> you can forget to check

Linters can help with this.

Specifically for this issue, linters also have many false positives. Some Go libraries trying to encourage a fluent style will accept an error for some logic also return it, so you can `return x.HandleError(err)` - but if you don't want to return it, you obviously don't care it returns what you just passed it. (I personally consider fluent methods a bad idiom in Go, but I also don't get to write all the Go code in the world or even in my project.)

There are also a lot of functions that return errors because Go's type system demands that if the interface returns two values `T, error`, every implementation must also - it won't auto-create a nil value for the second result. That's reasonable if you are committed to errors just being normal values. But such a restriction would not be necessary if the interface could be declared with a sum type - promotion of a `T` to a `Either<T, ...>` or `Just T` or so on would be fine for all types, not just error handling . Lots of infallible Writer implementations like bytes.Buffer and hash.Hasher suffer from this, and linters can't be aware of all such cases.

Sure, but given a choice, I'd rather work with primitives that are correct-by-construction, not correct-if-I-use-an-extra-tool-and-actually-act-on-its-advice.

If you do use a linter and have it set up so linter issues are fatal to the build, then you run into the issue that if the linter throws false positives, you have to add exclusion rules (if the linter even supports that) or downgrade linter issues back to warnings, and lose the benefit entirely.