Authorizations, feature control and messages

Feature control answers "can it happen?", authorization answers "may you do it?", and messages tell the user what went wrong. All three are visible to users, all three are easy to break, and all three are often tested only by clicking in CMQ. This lesson moves them into ABAP Unit, with Contoso's rules: a CLOSED agreement cannot be updated, deleted or closed again; only users authorized for company code 1910 may change the Alternadores del Norte agreements; every error tells the user which field to fix.

Testing instance feature control

Kenji's instance feature handler returns if_abap_behv=>fc-o-disabled for update, delete and closeAgreement when Status is CLOSED. There are two ways to test it, and the suite uses both.

1. Ask for permissions. GET PERMISSIONS ENTITY ZR_CMC_CoreAgreement with a request for update, delete and the action, for a CLOSED agreement created in the test. Assert the result reports each as disabled, and as enabled for an OPEN agreement.
2. Attempt the operation. As a consumer (not in local mode), MODIFY ENTITIES ... UPDATE the CLOSED agreement. Assert the key appears in failed and the change was not applied.
3. Unit-test the handler. Call the feature method as a friend, with the buffer double holding one OPEN and one CLOSED agreement, and assert the %features values per key.

Testing authorization

The authorization handler checks Contoso's authorization object for agreements (illustrative name ZCMC_CAGR, with company code and activity) using AUTHORITY-CHECK. A developer running the tests usually has full authorization, so a test that "checks" authorization proves nothing unless it restricts the user. SAP's authority check test helper, CL_AUNIT_AUTHORITY_CHECK, does that: the test gets a controller with get_controller( ), builds an authorization object set, and calls restrict_authorizations_to( ... ). The cheat sheet stresses that the helper only restricts the test user's authorizations; it never grants new ones.

Remember what the advanced RAP course noted: authorization methods are not called for every draft action, and read access is governed separately by CDS access control. Unit tests on doubles do not exercise access control the way real users experience it, so the 1710/1910 read separation is also tested with real test users in CMQ.

Testing messages

A RAP message is an entry in reported with a message object in %msg, an optional field marking in %element, and for drafts a %state_area. For the user, the field marking is what makes an error fixable. Contoso's message tests assert:

• The message class and number, read from the message object, not the text.
• The severity: an error for a rejected save, a warning for a deposit above the usual 45.00 per unit.
• The element: %element-ValidTo = if_abap_behv=>mk-on for validateDates, DepositPerUnit for validateDeposit.
• The pairing: every key in failed has at least one message in reported. A failure with no message leaves the user stuck.

Prechecks and the early rejection

Contoso's create operation has a precheck: a new agreement for a customer and product that already has an OPEN agreement is rejected before it enters the buffer. Prechecks run in the interaction phase, so the test asserts right after MODIFY ENTITIES: the %cid appears in failed with a message, and mapped has no entry for it. Arrange the duplicate through the CDS double (an OPEN agreement for 17100001 already in the data) rather than by creating and committing one first; the test stays fast and does not depend on the save sequence. A second test with a different product proves the precheck does not block legitimate agreements, which is the defect key users notice first.

What goes wrong in testing

• Testing as a superuser. Authorization tests without restrictions always pass.
• Local mode in tests. IN LOCAL MODE and PRIVILEGED skip checks; use them in tests only when that is the point.
• Greyed-out button as proof. A disabled button in the UI does not prove that the API rejects the operation.
• Failed without reported. Kenji's first close action rejected closed agreements silently; only a pairing assertion found it.