The glyph law

Most languages spell two different jobs with one symbol. = declares a field of a type and also stores a value into a variable. Those jobs happen at different times, and a reader should not have to guess which one a line is doing. Faber splits them.

• ← is a runtime event. It binds a value to a name, reassigns it, or mutates it. Every ← in a program happens when the program runs.
• = is a compile-time fact. It fixes the shape of a value inside a literal, or a declaration's metadata. Nothing happens at run time.
• → declares a function's return type; ⇥ declares its error channel. Both describe flow, and both are written where the flow is declared.
• ≡ tests equality, ↦ converts at run time and may fail, ∷ asserts a type at compile time.

The rule behind the split is one sign, one job. A glyph may have exact aliases, but it should not carry unrelated meanings. That is why a reader can scan a Faber function and see every data-flow operation at once: every ← is live, and every = is settled before the program starts.

What it looks like#

class Punctum {
    const int x
    const int y
}

main {
    const _ p ← Punctum { x = 10, y = 20 }
    var int count ← 0
    count ← count + 1
    print p.x, count
}

$ faber run
10 1

p ← Punctum { … } runs: it attaches a value to the name p. The x = 10 inside the braces does not run — it is the shape of the value being built, known while compiling. count ← 0 then count ← count + 1 are two runtime events in sequence. Nothing here is a matter of style; the glyph decides.

One canonical spelling, many sugars#

The same separation runs one level up, between what a construct means and how it is spelled. Faber defines one canonical spelling for each construct and accepts sugar spellings that parse to exactly the same node. The parser does not prefer one over the other; faber format --locale la re-emits the canonical form, while author mode preserves what was written.

Numeric containers follow it (tf32[4] and tensor<f32, [4]> are the same type), annotations follow it (@ option verbose … and the braced record are the same HirAnnotation), and reader locales follow it — Latin is the canonical surface, every pack is a rendering of it.

Why it matters#

Two properties fall out of the split:

• Local reasoning. Nothing about a line depends on distant context. A reader sees the declared shape and the runtime step separately, without knowing a library's internals.
• Lossless round-trips. Because the canonical form is defined and the sugar is a rendition of it, a program can be reformatted, converted, or re-emitted without changing its meaning. The glyphs and the type-first order survive every rendering; only the words change.

The full glyph inventory, including the comparison and logic operators, is in Glyphs and Latin. The design laws that keep these choices stable are in Design notes.