Unlike git, Mercurial considers an hg commit --amend which doesn't change the working copy to be an error.
The current behavior for this is fairly bad, since the user gets an exception wrapping the entire command ("Command Failed! ...") and it's pretty verbose and not obvious what has happened.
Some alternatives are:
- Detect this condition and raise a more tailored exception, like a UsageException.
- Detect this condition and succeed.
Although I tend to think (1) is the right approach in general (that is, arc x should usually behave like git x or hg x), I went with (2) here because we have a handful of amend callsites and they all assume git semantics (no-op amends are successful), and because I think Mercurial's behavior is a little silly (the working copy ends up in the correct / expected state, which seems fairly clearly like a success to me).