Add Sourcepoint first-party CMP integration

[ChristianPavilonis]

Summary

• Adds a new Sourcepoint CMP (Consent Management Platform) first-party proxy integration that routes cdn.privacy-mgmt.com and geo.privacymanager.io traffic through Trusted Server, eliminating third-party requests for consent management assets.
• Implements four rewriting layers (HTML attributes, JS response bodies, runtime config trap, client-side DOM guard) to ensure comprehensive URL coverage across static markup, dynamically loaded scripts, and Next.js hydration payloads.
• Ships disabled by default with full TOML configuration, integration documentation, 14 Rust unit tests, and 6 TypeScript tests.

Changes

File	Change
crates/trusted-server-core/src/integrations/sourcepoint.rs	New ~860-line integration module: route registration, CDN/geo proxy handlers, HTML attribute rewriter, JS body rewriter (rewrite_script_content), head injector (window._sp_ property trap), Accept-Encoding scoping, backend selection, and 14 unit tests
crates/trusted-server-core/src/integrations/mod.rs	Register sourcepoint module and wire it into the integration registry
crates/js/lib/src/integrations/sourcepoint/index.ts	TypeScript entry point that bootstraps the client-side script guard
crates/js/lib/src/integrations/sourcepoint/script_guard.ts	Client-side DOM insertion dispatcher handler that intercepts dynamically inserted <script>/<link> elements pointing at Sourcepoint CDN and rewrites them to first-party paths
crates/js/lib/test/integrations/sourcepoint/script_guard.test.ts	6 Vitest tests covering the script guard (rewrites, skips, and teardown)
docs/guide/integrations/sourcepoint.md	Integration documentation: overview, setup steps, config reference, architecture
docs/guide/integrations-overview.md	Updated overview with Sourcepoint in the Mermaid flowchart, performance characteristics table, and environment variables section
trusted-server.toml	Added [integrations.sourcepoint] config stanza (disabled by default)

Closes

Closes #145
Closes #344
Closes #345

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

Checklist

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

[aram356]

Summary

Adds a well-structured Sourcepoint CMP first-party proxy with four rewriting layers (HTML attributes, JS body, runtime config trap, client-side DOM guard). The implementation follows existing integration patterns closely, has 14 Rust + 6 JS tests, and ships disabled by default. A few items need attention before merge.

Blocking

🔧 wrench

• take_body_str() panics on non-UTF-8 upstream responses: response.take_body_str() at line 400 will panic if the upstream CDN returns invalid UTF-8 with a JS Content-Type. Read bytes and attempt conversion with a fallback instead. (sourcepoint.rs:400)
• CI failure — format-typescript: Prettier check is failing on test/integrations/sourcepoint/script_guard.test.ts. Must be fixed before merge.

❓ question

• Redirect Location headers not rewritten: The JS body rewrite is gated on status == 200. If the CDN returns a 3xx redirect with a Location pointing to cdn.privacy-mgmt.com, the browser follows it back to the third-party CDN, defeating the proxy. (sourcepoint.rs:394)

Non-blocking

🤔 thinking

• Upstream response headers forwarded verbatim: Set-Cookie and other headers from cdn.privacy-mgmt.com pass through to the client on the non-rewritten path. Same as Didomi, but worth confirming this is intended. (sourcepoint.rs:418)
• Regex matches mismatched quotes: The CDN URL pattern can match 'url" (single open, double close). Extremely unlikely in minified JS but could produce malformed output. (sourcepoint.rs:65)

♻️ refactor

• Redundant rewrite_sdk guard: Checked in both handles_attribute() and rewrite() — the second is unreachable. (sourcepoint.rs:437)
• Head injector config property names: Hardcoded Sourcepoint config keys could go stale. A test asserting known property names appear in the generated script would catch omissions. (sourcepoint.rs:467)

🌱 seedling

• No body size limit: The body is read entirely into memory with no upper bound. A Content-Length check would prevent OOM from unexpected large responses. (sourcepoint.rs:400)
• Missing validation test for invalid cdn_origin: No test proves that register() fails when cdn_origin is not a valid URL. (sourcepoint.rs:153)

⛏ nitpick

• Manual [sourcepoint] log prefix: Other integrations rely on the log crate's module path rather than a manual bracketed prefix. (sourcepoint.rs:355)

CI Status

• cargo fmt: PASS
• cargo test: PASS
• vitest: PASS
• format-typescript: FAIL (Prettier issue in script_guard.test.ts)
• format-docs: PASS
• browser integration tests: PASS
• integration tests: PASS
• CodeQL: FAIL (likely infra)

[prk-Jr]

Reviewed commit a211eb0. Three blocking issues require fixes before merge; the rest are non-blocking.

