[compile-time experiment] Dedupe repeated per-longhand code in generated properties.rs

[nicoburns]

Summary

Compile-time experiment: shrink the mako-generated properties.rs by factoring code that was emitted once per longhand into shared handwritten helpers, and measure the effect on build times.

What was deduplicated (all in helpers.mako.rs's longhand()/shorthand() defs):

• Cascade glue: the ~40-line per-property CSSWideKeyword match in each cascade_property is replaced by one call to a new shared PropertyDeclaration::cascade_simple_wide_keyword(context, id, is_inherited, inherit_or_reset, reinherit_with_zoom) in style/properties/mod.rs. Per-property behavior (inherit vs reset, zoom-dependent re-inheritance, debug asserts, for_non_inherited_property) is passed via fn pointers/flags. Zoom-dependent longhands emit a tiny local reinherit_with_zoom fn.
• Import boilerplate: the ~16 use lines per longhand module (and ~8 per shorthand module) become a single glob import of new longhand_prelude/shorthand_prelude modules.
• Vector-list machinery: the per-property generic OwnedList<T> struct + Animate/ComputeSquaredDistance/ToAnimatedZero/From impls emitted for each vector longhand are replaced by three shared generic types in new style/properties/longhand_lists.rs (NonEmptyCommaList, EmptyCommaList, EmptySpaceList), chosen per property in the template. The shared-ArcSlice ComputedList path (only text-shadow in servo) stays templated.
• from_gecko_keyword (already #[cfg(feature = "gecko")]) is now only emitted for gecko-engine builds.

Note: the shared list types now carry Animate/ToAnimatedZero blanket impls for any T meeting the bounds, whereas previously those impls only existed for longhands with a vector animation_type. This is a strict superset (no existing behavior changes).

Measurements

2-core linux VM, default (servo) features, protocol: cold rm -rf target && cargo build --timings; cargo check warm, then touch style/lib.rs && time cargo check (both with incremental on and with CARGO_INCREMENTAL=0).

Metric	Before	After
Generated properties.rs lines	123,136	109,723 (−10.9%)
Cold cargo build (total)	99.0s	97.9s
stylo unit time (cargo-timings)	53.4s	54.6s
touch style/lib.rs + cargo check (incremental)	8.3s	8.3s
same, CARGO_INCREMENTAL=0	20.4–20.9s	20.6–23.1s

Conclusion: the generated file shrinks by ~11%, but compile times are unchanged within noise. The deduplicated code was cheap-to-compile boilerplate; stylo's compile time appears dominated by the remaining giant match-based dispatch (PropertyDeclaration/AnimationValue/LonghandId impls), derives, and the per-field style-struct accessors, which are out of scope here (table-driven dispatch is being explored separately).

Testing

• cargo check: clean, no new warnings.
• cargo test --workspace --exclude stylo: all pass.
• cargo test -p stylo fails to compile on main as well (pre-existing PseudoElement::MozRubyText/FirstLine errors in selector_parser.rs tests), unchanged by this PR.

Link to Devin session: https://app.devin.ai/sessions/4acc993f357a4cd9ac8e17704aa71ed6
Requested by: @nicoburns

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring