Thanks for the review and new patch, David! Your approach makes sense and the patch looks good to me.

However, regarding backwards-compatibility, is that really a concern?

Currently the deserialization process systematically 1) Adds the 'httponly' and 'secure' dict keys to the cookie object and 2) Puts an empty string value for those keys, regardless of whether those flags are present or not in the loaded string. So currently nobody's code could possibly rely on any particular state or behavior in the cookie object to determine if those flags were originally present in the loaded string.

I guess I'm wondering what could possibly break in people's code if we now implemented a fully logical fix for this. What do you think?