// A TF_ExecutionContext stores knowledge about how to execute an operation.
// E.g. it could know whether we're in eager mode or graph mode, keeps track
// of gradient tapes, etc.
typedef struct TF_ExecutionContext TF_ExecutionContext;

// A TF_AbstractTensor is an input to an operation. E.g. it could be a union
// type of eager and graph tensors. It is also the result of executing an
// operation.

#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_operation.h"

// Wraps a pointer to an operation implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying operation object. Instead, call
// TFE_DeleteOp who calls Release() on the operation pointer and deletes
// the TFE_Op structure.

#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_tensor_handle.h"

// Wraps a pointer to a tensor handle implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying handle object. Instead, call
// TFE_DeleteTensorHandle who calls Release() on the handle pointer and deletes
// the TFE_TensorHandle structure.

//
// Otherwise, while traversing the tree, undeleted_file_count and
// undeleted_dir_count are updated if an entry of the corresponding type could
// not be deleted. The returned error status represents the reason that any one
// of these entries could not be deleted.
//
// Typical status codes:
//  * TF_OK - dirname exists and we were able to delete everything underneath
//  * TF_NOT_FOUND - dirname doesn't exist

              delete reinterpret_cast<DataType*>(data);
            },
            nullptr),
        TF_DeleteTensor);
    // TODO(allenl): Here and when executing regular operations, we could hold
    // on to one TFE_Op per device and just call TFE_ResetOp to avoid parsing
    // device names repeatedly.
    std::unique_ptr<TFE_Op, decltype(&TFE_DeleteOp)> const_op(

// APIs for generically dealing with op attributes (e.g. when forwarding them
// through custom device implementations).
//
// TODO(allenl): Currently these are black boxes, but we should have some way to
// inspect values. This would let people e.g. copy over most attributes and then
// modify some based on their values.

// A reference to an op's name -> attribute mapping
typedef struct TFE_OpAttrs TFE_OpAttrs;

// least `max_values`.
//
// The elements of values will point to addresses in `storage` which must be at
// least `storage_size` bytes in length.  Ideally, max_values would be set to
// TF_AttrMetadata.list_size and `storage` would be at least
// TF_AttrMetadata.total_size, obtained from TF_OperationGetAttrMetadata(oper,
// attr_name).
//
// Fails if storage_size is too small to hold the requested number of strings.

#include "tensorflow/c/conversion_macros.h"
#include "tensorflow/c/eager/immediate_execution_context.h"

// Wraps a pointer to a context implementation.
//
// WARNING: Since the underlying object could be ref-counted a user of this
// interface cannot destruct the underlying context object. Instead, call
// TFE_DeleteContext who calls Release() on the context pointer and deletes
// the TFE_Context structure.

  //
  // If false, a caller might implement pretty-printing by resolving and
  // iterating over the resulting tensor. This may still be viable if resolving
  // the handle loses information, but `SummarizeValue` would be more precise.
  virtual bool PreferCustomSummarizer() const { return false; }

  // Returns a string which summarizes the value of this TensorHandle, for
  // debugging. Does not include a shape or dtype.
  //

    // can do its work.
    //
    // Rather than re-entering and delegating Accumulate like this, we could
    // instead allow ForwardAccumulator some control over the current tape set
    // (so it can deactivate itself and activate its GradientTape). Currently
    // that is managed by the language binding and would require relatively
    // messy callbacks.
    call_state_.top().backward_tape->RecordOperation(