---
enabled: true
touches: [crates/dom/**, roadmap.md]
depends-on: []
---

# Read the inline style="" attribute into the cascade

`held-out-rerun`'s WPT re-measurement found that every remaining failure in that
corpus traces to one gap: `style="..."` attribute values are never read into the
cascade, only `<style>` block rulesets are. `TElement::style_attribute()` in
`crates/dom/src/style.rs` is a hard-coded stub returning `None` — Stylo's own hook
for this, left unimplemented since Phase 0's corpus styled everything through
`<style>`, never `style="..."`. `ElementData::attr` (`crates/dom/src/node.rs`)
already returns the raw attribute string verbatim, proven by `parse.rs`'s
`attributes_are_captured_with_name_and_value` test — no parser change is needed.

Implement `style_attribute()` to look up `self.element().attr("style")` and, when
present and non-empty, parse it with Stylo's own
`style::properties::declaration_block::parse_style_attribute` — the same function
Servo and Gecko use for this — then cache the resulting
`Arc<Locked<PropertyDeclarationBlock>>` somewhere `ElementData` can own it; today
it has nowhere to. `Document::parse` (`crates/dom/src/style.rs`) already
constructs the `UrlExtraData`, `QuirksMode` and shared lock this needs; retain
whichever of those the cache requires.

Stylo's rule collector already inserts the style attribute at the correct cascade
level the moment `style_attribute()` stops returning `None` —
`same_tree_author_normal()` after stylesheet author-normal rules, and
`same_tree_author_important()` separately for `!important` declarations. No
specificity or cascade-origin logic belongs in this codebase; do not add any —
correctness comes from Stylo's own rule tree, not from anything written here.

Requirements:
- A `style=""` attribute with no matching stylesheet rule applies on its own.
- An inline declaration overrides a stylesheet rule on the same property, and a
  stylesheet `!important` declaration still overrides a plain inline one. Model
  both as new tests in `crates/dom/src/computed.rs`, following
  `an_author_margin_on_body_overrides_the_user_agent_default`.
- `crates/dom/src/bin/style_check.rs`'s corpus (built from
  `crates/harness/src/cases.rs`) has no `style=` fixture today. Add one, or say
  in the Result section why it wasn't needed to prove the fix — don't leave the
  question unanswered.

The float-exclusion gap `held-out-rerun` also found on
`auto-margins-used-values-with-floats.tentative.html` (a `float: right` box
declared in a `<style>` block never excluded from the block formatting context)
is a separate, independent defect. Do not fix it here.

`cargo test --workspace` must pass. `style_check` and `box_check` must report the
same 2958/2958 corpus figure they do today, unless a new `style=` fixture was
deliberately added to the corpus for this work — if any other count moves, a
recipe was relying on the gap; report it in the Result section rather than
adjusting the recipe. Do not run `held_out_check` here; a future re-measurement
spec does that once this has landed, so the before-and-after is measured on a
committed base.