How do you parameterise tests with TestNG data providers?

Data providers turn a single test method into N parametrised invocations — each row of the returned 2D array becomes the arguments for one run. They're TestNG's answer to JUnit 5's @ParameterizedTest.

The basic shape:

public class LoginTest {
    @DataProvider(name = "loginCases")
    public Object[][] loginCases() {
        return new Object[][] {
            { "valid@example.com", "secret123",  true,  null },
            { "wrong@example.com", "secret123",  false, "User not found" },
            { "valid@example.com", "wrongpass",  false, "Incorrect password" },
            { "",                  "secret123",  false, "Email is required" },
        };
    }

    @Test(dataProvider = "loginCases")
    public void login(String email, String password,
                      boolean shouldSucceed, String expectedError) {
        new LoginPage(driver).loginAs(email, password);
        if (shouldSucceed) {
            assertTrue(new DashboardPage(driver).isLoaded());
        } else {
            assertEquals(new LoginPage(driver).errorText(), expectedError);
        }
    }
}

Reporting: TestNG names each invocation (login[0], login[1], …) so failures show which row failed. You can override naming with ITestContext for human-readable cases.

Sources for data providers:

• Inline arrays — fine for small fixed sets.
• CSV / Excel / JSON files — read in the data provider method. Useful for QA-owned test data.
• External classes — @DataProvider(name = "users", class = TestData.class).
• Database fetches — pull canonical test cases from a shared store.

Refinements:

• parallel = true runs the rows concurrently — combined with thread-safe page objects, you get test-level parallelism for free.
• A data provider can return Iterator<Object[]> to lazily generate rows for very large sets.

Common mistake: making the data provider too granular. If the test has 200 rows, ask whether you really want 200 separate test reports — often a few well-chosen equivalence-partition rows are clearer than exhaustive data.