@rtif-sdk/test-kit

Document builders and assertions for testing RTIF. Framework-agnostic — every assertion throws a plain Error with a focused message, so it works under any test runner. Depends on @rtif-sdk/core only.

npm install -D @rtif-sdk/test-kit @rtif-sdk/core

30 seconds

import { doc, p, h1, b, sel, assertDocEqual } from '@rtif-sdk/test-kit';
import { apply, invert } from '@rtif-sdk/core';

const before = doc(p('Hello ', b('world')), h1('Title'), p());
const op = { type: 'split_block', at: 3, id: 'x1' } as const;

assertDocEqual(apply(apply(before, op), invert(before, op)), before);

doc() returns a finished Document directly — no .build() step.

Builders

| Builder | Produces | |---|---| | doc(...blocks) | Document; no args → one empty paragraph | | p(...) | paragraph block | | h1(...) / h2(...) / h3(...) | heading with attrs: { level: n } | | blockquote(...) | blockquote block | | codeBlock(...) | code_block block | | li(...) | list block (one block per item) | | hr(opts?) | horizontal_rule block (empty content) | | block(type, ...) | any block type | | t(text, marks?) | span with optional marks | | b / i / u / s / code | bold / italic / underline / strikethrough / code span | | link(text, href) | span with { link: { href } } | | sel(anchor, focus?) | Selection; focus defaults to anchor |

Block builders take strings (unmarked spans), spans, and an optional { id?, attrs? } object first or last:

block('paragraph', 'Hello ', b('world'), { id: 'intro' })
li('first', { kind: 'ordered' })   // a non-{id,attrs} trailing object IS the attrs
h2('Subtitle', { id: 'sub' })
doc(p({ id: 'a' }, 'one'), p('two'))  // unpinned blocks get b1, b2, ... (per doc() call)

Ids are assigned deterministically and skip pinned ones. Builder output satisfies the normalization invariants the builders can guarantee locally; adjacent equal-mark spans are emitted as written (builders never merge).

Assertions

assertDocEqual(actual, expected, opts?)

Structural equality. Block ids are ignored by default — parsers and round trips regenerate ids, and most tests shouldn't care. Pass { compareIds: true } to catch id drift. On mismatch the error names the first differing block and shows just that block side by side, never the whole document:

Documents differ at block 1:
actual:   {"type":"paragraph","spans":[{"text":"wrld"}]}
expected: {"type":"paragraph","spans":[{"text":"world"}]}

assertValidDoc(doc)

Checks all six normalization invariants (block/span shape, no stray empty spans, no adjacent equal-mark spans, no {} marks/attrs, unique ids, ≥ 1 block) and names the violated invariant and offending block.

docToText(doc)

Compact debug rendering: blocks joined by \n; unmarked spans are raw text; marked spans render as <keys>text</> with keys sorted and non-true values appended as =JSON:

docToText(doc(p('Hello ', b('world')), p(link('here', '/x'))));
// 'Hello <bold>world</>\n<link={"href":"/x"}>here</>'

Handy in failure messages and snapshot-ish comparisons:

expect(docToText(result)).toBe('Title\nHello <bold|italic>world</>');