---

Blocking (must fix before merge)

B1 — CI: format-typescript still failing on current HEAD

The latest commit (a211eb02) re-introduced a Prettier violation in crates/js/lib/test/integrations/sourcepoint/script_guard.test.ts. CI's format-typescript check will fail on this commit. Fix with:

cd crates/js/lib && npm run format -- --write

then commit the result.

B2 — SSRF via cdn_origin (inline comment on line 96)

See inline comment. A syntactically valid but host-unrestricted cdn_origin allows proxying to any IP including 169.254.169.254 (cloud metadata). Needs a custom host validator.

B3 — PUT/PATCH unreachable (inline comment on line 337)

routes() only registers GET and POST; the PUT/PATCH arms in handle() are dead code. Either register the methods or remove the arms.

---

Important (non-blocking)

I1 — Single-quoted '/unified/' missed by SP_ORIGIN_UNIFIED_PATTERN (inline on line 81) — fix is a one-character change to the regex character class.

I2 — BackendConfig::from_url per request (inline on line 357) — pre-compute in new() following the pattern in aps.rs / prebid.rs.

I3 — PR checklist says "Uses tracing macros" but codebase uses log crate

The checklist item reads "Uses tracing macros (not println!)" but CLAUDE.md specifies log. The source correctly uses log::info! throughout. The checklist entry should read "Uses log macros (not println!)" to avoid confusion for future contributors.

I4 — JS-rewrite path drops all upstream headers (inline on line 403) — Response::new() discards Vary, CORS, and security headers. Start from response.clone_without_body() instead.

I5 — No test for build_target_url with a path-bearing cdn_origin (inline on line 337) — set_path silently drops any path prefix in cdn_origin. Should be tested and either fixed or documented as an explicit constraint.

---

Suggestions (non-blocking)

S1 — URL normalisation logic duplicated across Rust and TypeScript

parse_sourcepoint_url in sourcepoint.rs and normalizeSourcepointUrl in script_guard.ts implement identical protocol-relative URL normalisation in parallel. Not a defect, but a cross-reference comment in each file would prevent future fixes being applied to only one side.

S2 — Redundant rewrite_sdk guard in rewrite() (inline on line 437) — unreachable because handles_attribute already gates on it. Either remove with a comment or document the intentional belt-and-suspenders.

S3 — No test for single-quoted CDN URLs (inline on line 81)

S4 — Unchecked WASM build in PR checklist

The test plan has an unchecked - [ ] WASM build item, but CI's WASM build passes. The checkbox should be checked to accurately reflect build status.

S5 — register missing # Examples doc section (inline on line 314) — CLAUDE.md requires this for all public API functions.

[ChristianPavilonis]

Review feedback addressed

All blocking and important findings have been fixed across three commits. Replies posted on each inline thread.

Blocking (fixed)

• SSRF guard — Custom validate_cdn_origin() restricts host to *.privacy-mgmt.com
• PUT/PATCH dead code — Removed unreachable arms
• Prettier — Formatting fixed

Important non-blocking (fixed)

• UTF-8 safety — take_body_bytes() + String::from_utf8() with graceful fallback
• Redirect rewriting — 3xx Location headers containing CDN host are rewritten
• CORS headers — JS-rewrite path now preserves upstream CORS headers
• Single-quote regex — SP_ORIGIN_UNIFIED_PATTERN handles both ' and " quotes
• Log prefix — Normalised to Sourcepoint: style

Suggestions (fixed)

• Removed redundant rewrite_sdk guard, added invariant comment
• Added # Examples to register() docs
• Added validation tests for cdn_origin
• Added test for single-quoted unified path

Intentionally skipped (with rationale in thread replies)

• Regex mismatched quotes — documented limitation, no real-world impact
• BackendConfig::from_url per-request — matches APS/Prebid pattern
• Body size limit — future enhancement
• Config property name test — covered by existing head injector test

All CI checks pass locally: cargo fmt, clippy, cargo test, vitest, Prettier.