Newtypes #1015
Newtypes #1015
Conversation
robdockins
force-pushed
the
newtypes
branch
4 times, most recently
from
298e2cc
to
0264ada
Compare
|
@pnwamk, @david-christiansen, There are some nontrivial questions about how to represent values of newtypes in the cryptol remote API, and I'd love you have your opinions about what we should do here. |
|
@yav, @brianhuffman, aside from the questions about the remote API, I think this PR is ready to go. |
|
Can we have a video chat about it in the first week of January? Just so there's time to go back and forth a bit about what the feature is used for and its history so we can pick the right trade-offs. |
Sure, I'll set something up. |
There was a problem hiding this comment.
I only looked at the diff on github, so not in a lot of detail, but it looks reasonable to me.
In the longer run we may want to represent newtypes with a single field a bit more efficiently, but let's worry about that if it every comes up. I'd say merge it in, this way we'll get to test it by using it.
| r = x.real * y.real - x.imag * y.imag | ||
| i = x.real * y.imag + x.imag * y.real | ||
|
|
||
| cplxEq : CplxQ -> CplxQ -> CplxQ |
There was a problem hiding this comment.
Note to self : should be cplxEq : CplxQ -> CplxQ -> Bit
src/Cryptol/Eval/Reference.lhs
Outdated
|
|
||
| At runtime, newtypes values are represented in exactly | ||
| the same was as records. The constructor function for |
the fields of the newtype body, and push newtypes into various parts of the evaluator, random testing, etc.
and use `TValue` instead of `Type` in more places.
in scope. Also, demote the `NewtypeEnv` to a type synonym, as the newtype was just annoying.
of `Type`. This allows us to directly carry the `Newtype` instead of having to look it up in a table at use sites.
functions. Newtype information is now propigated directly into types via the typechecker instead of being looked up separately.
make it responsive to the `show-examples` option.
by the newtype body.
We've had partial support for a while now for a
newtypeconstruct, but that support was never fully fleshed out. This PR is for completing the job, writing associated documentation, test cases, etc.CF #870