10. The resolved document

The resolver turns one source file into a resolved document. This chapter defines its exact shape and how two implementations are compared. This is what makes Level 2 testable without a browser.

10.1 Inputs

DOC-1. The resolver takes: the source text, the file name, the token registry (4.5) and the motion registry (7.3). The registry defaults are the built-in tokens and no motions.

10.2 The document

{
  errors:        [ { file, line, message, hint? } ],      // parser errors, then resolver errors
  warnings:      [ { file, line, message, hint? } ],
  rules:         [ Rule ],
  shadowRules:   [ Rule ],
  raws:          [ { css, file, line, order } ],
  fontLinks:     [ { url, line, file } ],
  motionNames:   [ name ],                                // motions defined in this file, in source order
  motionFrames:  { name: { essential, frames: [ { offset, props } ] } },
  mediaQueries:  [ query ],                               // distinct, in first-use order
  tokens:        { key: value },                          // the final token map
  registryRule:  Rule
}
Rule = { id, sel, media, signals, containers, layer, order, decls, file, line, pseudo, states }
decl = { prop, value, line, via?, fromGradient? }

DOC-2. rules holds the ordinary rules and the token rules from tokens blocks, in the order they were emitted. shadowRules holds pseudo-element rules (5.8). raws holds raw and fonts font-face blocks in emission order. The registry rule is not in rules.
DOC-3. mediaQueries lists every distinct entry of media found in rules and shadowRules, in the order first seen, rules first.
DOC-4. tokens is the final map of 4.3 (every key, with the last definition in source order). Keys are compared as a set; the canonical form sorts them.
DOC-5. A declaration may carry via (the semantic property that produced a display value) and fromGradient (true for the background-image from gradient:). Both are part of the record.
DOC-6. states is a set of state words. It is written as a sorted array.
DOC-7. ok is not stored. It is errors.length == 0.

10.3 Canonical form

To compare two documents they are written as JSON in this form.

DOC-8. Object keys are sorted by code point. Arrays keep their order, except states and the tokens object, which are sorted.
DOC-9. Numbers are written as the shortest decimal that reads back to the same double, with no exponent for values between 1e-6 and 1e21, and no -0. Infinity is the string "Infinity" (an iterations of infinite).
DOC-10. A field that is undefined is omitted. A field that is null is null.
DOC-11. Strings are compared as UTF-8 text.
DOC-12. Two implementations are equal on a source when their canonical JSON is byte-identical, after removing lint warnings from implementations that declared lint: false (DIA-12).

10.4 The corpus

The comparison needs inputs. The corpus is a folder of .dssl files:

every stylesheet the site ships (public/styles/*.dssl),
every fenced dssl block from the guide, the readme and this spec,
the fixtures of the existing test suites,
one file per diagnostic message, written to trigger it.

For each input, scripts/native-dump.mjs writes the canonical JSON of the reference engine's output to native/expected/. The Rust test native/tests/differential.rs resolves the same input and compares.

DOC-13. The expected files are generated. They are committed so that a change in the reference engine shows up in review as a change in expected output, and so the Rust tests run without Node.
DOC-14. A divergence (chapter 12) is excluded from the comparison by listing the input in native/corpus/EXCLUDED.md with the divergence id. An excluded input is still run, and its diff is printed, so a fix is noticed.
DOC-15. Once the harness exists, native/PARITY.md lists, per chapter, how many corpus inputs pass, fail and are excluded. It is generated by the test run and not written by hand.

10.5 What the document does not contain

The syntax tree (Level 1 compares it separately, 2.9).
Behaviors (they are expanded during the walk and do not appear in the output).
Anything about a particular element. Matching is stage 3 and needs a document.