Re: incorrect signed data

[Roland McGrath]

I think you're right on both counts, but I haven't delved into the
potential weirdness from changing sign-extension behavior.