Gabriel Scherer

We discussed this PR again today during an informal type-system-refactoring meeting. ### Backtrack vs. fallback The first thing to say is that I realized that I had missed a distinction:...

I agree that dealing with gadts optimally within `build_as_pat` is harder than the treatment we do for `type_pat`, and closer to what we would need to do to actually export...

We could definitely add `Dynarray.{to,of}_iarray` with the same behavior as `Dynarray.{to,of}_array` -- that is, making a copy. Same for `Dynarray.append_iarray`. (In retrospect I regret the name `unsafe_to_iarray` for a function...

I haven't tried, in part because I am no expert in writing concurrent data structures. My mental model is that a simple contextual version (that takes a mutex) would have...

Yes, I believe that marking the argument of the `unsafe_to_iarray` continuation `local` (so it cannot escape) would let us remove the `unsafe_` prefix.

Obviously we should inspect the user-agent to tell if the web browser displays the beginning of the title, or the end of the title ;-) (I'm okay with this proposed...

This is a duplicate of #28 so it could be closed. In #28 there is a proposed temporary fix for this issue: ``` pip install git+https://github.com/TimothyElder/pandoc-xnos --force-reinstall ```

I believe that the right long-term fix for this issue would be to use weak atomic read/write on an atomic variables, that is, acknowledge that the cache is used concurrently...

Computing a stable matching is a clever trick to give several fix suggestions as one. But today we don't typically do this, we stop on the first error. Suppose we...

I don't necessarily mean that there should be at most one renaming suggestion -- having several is fine. I am wondering if we can significantly reduce the complexity if we...