default_visibility = ["//visibility:private"],
    licenses = ["notice"],
)

# Filesystem implementation for POSIX environments: Linux, MacOS, Android, etc.
tf_cc_shared_object(
    name = "libposix_filesystem.so",
    framework_so = [],
    linkstatic = False,
    visibility = ["//visibility:public"],
    deps = [":posix_filesystem_impl"],
)

# The real implementation of the filesystem.
cc_library(

It can't handle nested models very well. So, if the JSON body in the request is a JSON object that has inner fields that in turn are nested JSON objects, it cannot be properly documented and validated.

!!! check "Inspired **FastAPI** to"

# Libtensorflow archives) to GCS buckets. TFCI Mac VMs need to authenticate as
# a service account that has the right permissions to be able to do so.
set +x
if [[ -n "${GOOGLE_APPLICATION_CREDENTIALS:-}" ]]; then
  # Python 3.12 removed the module `imp` which is needed by gcloud CLI so we set
  # `CLOUDSDK_PYTHON` to Python 3.11 which is the system Python on TFCI Mac
  # VMs.
  export CLOUDSDK_PYTHON=$(which python3.11)

The same models are shared among requests, so, it's not one model per request, or one per user or something similar.

Let's imagine that loading the model can **take quite some time**, because it has to read a lot of **data from disk**. So you don't want to do it for every request.

  }

  public void testRoundToDouble_maxPreciselyRepresentablePlusOne() {
    double twoToThe53 = Math.pow(2, 53);
    // the representable doubles are 2^53 and 2^53 + 2.
    // 2^53+1 is halfway between, so HALF_UP will go up and HALF_DOWN will go down.
    new RoundToDoubleTester(BigDecimal.valueOf((1L << 53) + 1))
        .setExpectation(twoToThe53, DOWN, FLOOR, HALF_DOWN, HALF_EVEN)

Each section gradually builds on the previous ones, but it's structured to separate topics, so that you can go directly to any specific one to solve your specific API needs.

It is also built to work as a future reference.

So you can come back and see exactly what you need.

## Run the code

All the code blocks can be copied and used directly (they are actually tested Python files).

* There's a subdirectory `app/routers/` with another file `__init__.py`, so it's a "Python subpackage": `app.routers`.
* The file `app/routers/items.py` is inside a package, `app/routers/`, so, it's a submodule: `app.routers.items`.
* The same with `app/routers/users.py`, it's another submodule: `app.routers.users`.
* There's also a subdirectory `app/internal/` with another file `__init__.py`, so it's another "Python subpackage": `app.internal`.

            return dstIndex - start;
        }

        /* The message provided to batch has a batchLimit that is
         * higher than the current batchLevel so we will now encode
         * that chained message. Before doing so we must increment
         * the batchLevel of the andx message in case it itself is an
         * andx message and needs to perform the same check as above.
         */

  protected void setUp() throws Exception {
    // Load the "normal" copy of SettableFuture and related classes.
    SettableFuture<?> unused = SettableFuture.create();
    // Hack to load AbstractFuture et. al. in a new classloader so that it tries to re-read the
    // cancellation-cause system property. This allows us to test what happens if reading the
    // property is forbidden and then continue running tests normally in one jvm without resorting

      // the operation would be a no-op, so exceptions are allowed but not required
    }
    expectUnchanged();
  }

  @MapFeature.Require(absent = SUPPORTS_PUT)
  public void testReplaceEntry_unsupportedAbsentKey() {
    try {
      getMap().replace(k3(), v3(), v4());
    } catch (UnsupportedOperationException tolerated) {
      // the operation would be a no-op, so exceptions are allowed but not required
    }