Sam A. Horvath-Hunt

This appears to affect all distros since we started [building in a Nix shell](https://github.com/unsplash/intlc/commit/404b38535ac24044b404e3363f4cfa46689a0f98). Linux distros can use `patchelf` like above with `gmp` and `libffi`.

0.6.1 resolves the regression in dynamic linking in 0.6.0. Most users should find it once again works out of the box. Users in non-FHS environments such as NixOS will continue...

Here's a starting point to building with Nix (dynamically linked): ```patch diff --git a/cli/CLI.hs b/cli/CLI.hs index 456de43..a5e1a9b 100644 --- a/cli/CLI.hs +++ b/cli/CLI.hs @@ -1,8 +1,5 @@ -{-# LANGUAGE TemplateHaskell #-}...

This looks easier said than done. We only parse and keep AST annotations/offsets for ICU messages themselves. The JSON is parsed as a quick conversion roughly `Text -> Map`. The...

Note also that `js`, `jsx`, and `dts` backends would now be fairly trivial to add, particularly if we changed the TypeScript output to be simply `js(x)` and `dts`.

Before I forget my idea was that potentially output could look something like: ```ts // Now { x: string; y: number } // After { x: string } & {...

That's interesting. I suppose it'd be analagous to `other`. Where `other` widens e.g. `'a' | 'b'` to `string` (encoded conveniently via a union), `null` could widen/unionise to `null`. Not just...

I agree, nullability warrants its own issue: https://github.com/unsplash/intlc/issues/105 I've also raised this... https://github.com/unsplash/intlc/issues/106

This'd break virtually all the TS tests, but here's roughly what a passing test with this feature would look like: ```diff diff --git a/test/Intlc/EndToEndSpec.hs b/test/Intlc/EndToEndSpec.hs index 4bc1c57..b443008 100644 --- a/test/Intlc/EndToEndSpec.hs...

If we decide to go this route we may consider a change in JSON structure: ```json { "locale": "en-US", "messages": { "header": { "message": "A very fun header" } }...