Download the PHP package univapay/univapay-sdk-compat without Composer

univapay/univapay-sdk-compat

日本語版はこちら

A runtime compatibility layer that reimplements the public surface of the legacy, hand-written univapay/php-sdk — the same class names, method signatures, public properties, enum style, exceptions, and polling behavior — on top of univapay/client-sdk (the new, APIMatic-generated transport engine) as its transport. It exists so that a codebase migrated by univapay/univapay-sdk-migrate keeps working, unmodified, on the new engine.

What this is

univapay/php-sdk will not be updated again. Every new API feature (v1.1 and later) lands only in univapay/client-sdk. This package lets existing integrators reach that new engine without rewriting call sites: it is a drop-in, namespace-only swap — Univapay\* → Univapay\Compat\* — that keeps every construct your code already uses (Money\Money values, ChargeStatus::SUCCESSFUL() identity comparisons, public property access, awaitResult()/chained calls, catch blocks) behaving exactly as it did before. Internally, every compat method builds a typed request against univapay/client-sdk and sends it through that engine's real HTTP transport — nothing here talks to the API directly. See Architecture for how responses are then hydrated back into the old SDK's shapes.

You do not install this package by hand in the normal case — see Install.

Install

Normal path: run the migration tool. univapay/univapay-sdk-migrate requires this package, rewrites your code's imports to point at it, and removes univapay/php-sdk, all in one command:

See the univapay-sdk-migrate README for what that command does, its flags, and its report format.

Manual path. If you are not using the migration tool — a fresh integration deliberately targeting the old SDK's API shape, or a codebase already rewritten by hand — install directly:

The two packages can be installed side by side — their autoload roots (Univapay\ vs. Univapay\Compat\) don't collide, and this package never references an old-SDK class directly. univapay-sdk-migrate relies on that: it requires this package before removing univapay/php-sdk, so both are present for one step while Rector still needs the old SDK's classes loadable for receiver-type resolution. Once your migration is done, remove univapay/php-sdk — there's no reason to keep it installed, but nothing breaks if it lingers alongside this package. Requires PHP >=7.2 (matching univapay/client-sdk's own floor) and moneyphp/money ^3.3 || ^4.0.

Supported surface matrix

Almost everything the old SDK exposed works. A small, fixed set of methods compile and are reachable but throw Errors\UnivapayUnsupportedFeatureError at call time, because the new engine has no equivalent API to call through to:

"Permanent throw" means feature-frozen: these will not gain support in a future compat release. They are reachable only through the native SDK (see Migrating off the compat layer), and the migration tool flags every call site that reaches one so it's a reviewable line instead of a runtime surprise.

Behavior deltas

Compat is not a byte-for-byte replay of the old SDK — a small number of documented, deliberate differences exist.

Beyond those, several other differences are deliberate:

• UnivapayNetworkError replaces the old WpOrg\Requests\Exception retry target. A genuine transport failure (DNS, connection refused, timeout before any response) surfaces from the new engine as ApiException with getCode() === 0. The old SDK's NetworkRetryHandler matched on WpOrg\Requests\Exception, a class that never appears on this transport — so that retry path was silently dead. Compat's NetworkRetryHandler targets Errors\UnivapayNetworkError instead, which Support\ExceptionMapper raises specifically for this case (not UnivapayServerError, which would mislabel a network failure as a 5xx). The migration tool flags any consumer code that still catches WpOrg\Requests\Exception for manual review.
• 10-second timeout, matching what integrators have always experienced. The old SDK's transport (rmccue/requests) defaulted to 10s and never exposed the knob. The new engine's own default is 30s; Support\Bridge pins it back to 10 so nothing that depended on that ceiling changes behavior.
• Retry-safe idempotency. The old SDK generated one idempotency key per logical call and reused it on every retry within that call. The new engine's IdempotencyCallback mints a fresh key per HTTP request by default — combined with the default retry cascade (rate-limit + network retries, up to 4 attempts), a timed-out-but-actually-processed POST /charges could create up to 4 real charges before this fix. Support\ApiCaller generates one key per logical call, outside the retry loop, and passes it explicitly on every attempt — exactly what the old SDK did.
• Error mapping goes through ApiResponse::isError(), not a caught exception. Every generated API method's response handler is configured to return an error response rather than throw one — a 4xx/5xx from univapay/client-sdk comes back as a plain, non-throwing ApiResponse whose isError() is true, not an exception. Support\ApiCaller checks for that on every call and maps it via Support\ExceptionMapper into the same Errors\* hierarchy the old SDK exposed (UnivapayNotFoundError, UnivapayRequestError, etc.) — including the old SDK's own quirk that 404 responses carry no decoded error body (only 400/401/403 do). A genuine transport failure (no HTTP response at all) is the one case that does still throw, and is mapped separately into UnivapayNetworkError above. See docs/ARCHITECTURE.md for the full mechanism and why it matters.

