Refactor overlapping test coverage for streaming and creative rewrites

[ChristianPavilonis]

Summary

• Refactor overlapping UTF-8 boundary tests in streaming_replacer into shared helpers and consolidated coverage-focused cases.
• Consolidate repetitive srcset and imagesrcset creative rewrite tests into shared helpers plus table-driven case matrices.
• Keep runtime behavior unchanged while reducing duplicated test setup and preserving multibyte, emoji, spacing, relative path, and descriptor coverage.

Changes

File	Change
crates/trusted-server-core/src/streaming_replacer.rs	Added reusable chunk-processing test helpers and replaced overlapping UTF-8 boundary tests with consolidated case-based coverage, including explicit multibyte and emoji boundary splits.
crates/trusted-server-core/src/creative.rs	Added shared srcset/imagesrcset test helpers and merged overlapping attribute rewrite tests into compact table-driven cases while keeping integration-specific tests separate.

Closes

Closes #457
Closes #458

Test plan

• cargo test --workspace
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo fmt --all -- --check
• JS tests: cd crates/js/lib && npx vitest run
• JS format: cd crates/js/lib && npm run format
• Docs format: cd docs && npm run format
• WASM build: cargo build --package trusted-server-adapter-fastly --release --target wasm32-wasip1
• Manual testing via fastly compute serve
• Other: Focused Rust test runs for the new consolidated test cases

Checklist

• Changes follow CLAUDE.md conventions
• No unwrap() in production code — use expect("should ...")
• Uses tracing macros (not println!)
• New code has tests
• No secrets or credentials committed

[aram356]

Summary

Pure test refactor consolidating overlapping srcset/imagesrcset attribute tests in creative.rs and UTF-8 boundary tests in streaming_replacer.rs. The 1:1 mapping from old tests to new table-driven cases is faithful, and several imagesrcset cases gain stronger assertions (>= 2 rewritten candidates instead of >= 1). Two issues need attention before merge — see inline comments.

Blocking

🔧 wrench

• Multibyte boundary test still doesn't split mid-character: see inline at streaming_replacer.rs:468. The PR fixes the analogous emoji bug but leaves the multibyte case mis-named.
• Helper assertion message contradicts its check: see inline at creative.rs:798. >= 2 with message expected two should be == 2.

Non-blocking

🤔 thinking

• <div> → <img> element switch in helper: see inline at creative.rs:781. Functionally equivalent given the [imagesrcset] selector, but neither old nor new tests cover the realistic <link rel="preload" as="image" imagesrcset="…"> path. Worth a follow-up issue.
• Loss of cargo test <single-case> granularity: collapsing 8 standalone tests into 2 table-driven tests means a failing inner case can't be re-run in isolation — only the umbrella test. The case_name in assertion messages mitigates this for log-reading, but bisecting harder failures is more friction. Acceptable tradeoff; flagging as a real downside without a parameterized-test macro (rstest or similar).

📝 note

• Test-name surface shrinks: cargo test output loses 6 named tests (8 srcset/imagesrcset → 2 umbrellas, plus the streaming consolidation). Any CI dashboard or flake tracker keyed on test names will need to be aware.

CI Status

• cargo fmt: PASS
• cargo clippy: PASS
• cargo test: PASS
• vitest: PASS
• format-typescript / format-docs: PASS
• integration tests / browser integration tests: PASS

[aram356]

🔧 wrench — test_process_chunk_boundary_in_multibyte_char still doesn't split mid-multibyte.

Byte 23 falls between ø (bytes 21–22) and r — i.e., on a valid UTF-8 boundary, not mid-character. The test name claims to exercise a split inside a multibyte char, but it doesn't. Since you correctly fixed the analogous emoji case in this same PR (the original test_process_chunk_emoji_boundary never split at all), please apply the same fix here.

The replacer's host-fallback rule (example.com → new.com) keeps the assertion green even when the URL-prefix replacement is severed, so the existing assertion still works.

Fix:

let result = process_with_explicit_splits(
    create_url_replacer("example.com", "https://example.com", "new.com", "https"),
    content,
    &[22, content.len()],
);

(byte 22 is in the middle of ø — first byte was 21).

[aram356]

🔧 wrench — Assertion message and check are inconsistent.

out.matches("/first-party/proxy?tsurl=").count() >= 2,
"case `{}` expected two rewritten {} candidates: {}",

The message says expected two but the predicate accepts ≥ 2. All current cases produce exactly two rewritten candidates, so tightening the check is the correct fix and would catch any future regression that quietly rewrites a third (e.g., relative-candidate) entry.

Fix:

assert_eq!(
    out.matches("/first-party/proxy?tsurl=").count(),
    2,
    "case `{}` expected exactly two rewritten {} candidates: {}",
    case_name,
    attr_name,
    out
);

[aram356]

🤔 thinking — The new helper wraps both srcset and imagesrcset in <img>. The old imagesrcset tests used <div>. Both are HTML-spec-illegal placements — imagesrcset belongs on <link rel="preload" as="image"> — but both rely on the element-agnostic [imagesrcset] selector at creative.rs:660, so runtime behavior is identical.

Coverage gap (pre-existing, not introduced here): the realistic <link rel="preload" as="image" imagesrcset="…"> path also flows through the link[href] handler at creative.rs:556-561. Neither the old nor new tests exercise that path. Worth a follow-up issue rather than expanding this PR.