2021-06-21: Selector trouble

[nikodemus]

I knew this was coming, but I'd forgotten.

Foolang uses Lisp/Smalltalk style interned strings, aka selectors aka symbols as message keys / method identifiers, which have an associated literal syntax.

#aMessage sendTo: anObject is same as anObject aMessage

Plan is that the compiler would not bother emitting code for methods that can never be called - because we know which selectors have no references to them outside the methods identified by them. Smaller code, faster compile times, all good.

This means that there must not be a way for user code to intern selectors at runtime.

Except that the interpreter must be able to do that: otherwise it cannot go from the string "1 + x" to executing #+ sendTo: 1 withArguments: [x]. (The interpreter can intern new selectors just fine by keeping a table of its own, but existing ones are the problem.)

There are a few options:

1. Just support Selector#intern: for now, and worry about removing it later. (Easy.)
2. Add an explicit table of all host selectors to interpreter, so that it can intern existing ones by looking up the name there. (Tedious but simple.)
3. Add a separate class SelectorTable which does the interning: having a reference to it causes there to be a reference to all selectors, preventing the compiler from ignoring methods, etc. (Most work by far.)

I'm really temped by number 1, but I know I will regret it later...

[nikodemus]

(I picked #1, I will regret it later - but there's enough stuff that needs to be reworked that this one is a drop in a bucket...)