Replace bitflags! in the style crate with a lean crate-local macro

[nicoburns]

Summary

Replaces the bitflags 2.x dependency of the style crate with a lean crate-local bitflags! macro (style/flags.rs) that generates only the API subset the crate actually uses. The macro keeps the same name and invocation syntax (both the struct-defining and impl-only forms), so all 55 call sites are unchanged apart from:

• struct-form sites that had #[derive(Debug)] inside the invocation drop that derive — the macro always generates a Debug impl that lists contained flag names (matching bitflags' output, e.g. RestyleHint(RESTYLE_SELF | RECASCADE_SELF));
• servo/restyle_damage.rs replaces <ServoRestyleDamage as bitflags::Flags>::Bits::MAX with u16::MAX.

Per type, the macro generates the flag consts, a private (name, flags) table, empty/all/bits/from_bits/from_bits_truncate/from_bits_retain/is_empty/is_all/intersects/contains/intersection/union/difference/symmetric_difference/complement/insert/remove/toggle/set/iter/iter_names, and the 9 bitwise operator impls — ~120 expanded lines instead of bitflags' ~2,500 (no hidden internal type, forwarding macros, iterator types, Binary/Octal/Hex impls, from_name parsing, or serde/arbitrary/bytemuck hooks). all()/Debug are table-driven so #[cfg]-gated flags keep working.

Only the style crate is converted; selectors/, style_traits/ and stylo_dom/ still use the bitflags crate (which also remains a transitive dependency via cssparser), but bitflags is removed from style/Cargo.toml.

Measurements

Environment: Linux x86_64 VM, rustc 1.99.0-nightly (504869653 2026-08-03), CARGO_INCREMENTAL=0, default (servo) features, dependencies pre-built, touch style/lib.rs between runs, 3 runs each.

-Zmacro-stats for the style crate (sum of all bitflags-related macro rows):

	Uses	Expanded lines	Expanded bytes
before (bitflags 2.13.1)	~9,900	135,768	5,063,114
after (local macro)	100	11,810	492,312
delta		−91.3%	−90.3%

Wall times for the stylo crate itself (deps warm):

	before (median of 3)	after (median of 3)
cargo +nightly check -p stylo	19.87s (19.71 / 19.87 / 32.07)	19.42s (19.34 / 19.42 / 19.95)
cargo +nightly build -p stylo (debug)	39.26s (39.19 / 39.26 / 39.38)	38.46s (38.27 / 38.46 / 39.28)

Honest takeaway: despite the ~90% reduction in macro-expanded code (~124k lines / ~4.6 MB less for the compiler to expand, parse and type-check), the wall-time impact is small — roughly 0.5–1s (~2%) on both check and build, at or near run-to-run noise on this machine. Macro expansion/parsing of the bitflags output evidently isn't a dominant cost for this crate; the bulk of the time goes elsewhere (type checking the crate's own code, codegen, linking metadata).

Testing

• cargo +nightly check -p stylo and cargo check --workspace pass.
• cargo check --all-targets -p stylo has 10 pre-existing errors on main (test code referencing PseudoElement::MozRubyText/FirstLine); the same errors, and no new ones, occur with this change.
• The gecko feature can't be built locally (needs mozbuild); the gecko-cfg'd sites (url/gecko.rs, gecko/*.rs, gecko-only flags in values/specified/text.rs) were converted textually with care, but are unverified by the compiler.

Link to Devin session: https://app.devin.ai/sessions/d53d0b1a1f36492486a5ce4537362adc
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