Opaque by default

[nad]

Opaque by default

[jespercockx]

Where would you put the unfolding statement if there is no opaque block? Currently the syntax is:

opaque
  unfolding ...
  x = ...
  y = ...

Without the opaque it would not be clear if the unfolding declaration applies only to x or also to y.

[nad]

If you want to use unfolding, then you should use opaque.

[nad]

This feature request was discussed during a recent developer meeting. Someone (@jespercockx?) suggested that there should also be non-pragma syntax corresponding to --opaque, and @plt-amy suggested that one could use opaque module to make "everything" inside a module opaque. I wonder if it might make sense to instead use attributes for this: one can already use module @0 to make a module erased, perhaps the notation could be module @opaque.

That raises the question of whether one should also allow this notation for individual functions:

@opaque f : …

Note that I don't think the possibility to use opaque module or module @opaque makes the flag --opaque pointless: one can put this flag in a project's .agda-lib file to make things opaque by default, similarly to how one can put StrictHaskell in a .cabal file to make code strict by default in a Haskell project.

[JacquesCarette]

These proposed extensions sure make a lot sense to me.

[xekoukou]

I don't know the exact functioning of opaque. I believe it abstracts the implementation of a function. It forces the user to manually unfold something.

I agree that it should be the default. I am concerned with teaching people the correct way of programming in agda.
For example, instead of unfolding, would it be better to create an abstraction? How can creating an abstraction be more natural and easy than unfolding?

For example, you have a specific implementation of Integers. And you want to prove something that is only affected by properties of integers being an abelian group. How can we teach/guide people not to unfold and just create the abstraction?