type definitions, interfaces or schemas open or referenced.
• Writing a clear function signature and docstring before asking for the body.
• Asking for the whole file at once rather than a function at a time.
• Using longer variable names so the model has more to read.

• The developer should disable the borrow checker.
• Copilot does not support Rust.
• Switching model will eliminate the problem.
• It is a known rough edge — the borrow checker enforces constraints the model does not reliably track, so expect to iterate with compiler errors as the feedback loop.

• Nothing special; shell suggestions are as safe as any other.
• Run it in the project directory; version control will recover anything lost.
• Run it as root so permissions do not interrupt.
• Test the pattern with a non-destructive command first — echo or ls — so you can see exactly what it would match.

• Leave it; Python evaluates defaults per call.
• Fix it — the default is created once and shared across calls, which is a genuine bug even though the code runs.
• Leave it, but add a comment noting the risk.
• Report it to GitHub as a model defect.

• It avoids needing to review the implementation.
• The test states the expected behaviour precisely, which both constrains the generated implementation and gives you an independent way to check it.
• Copilot writes tests better than implementations.
• Tests are quicker to generate.

• Disable Copilot for large projects.
• Rename the interfaces so they are easier to guess.
• Accept the suggestions and refactor them onto the interfaces afterwards.
• Bring the relevant interfaces into context — open them or reference them — so the suggestion is written against the contracts you actually have.

• Copilot cannot generate valid C++.
• C++ has no way to detect memory errors.
• Mistakes produce undefined behaviour that may not fail at the point of the error, or at all during testing.
• Modern C++ removed the need for review.

• Generated code can be idiomatic in shape while omitting the substance the idiom exists for.
• The compiler will reject it.
• Go does not require error checking.
• It means the model chose the wrong language.