In this case, you can use `dict`:

{* ../../docs_src/extra_models/tutorial005_py310.py hl[6] *}

## Recap { #recap }

Use multiple Pydantic models and inherit freely for each case.

              public ImmutableList<AclEntry> value() {
                return acl;
              }
            };
        return () -> attribute;
      } catch (IOException e) {
        // We throw a new exception each time so that the stack trace is right.
        return () -> {
          throw new IOException("Could not find user", e);
        };
      }
    }

    private static String getUsername() {
      /*

* `C:\Program Files\Python312\Scripts`
* `C:\Program Files\Python312`
* `C:\Windows\System32`

////

When you type a **command** in the terminal, the operating system **looks for** the program in **each of those directories** listed in the `PATH` environment variable.

For example, when you type `python` in the terminal, the operating system looks for a program called `python` in the **first directory** in that list.

// rows in some manner. Requires all input rows to be processed,
// before a result is returned.
//
// Row function - An expression that depends on a value in the
// row. They have an output for each input row.
//
// Some types of a queries are not valid. For example, an aggregation
// function combined with a row function is meaningless ("AVG(s.Age) +
// s.Salary"). Analysis determines if such a scenario exists so an

   * because underutilization = free resources for the taking. If we are primarily interested to
   * deal with overflow, then stored permits could be given out /slower/ than fresh ones. Thus, we
   * require a (different in each case) function that translates storedPermits to throttling time.
   *
   * This role is played by storedPermitsToWaitTime(double storedPermits, double permitsToTake). The

We will create a FastAPI **dependency** with `yield` that will provide a new `Session` for each request. This is what ensures that we use a single session per request. 🤓

Then we create an `Annotated` dependency `SessionDep` to simplify the rest of the code that will use this dependency.

      b.append(',').append(' ');
    }
  }

  /**
   * Saves the state to a stream (that is, serializes it).
   *
   * @serialData The length of the array is emitted (int), followed by all of its elements (each a
   *     {@code double}) in the proper order.
   */
  private void writeObject(ObjectOutputStream s) throws IOException {
    s.defaultWriteObject();

    // Write out array length
    int length = length();

   * plus the given elements, according to {@link #expectContents(java.util.Collection)}. In other
   * words, for the default {@code expectContents()} implementation, the number of occurrences of
   * each given element has increased by one since the test collection was created, and the number
   * of occurrences of all other elements has not changed.
   *

And whenever you output that data, even if the source had duplicates, it will be output as a set of unique items.

And it will be annotated / documented accordingly too.

## Nested Models { #nested-models }

Each attribute of a Pydantic model has a type.

But that type can itself be another Pydantic model.

So, you can declare deeply nested JSON "objects" with specific attribute names, types and validations.

      FeatureSpecificTestSuiteBuilder<
              ?, ? extends OneSizeTestContainerGenerator<Map<K, V>, Entry<K, V>>>
          parentBuilder) {
    // TODO: Once invariant support is added, supply invariants to each of the
    // derived suites, to check that mutations to the derived collections are
    // reflected in the underlying map.

    List<TestSuite> derivedSuites = super.createDerivedSuites(parentBuilder);