Differences from the Web SDK
Android SDK v1.4.0 — Differences from the Web SDK
The Android SDK follows the Web SDK as closely as the platform allows — same event names, same wire contract, same consent semantics. Where they diverge, it's a deliberate consequence of not running inside a browser DOM, or of an iteration boundary for this first release. This table is the authoritative list; every row below is discussed in more depth in the chapter its right-hand column links to.
| Web SDK | Android | Why |
|---|---|---|
page(), SPA history hooks (pushState/popState interception) | screen(name) + automatic Activity tracking (ch. 3, ch. 2) | There's no browser History API on Android. screenTracking = AUTO substitutes ActivityLifecycleCallbacks (one screen() per Activity resume, named after the Activity's simple class name) for SPA route-change detection; screenTracking = OFF is the manual escape hatch for apps whose navigation model (Compose Navigation, single-Activity, Fragments) doesn't line up with Activity boundaries. |
Tracker script activation (data-optima-tracker DOM attribute), managed script/html/page tags | Only the gate (consent.can, consent.changes) and the pixel kind (ch. 4, ch. 3) | A managed script/html tag means "inject and run this JavaScript/markup on the page" — a concept with no equivalent on a native screen. A pixel (a plain GET request) has no such dependency and fires natively; script/html/page-kind rows are recognized but intentionally inert, logged once at debug ("managed script/html tags are not runnable on Android; ignored"). |
web_personalization items | Dropped with one debug line (ch. 5) | Server-side template personalization rewrites DOM content that simply doesn't exist as a concept on a native screen. |
Cookies, cookieDomain config | SharedPreferences only | Android apps don't share a cookie jar with anything; there's no cross-subdomain identity problem for a cookieDomain setting to solve, and nothing analogous to configure. |
Global Privacy Control (navigator.globalPrivacyControl) | None on Android | GPC is a browser-signal standard with no OS-level Android equivalent to read. |
Google Consent Mode dataLayer reader (automatic gtag/dataLayer observation) | Manual setGCM only; TCF is read automatically from IABTCF_* prefs (ch. 4) | Google Consent Mode is a dataLayer/gtag.js convention specific to web pages; there's no equivalent runtime object to observe on Android. TCF, by contrast, has a real cross-platform convention (IABTCF_TCString in default SharedPreferences on Android, localStorage on web) that every compliant CMP SDK already writes to, so the SDK reads it the same way on both platforms. |
codeMode inline/sandbox, CSP nonces, content:blocked event | htmlMode WEBVIEW/OFF; HTML runs in an isolated WebView (ch. 2, ch. 5) | A browser page executing operator-authored script inline is a same-origin risk that CSP/sandboxing/nonces exist to contain. A native app has no "page origin" for that script to inherit in the first place — isolating it into its own WebView (no shared cookies/storage with the host app) achieves the same containment goal with a platform-native mechanism instead of a web-specific one, so there's no content:blocked event to fire — the isolation is structural, not something a customer's CSP can block. |
forms.render (a full native-DOM form renderer) | Not in this iteration — forms.load (headless) + the hosted page in a Custom Tab (ch. 6) | Rendering an arbitrary field-driven form natively, matching a design system you don't control, is a materially larger surface than the rest of this SDK's UI (which mirrors a small, fixed set of content-item shapes). Headless loading plus a Custom Tab link to the same hosted page the Web SDK's own non-JS fallback uses covers every form today without committing to a native form-widget API prematurely; a native renderer is a plausible follow-up once real usage informs what it should look like. |
| — | push.* (setToken, parse, trackOpen) | Mobile push has no web equivalent in this SDK's scope — it exists purely because mobile apps need it, not because it mirrors something the Web SDK does differently. See chapter 7. |
What's identical
Everything not in the table above is the same on purpose: track/identify/
reset/deviceId/flush/the event names and payloads, the consent
omit-vs-[] declaration rule, the outbox batching numbers (20 hits / 2s
coalesce), the retry backoff ladder (1s/2s/5s/15s/60s), the display rule
schema (trigger/frequency/dismissible/ttl_days/placement), campaign
mobile_inapp impression/click/dismiss feedback, the tracker-gate
consent.can pattern, and the wire contract itself
(docs/oapi3_edge.yml/contract/oapi3_edge.yml) — a change on one side of
that contract is a change on both. If you've integrated the Web SDK before,
almost everything here should feel immediately familiar; this table is
deliberately the complete list of what doesn't.