// It turns out that creating a new String is usually as fast, or faster
    // than wrapping cbuf in a light-weight CharSequence.
    target.append(new String(cbuf, off, len));
  }

  /*
   * Override a few functions for performance reasons to avoid creating unnecessary strings.
   */

  @Override
  public void write(int c) throws IOException {
    checkNotClosed();
    target.append((char) c);
  }

  @Override

### Dependencies with `yield` and `scope` { #dependencies-with-yield-and-scope }

In version 0.121.0, FastAPI added support for `Depends(scope="function")` for dependencies with `yield`.

Using `Depends(scope="function")`, the exit code after `yield` is executed right after the *path operation function* is finished, before the response is sent back to the client.

	if f.Ref == nil {
		f.Ref = make([]*Ref, 0, 8)
	}
	f.walk(ast2, ctxProg, (*File).validateIdents)
	f.walk(ast2, ctxProg, (*File).saveExprs)

	// Accumulate exported functions.
	// The comments are only on ast1 but we need to
	// save the function bodies from ast2.
	// The first walk fills in ExpFunc, and the
	// second walk changes the entries to
	// refer to ast2 instead.
	f.walk(ast1, ctxProg, (*File).saveExport)

#define TENSORFLOW_C_EAGER_GRADIENTS_INTERNAL_H_

#include "tensorflow/c/eager/gradients.h"

namespace tensorflow {
namespace gradients {
namespace internal {

// Helper functions which delegate to `AbstractOperation`, update
// the state of the ForwardOperation and call the tape as appropriate.
// These APIs are mainly to facilitate testing and are subject to change.

}

// negate negates a numeric value
func (v *Value) negate() {
	switch x := v.value.(type) {
	case float64:
		v.value = -x
	case int64:
		v.value = -x
	}
}

// Value comparison functions: we do not expose them outside the
// module. Logical operators "<", ">", ">=", "<=" work on strings and
// numbers. Equality operators "=", "!=" work on strings,
// numbers and booleans.

*   `tf.function`:

    *   `tf.function` now uses the Python inspect library directly for parsing the signature of the Python function it is decorated on. This change may break code where the function signature is malformed, but was ignored previously, such as:
        *   Using `functools.wraps` on a function with different signature

It is designed to have functions that receive two parameters, one "request" and one "response". Then you "read" parts from the request, and "write" parts to the response. Because of this design, it is not possible to declare request parameters and bodies with standard Python type hints as function parameters.

// GraphDef type. These functions must be deleted by calling TF_DeleteFunction.
//
// If `mutate_proto_func` is non-NULL, run it over each FunctionDef proto,
// before creating a TF_Function out of the possibly mutated proto.
static std::vector<UniqueFuncPtr> CreateFunctionsFromTextProto(
    const char* text_proto,
    std::function<void(FunctionDef*)>* mutate_proto_func, TF_Status* status) {

//     -   ObjectKey := DAREv2_Dec(KeyEncKey, SealedKey)
//     - object_data := DAREv2_Dec(ObjectKey, enc_object_data)
//     Output: object_data
//
// ### SSE-S3 and KMS
//
// SSE-S3 requires that the KMS provides two functions:
//
//  1. Generate(KeyID) -> (Key, EncKey)
//
//  2. Unseal(KeyID, EncKey) -> Key
//
//  1. Encrypt:
//     Input: KeyID, bucket, object, metadata, object_data
//     -     Key, EncKey := Generate(KeyID)

    (For example, a function in a library to compute square roots has
    a purpose that is entirely well-defined independent of the
    application.  Therefore, Subsection 2d requires that any
    application-supplied function or table used by this function must
    be optional: if the application does not supply it, the square
    root function must still compute square roots.)