[names] Add a section that clarifies how to emit names for JS

[szuend]

This is a first draft that clarifies on which JS tokens we expect mappings with a name.

I tried to stick to the ECMA-262 AST on a level where it makes sense, but let me know what you think. I also used lang

A nice tie-in with the "Scopes" proposal is, that the ( token that shows up in the definition is also what we'll probably recommend as the "generated range" start, so that would fit rather well together.

Thinks that still need clarification:

• Phrasing around the conditions "when" to emit a mapping with a name. Since we don't know the source language we need to keep it generic but we should make it clear that it should also make sense semantically.
• The current phrasing uses "should" and "may". I'm not sure if we should phrase it as "must"?
• Should we spell out all the AST constructs for variables where we expect mappings with a name?
• Figure out if we want to allow more complex expressions for variable names, e.g. if a minifier turned obj.x into f.
• Any other open question you might come up with :)

[szuend]

Hey all, I incorporated the feedback form yesterday's meeting. Thanks a ton for the fruitful discussion! In particular:

• For functions/methods/generators emit the named mapping on the
  binding identifier by default (or opening parenthesis if no identifier is present).

• Special case single argument arrow functions that have neither
  a binding identifier or an opening parenthesis.

• Introduce a definition for "named mapping". I.e. a decoded
  mapping entry where the name field is not null.

• Call out that generators may emit additional "named mappings"
  as they require. Either to support existing tools or to fit
  their requirements.

English is not my native language and writing spec text is not my strong suite, so I'd welcome some word smithing/rephrasing to clear up the meaning. I also spelled out some more things that require review. Thanks!

[rbuckton]

In general, TypeScript does not transpile a function to an arrow function in any case. Instead, we might transpile an arrow function with type information to an arrow function without, or an arrow function to a function as part of down-leveling to ES5.

[szuend]

Replaced with "A minifier renaming functions and variables or removing function names
from immediately invoked function expressions. "

[rbuckton]

This paragraph is confusing and doesn't properly address the concern I raised. It now says to emit a name mapping on (a) either the BindingIdentifier or ( of an arrow function, or (b) the => if it is a single argument arrow function with neither a BindingIdentifier nor Arguments.

(b) is never true as a single argument arrow function has only a BindingIdentifier, and thus (a) is only the ever case that is true.

Correct me if I'm wrong, but I believe we discussed in the meeting that there are two possible places to emit a name mapping for an arrow function or async arrow function:

1. For an ArrowFunction or AsyncArrowFunction with Arguments, the name mapping is attached to the opening (. A generator may opt to also emit a name mapping attached to the => for consistency with (2).
2. For an ArrowFunction or AsyncArrowFunction with a SingleNameBinding, the name mapping is attached to the =>.

[szuend]

I'm embarrassed to say that I don't know what I was thinking when I wrote this.

I fixed it up as proposed but with concrete references to the productions.