Changelog: tramaj-js

tramaj-js 0.2.02026-10-15

TypeScript port: parser, evaluator (concrete and symbolic modes), Node JSON codec and static analyses, with language v5. The first release on npm.

• added Hand-written port checked against `corpus/cases` by `test/corpus.test.ts`.
• added Object destructuring patterns in `@` bindings and lambda parameters (decisions section 17): a parser lowering to plain `Let`/`Lambda`, checked by shared corpus cases 100-118.
• changed Breaking: numbers are two types, an integer and a float (decisions section 18). A literal and a JSON number are typed by their text: `3` is an integer, `3.0` and `3e0` are floats. Integers cover the guaranteed range, `-(2^53 - 1)` to `2^53 - 1`; one outside it is a parse error as a literal and a `TypeMismatch` in the context.
• changed Breaking: a number keeps its type when written. `str(1.0)` is `1.0` where it was `1`, and every JSON boundary keeps a float a float.
• changed Breaking: `eq` never equates an integer with a float, and `lt`/`lte`/`gt`/`gte` take two integers or two floats; a mixed pair is a `TypeMismatch`.
• changed Breaking: the `Json` type gains `Float` and `bigint`. A plain `number` that is a safe integer is an integer and any other is a float; the float `3.0` is `new Float(3)`, and a result holds a `Float` for a whole-valued float. `isJsonObject` is false for a `Float`.
• added `parseJson` and `stringify`, a JSON reader and writer that keep `3` and `3.0` apart, which `JSON.parse` and `JSON.stringify` do not; `float`, `isInteger`, `isFloat`, `numberValue`, `classifyNumber`, `normalizeNumbers`, `formatInteger` and `formatFloat`. `formatNumber` is removed.
• changed Breaking: `"$term"` is a reserved object key, `int` and `float` replace `number` as v4 type primitives, and `nodeFromJson` rejects a number outside the value domain. In the AST, `IntLit` and `FloatLit` replace `NumberLit`, and `ScalarInt` and `ScalarFloat` replace `ScalarNum`; in `Value`, `int` and `float` replace `number`.
• added The arithmetic profile: `sum`, `product`, `negate`, `quotient`, `inverse`, `floor-quotient`, `modulo`, `floor`, `real` and `round`, and the `NotRepresentable` error. It is off by default: `Options`, `defaultOptions`, `evalProgramWith` and `runProgramWith` turn it on. `round(x)` is the integer nearest to `x`, ties away from zero (decisions section 20).
• added Terms: over a symbol an arithmetic builtin evaluates to `{"$term": op, "arguments": [...]}`, with its operands as written.
• added `arithmeticNames`, `arithmeticOps` and `deepArithmeticOps`, the arithmetic names a program references.
• added `sort-by(list, fn)` and `sort-by-descending(list, fn)` (decisions section 20) order a list by the key `fn` gives each element. Keys are all integers, all floats or all strings, compared by code point; anything else is a `TypeMismatch`, and a symbol or a term a `NotConcrete`. Both are stable, each on its own terms, and `fn` is applied once per element, in index order. They are core forms, in every profile.
• added `format-number(x, decimals, group)` writes an integer or a float in positional decimal notation, with `decimals` digits (0 to 20) after the point and `group` between the groups of three digits of the integer part. It rounds the exact value of the number, ties away from zero, and never writes an exponent or a negative zero. An ordinary builtin, in every profile.
• changed Breaking: `Expr` gains the constructor `SortBy` (`descending`, `collection`, `fn`), which `subExprs` and every analysis traverse. `sort-by` and `sort-by-descending` are special-form names, as `map` is: a program that bound one and called it now gets the form, or a parse error without exactly two arguments. `format-number` is a builtin name and `round` joins `arithmeticNames`; a program that binds either still shadows it.
• added `emittedConstraintCount`, the number of constraints an evaluation emitted before equal ones are made one. It is there for tests, which have no other way to count the applications of a function.