Webhook notes

UnivapayClient::parseWebhookData() reproduces the old SDK's dispatch and its corner cases verbatim — including ones that look like bugs but are pinned, intentional behavior:

• Transfer events hydrate; everything else about Transfer still throws. transfer_created/transfer_updated/transfer_finalized webhook payloads hydrate a real Resources\Transfer object regardless of the fact that Transfer itself is unsupported for direct API access (see the surface matrix above) — the webhook channel keeps delivering that data independent of whether this transport engine exposes a Transfers API. Any subsequent call on that object — fetch(), update(), listLedgers(), listStatusChanges() — throws UnivapayUnsupportedFeatureError. Parsing the payload does not make the resource supported.
• Three current token event types have no compat enum case. The live API's TokenEvent discriminator now includes token_three_d_s_updated, token_cvv_auth_check_updated, and token_replaced — additions made after the old SDK's Enums\WebhookEvent was last updated, so none of the three exist in compat's ported version of that enum either (it mirrors the old SDK, not the current spec). A webhook delivery carrying one of these three event types will raise Errors\UnivapayUnknownWebhookEvent, exactly as any other unrecognized event string would. If your integration needs these events, handle them via native() (see below) instead of parseWebhookData().
• A merchant-level app token receiving a store-scoped event gets UnivapayInvalidWebhookData, not a clearer error. TOKEN_*, REFUND_FINISHED, and CANCEL_FINISHED events require a store-scoped JWT, exactly as the old SDK's context lookups did; the guard that enforces this fires inside the same try block that a broad catch funnels into UnivapayInvalidWebhookData — so that's what a merchant-JWT client sees, not a more specific "wrong token type" error. This is reproduced exactly, not cleaned up, because the old SDK's own behavior here is what any existing integration has already coded around.
• customs_declaration_finished has an enum case but no parser — it also maps to UnivapayInvalidWebhookData. The event type is recognized (it doesn't raise UnivapayUnknownWebhookEvent), but there has never been a resource type for it to hydrate into, in the old SDK or here.

Architecture

Requests are built typed, against univapay/client-sdk's own generated models — every field your code sets goes through the same validation and serialization the native SDK would use. Responses, by contrast, are hydrated from the raw captured wire body through the old SDK's own ported JSON schema parsers, not through the generated SDK's typed response models — the only way to guarantee wire-for-wire parity with what the old SDK's battle-tested parsers already handled (including shapes the current spec doesn't describe yet). See docs/ARCHITECTURE.md for the full request/response diagram and the confinement boundary that keeps raw-body access contained to a reviewed allowlist of files.

Migrating off the compat layer

Compat is not meant to live forever. UnivapayClient::native() returns the exact UnivaPay\UnivapayClientSdkClient instance this client already built internally to make its own calls — same auth, base URL, and 10-second timeout as the compat surface, never a second, separately configured client:

This enables mixed mode: migrate call sites file by file, rewriting each one against native()'s typed API while everything not yet touched keeps calling the compat facade exactly as before. Both paths share one engine, so there is no drift between them during the migration window — a charge created through native() is visible to code still reading it through compat, and vice versa.

A full construct-by-construct migration reference — Money → int + currency string, ChargeStatus::SUCCESSFUL() identity comparisons → string constants, $charge->status → typed getters, awaitResult() → pollCharge(), paginated lists → cursor parameters, parseWebhookData() → typed webhook handler classes, and more, each with a before/after snippet — is tracked in the portal guide's Phase 2 section, not duplicated here.

Deprecation notices

Opt-in runtime signal that helps a team find its own remaining compat call sites before doing the phase-2 migration onto native() above.

• UnivapayClientOptions::$deprecationNotices — bool, defaults to false. Off is zero overhead: no backtrace, no bookkeeping, just one boolean check.
• On, every public compat method (UnivapayClient methods, resource methods — fetch()/patch()/capture()/cancel()/awaitResult()/createRefund()/etc. — list methods, parseWebhookData()) triggers one trigger_error(..., E_USER_DEPRECATED) the first time it's reached from a given line of your code; calling that same line again stays silent, a different line notifies again. Each message names the compat method and its native-SDK equivalent.
• No cascade. A compat method that calls other compat methods internally — createCharge()'s own token-fetch-then-create flow, for instance — emits only the one notice for the call your code actually made, not one per internal step.
• native() never notifies. It's the escape hatch this feature is steering you towards, not a call site to flag.
• Turn it on in a staging/dev environment, exercise real traffic, and the notices surface exactly the call sites univapay-sdk-migrate's --phase2 Rector set will also want you to look at — the two are meant to be used together, not as alternatives to each other.

Versioning and sunset policy

Compat is feature-frozen as of 1.0: every new API capability lands in univapay/client-sdk only, reachable through native(). Compat itself continues to receive bugfixes and follows the engine SDK's own version bumps, for as long as integrators still depend on the old SDK's surface — there is no forced end-of-life date. That asymmetry (old surface stays exactly as it is; new capability only exists on the other side of native()) is the intended, gradual pressure to migrate, not a cliff.

License

MIT.