Now it properly introduces type variables with CLet and checks cmdMap
and subMap

This was very tricky. All of the variables need to be allocated into
the latest rank, and this t...

The old version would rebuild the Content for each var. The new one
just traverses the structures.

The idea to handle JSON is to just give the full text of the error
message rather than breaking ...

For example, when there is a pattern match in a function arg, but the
type has other options it ...

There is now a bit more metadata about (1) the expression that
generated the constraint and (2) ...

The main thing that was tricky here was that the way I destructured
patterns needed to change. B...

I believe it would require 8 levels of nested lets to use up all the
initial slots. The logic wi...

This is really important for making sure there’s not a bunch of
pointless garbage kept around! J...

The necessary metadata just got centralized into an already existing
thing.

This commit also introduces AST.Canonical.CtorOpts which caches which
constructor optimizations ...

Now that the AST tracks whether something is qualified or not, it makes
sense to change the stru...

Track qualified and unqualified variables as separate AST nodes, and
mark whether the names are ...

This way when the Elm.Name implementation gets optimized, there won’t
be any unneeded allocation...

I don’t think the compiler generated this operator for anything in any
prior version, which woul...

The root motivation was like this:

1. I want as much control as possible over error messages.
2...

The problem was handling destructuring bindings, like (let (x,y) = … in
…)

Before I generated a...

I really wonder how much of the language can be taught in these hints…

Primarily this starts tracking the location of any given local
definition. That way, if there is...

Probably not a big deal. Possibly the (>>=) usage is not as good.
Unclear.

The main thing is th...

In the case that there is no prefix, you need to detect reserved words!

Will come back and add things when those things actually work.

This error is for when folks say (Just) rather than (Maybe(..))

It does not mean it wasn’t foun...

The other version allowed ONE or more.

Turns out it doesn’t make that one error message better, so it’s better
to not allocate a spot f...

Man, wouldn’t it be cool if the compiler would detect this and tell you
about it at compile time :P

In the olden times, we had a series of definitions like this:

var Elm = {};
Elm[‘Thing’] = Elm[...

This way you do not get a bunch of unnecessary stuff in prod mode.

Figure out the trie structure of modules, generate the Builders for
anything that needs to know ...

I don’t really like having this Elm.Utils file in the first place, so
the goal would be to remov...

The new version needs to know which canonical modules are available.
From there it can figure ou...

The old KChunk did not distinguish between Elm variables and kernel
variables. This is necessary...

I believe this was nice because it hid the A.Located part. Now that
that is part of the error it...

Checks for duplicates, only in exports, only in docs, missing type
annotations, missing docs. I ...

Also explicitly models whether the user puts a `module` declaration
line or not.

This way they definitely get allocated only once

So the Name.Tracker type only needs to track one set, rather than two
for each sort of dependency.

It now lives in the object graph nodes, so it makes sense to have all
the details in one place.

It is used in a decent number of places in code gen

When we have type information, it’ll be possible to do this more often.
Nonetheless, this should...

The code gets a decent bit simpler thanks to the improvements in
earlier phases:

- There is no ...

For example, each case expression saves a temporary variable just in
case. That means we do all ...

This makes it easier to delay decisions about how code should be
generated.