Currently, they are not if we have unused parameters. The trait definition will have the unused parameters while the impl will not.

I don't have a preference, but they should be consistent.

I see that currently it is problematic because in the function generating the trait you don't do the analysis, so you donno if a component is used or not. I would just remove the --weight-trait flag and always write the trait and impl together.
then we can actually have tests that ensure what they produce compiles.

Currently as @shawntabrizi said the trait definition is just a helper and it should be manually tweaked. I think it can and should be fully automatic.