Arnaud Spiwack

> It seems `jupyter-client` is not for writing a kernel, but a client to send commands to a kernel. I couldn't find a dedicated lib, or extract a stand-alone subpart...

Indeed. There is no screenshot, but I imagine, from the description, that the interface will be pretty stripped down. You may have to work some for the syntax highlighting bit...

For what it's worth, I prefer using `TypeInType` systematically. `PolyKinds` is, I believe, more or less on a path to deprecation anyway.

Oh dear! I'm really sorry. I had missed your issue. There is no official benchmark, as of now. Nor is it quite clear to me what benchmark would be meaningful....

I gather that the problem that you are highlighting is solely on the `IO` implementation? Or does the same happen with, say `StateT Maybe`? What behaviour would you expect in...

We could, presumably, implement `modify'` using `state`, though. And implement `state` with `modifyRef`, though this is not atomic on `IORef` by default (which uses `modifyIORef`, rather than `atomicModifyIORef`. That being...

> Hmm, I might be misreading, but I don't see how. The following kind of interleaving is the issue I'm referring to, where `modify'` is not atomic. On `MVar` that...

(I don't know what the class would not be in `Capability.Writer.Internal`) The class methods are not an internal thing however, they very much matter if you need to define manual...

I don't think internal modules should have any sort of consistency. They are, by definition, not part of my API. Rather, use-at-your-own-risk extras.

``` > :info Stream type role Stream nominal nominal nominal data Stream (f :: * -> *) (m :: * -> *) r ``` Indeed… It's the inferred role too,...