I also meant to say...

On Fri, Sep 17, 2010 at 3:04 PM, Radu Grigore <radugrigore@gmail.com> wrote:
> 1. Never generalize type variables that appear under ref.[...]
> 2. Keep track of side-effects, [...]

These are conservative, in the sense that they accept all pure
programs that pass Hindley-Milner.

> 3. Only generalize type variables if e1 [is a value]
> 4. [...] generalize type variables when they appear only in covariant
> positions.

These are *not* conservative.