* is precisely representable as a {@code double}, its {@code double} value will be returned;
   * otherwise, the rounding will choose between the two nearest representable values with {@code
   * mode}.
   *
   * <p>For the case of {@link RoundingMode#HALF_EVEN}, this implementation uses the IEEE 754

    assertThrows(IllegalArgumentException.class, () -> RateLimiter.create(1.0, -1, NANOSECONDS));
  }

  @AndroidIncompatible // difference in String.format rounding?
  public void testWarmUp() {
    RateLimiter limiter = RateLimiter.create(2.0, 4000, MILLISECONDS, 3.0, stopwatch);
    for (int i = 0; i < 8; i++) {
      limiter.acquire(); // #1
    }

    assertThrows(IllegalArgumentException.class, () -> RateLimiter.create(1.0, -1, NANOSECONDS));
  }

  @AndroidIncompatible // difference in String.format rounding?
  public void testWarmUp() {
    RateLimiter limiter = RateLimiter.create(2.0, 4000, MILLISECONDS, 3.0, stopwatch);
    for (int i = 0; i < 8; i++) {
      limiter.acquire(); // #1
    }

        assertEquals(LongMath.pow(i, exp), valueOf(i).pow(exp).longValue());
      }
    }
  }

  @J2ktIncompatible // J2kt BigDecimal.divide also has the rounding bug
  @GwtIncompatible // TODO
  @AndroidIncompatible // TODO(cpovirk): File BigDecimal.divide() rounding bug.
  public void testDivNonZero() {
    for (long p : NONZERO_LONG_CANDIDATES) {
      for (long q : NONZERO_LONG_CANDIDATES) {

        for (RoundingMode mode : ALL_SAFE_ROUNDING_MODES) {
          // Skip some tests that fail due to GWT's non-compliant int implementation.
          // TODO(cpovirk): does this test fail for only some rounding modes or for all?
          if (p == -2147483648 && q == -1 && intsCanGoOutOfRange()) {
            continue;
          }
          int expected =

        model is called for the first time.
*   `tf.linalg`:
    *   Add `CompositeTensor` as a base class to `LinearOperator`.
*   `tf.lite`:
    *   Fix mean op reference quantization rounding issue.
    *   Added `framework_stable` BUILD target, which links in only the
        non-experimental TF Lite APIs.
    *   Remove deprecated Java `Interpreter` methods:

    private final CountDownLatch stopLatch = new CountDownLatch(1);

    private volatile boolean running = false;

    @Override
    public @Nullable Void call() throws InterruptedException {
      running = true;
      startLatch.countDown();
      stopLatch.await();
      running = false;
      return null;
    }

    void waitForStart() throws InterruptedException {

If you need to have two independent FastAPI applications, with their own independent OpenAPI and their own docs UIs, you can have a main app and "mount" one (or more) sub-application(s).

## Mounting a **FastAPI** application { #mounting-a-fastapi-application }

"Mounting" means adding a completely "independent" application in a specific path, that then takes care of handling everything under that path, with the _path operations_ declared in that sub-application.

{* ../../docs_src/behind_a_proxy/tutorial004.py hl[9] *}

and then it won't include it in the OpenAPI schema.

## Mounting a sub-application { #mounting-a-sub-application }

# Benchmarks { #benchmarks }

Independent TechEmpower benchmarks show **FastAPI** applications running under Uvicorn as <a href="https://www.techempower.com/benchmarks/#section=test&runid=7464e520-0dc2-473d-bd34-dbdfd7e85911&hw=ph&test=query&l=zijzen-7" class="external-link" target="_blank">one of the fastest Python frameworks available</a>, only below Starlette and Uvicorn themselves (used internally by FastAPI).