environ_cp['TF_NEED_CUDA'] = '0'
  else:
    environ_cp['TF_NEED_CUDA'] = str(
        int(get_var(environ_cp, 'TF_NEED_CUDA', 'CUDA', False)))
  if (environ_cp.get('TF_NEED_CUDA') == '1' and
      'TF_CUDA_CONFIG_REPO' not in environ_cp):

    set_action_env_var(
        environ_cp,
        'TF_NEED_TENSORRT',
        'TensorRT',
        False,
        bazel_config_name='tensorrt')

This can be very useful for setting up **resources** that you need to use for the whole app, and that are **shared** among requests, and/or that you need to **clean up** afterwards. For example, a database connection pool, or loading a shared machine learning model.

## Use Case

Doing this provides the following specific benefits:
- **Reduce the API surface** - We no longer need to maintain two methods.
- **Consistency** - All languages have consistent access to the same APIs and ergonomics in the DSL.
- **Reduce the size of the Gradle distribution** - We no longer need to carry multiple standard libraries for different languages.

## Decision

### Short

It has sensible **defaults** for everything, with optional configurations everywhere. All the parameters can be fine-tuned to do what you need and to define the API you need.

But by default, it all **"just works"**.

### Validation

* Validation for most (or all?) Python **data types**, including:
    * JSON objects (`dict`).

    // it might need this accessor's bytecode.
    //
    // If an accessor function is defined in a different module,
    // IDE tries to acquire its bytecode via the index, however,
    // the index doesn't cover classfiles from compiler output,
    // so the lowering fails.
    //
    // To solve the problem, we need to find all delegated properties with inline accessors

/**
 *
 * @since 4.0.0
 */
@Experimental
public interface LocalRepositoryManager extends Service {

    /**
     * Gets the relative path for a locally installed artifact.
     * Note that the artifact need not actually exist yet at
     * the returned location, the path merely indicates where
     * the artifact would eventually be stored.
     *
     * @param session The session to use, must not be {@code null}.

Use the `TestClient` object the same way as you do with `httpx`.

Write simple `assert` statements with the standard Python expressions that you need to check (again, standard `pytest`).

```Python hl_lines="2  12  15-18"
{!../../../docs_src/app_testing/tutorial001.py!}
```

!!! tip
    Notice that the testing functions are normal `def`, not `async def`.

    implementation(project(":analysis:analysis-api"))

    implementation(commonDependency("org.jetbrains.kotlin:kotlin-reflect")) { isTransitive = false }

    /*
     We do not need guava in the generator, but because of a bug in the IJ project importing, we need to have a dependency on intellijCore
     the same as it is in `:fir:tree:tree-generator` module to the project be imported correctly
     */
    compileOnly(intellijCore())

import org.apache.maven.plugin.descriptor.MojoDescriptor;

// TODO lifecycles being executed
// TODO what runs in each phase
// TODO plugins that need downloading
// TODO project dependencies that need downloading
// TODO unfortunately the plugins need to be downloaded in order to get the plugin.xml file. need to externalize this
//      from the plugin archive.
// TODO this will be the class that people get in IDEs to modify

/**

# to build the repo. The only dependencies in this mode are to have make and
# docker. If you'd rather build with a local tool chain instead, you'll need to
# figure out all the tools you need in your environment to make that work.
export BUILD_WITH_CONTAINER ?= 0

ifeq ($(BUILD_WITH_CONTAINER),1)

# An export free of arguments in a Makefile places all variables in the Makefile into the