[bitc-dev] Effect variables cannot be eliminated

[Jonathan S. Shapiro]

My statement that effect variables can be eliminated in the types proves
to be wrong. Matt M at the lambda-the-ultimate wiki distilled a
counter-example:

  http://lambda-the-ultimate.org/node/2924#comment-43177

So this means that function types need to carry effect variables in at
least some cases. In principle, it also means that a complete typing of
an expression needs to have both an associated effect and a type, but I
think we can elide printing those most of the time.


shap