def all(*inputs):  # pylint: disable=redefined-builtin
    """Check if value is different than 0 everywhere along the band axis.

    For only one image, this function checks that all bands of the image are True (i.e. !=0) and outputs
    a singleband boolean raster
    For several images, this function checks that all images are True (i.e. !=0) and outputs
    a boolean raster, with as many bands as the inputs

    Args:
        inputs: inputs can be 1) a single image or 2) several images, either passed as separate arguments
                or inside a list

    Returns:
        AND intersection

    """
    # If necessary, flatten inputs
    if len(inputs) == 1 and isinstance(inputs[0], (list, tuple)):
        inputs = inputs[0]

    # Add support for generator inputs (to have the same behavior as built-in `all` function)
    if isinstance(inputs, tuple) and len(inputs) == 1 and inspect.isgenerator(inputs[0]):
        inputs = list(inputs[0])

    # Transforming potential filepaths to pyotb objects
    inputs = [Input(inp) if isinstance(inp, str) else inp for inp in inputs]

    # Checking that all bands of the single image are True
    if len(inputs) == 1:
        inp = inputs[0]
        if isinstance(inp, logicalOperation):
            res = inp[:, :, 0]
        else:
            res = (inp[:, :, 0] != 0)

        for band in range(1, inp.shape[-1]):
            if isinstance(inp, logicalOperation):
                res = res & inp[:, :, band]
            else:
                res = res & (inp[:, :, band] != 0)

    # Checking that all images are True
    else:
        if isinstance(inputs[0], logicalOperation):
            res = inputs[0]
        else:
            res = (inputs[0] != 0)
        for inp in inputs[1:]:
            if isinstance(inp, logicalOperation):
                res = res & inp
            else:
                res = res & (inp != 0)

    return res

def any(*inputs):  # pylint: disable=redefined-builtin
    """Check if value is different than 0 anywhere along the band axis.

    For only one image, this function checks that at least one band of the image is True (i.e. !=0) and outputs
    a single band boolean raster
    For several images, this function checks that at least one of the images is True (i.e. !=0) and outputs
    a boolean raster, with as many bands as the inputs

    Args:
        inputs: inputs can be 1) a single image or 2) several images, either passed as separate arguments
                or inside a list
    Returns:
        OR intersection

    """
    # If necessary, flatten inputs
    if len(inputs) == 1 and isinstance(inputs[0], (list, tuple)):
        inputs = inputs[0]

    # Add support for generator inputs (to have the same behavior as built-in `any` function)
    if isinstance(inputs, tuple) and len(inputs) == 1 and inspect.isgenerator(inputs[0]):
        inputs = list(inputs[0])

    # Transforming potential filepaths to pyotb objects
    inputs = [Input(inp) if isinstance(inp, str) else inp for inp in inputs]

    # Checking that at least one band of the image is True
    if len(inputs) == 1:
        inp = inputs[0]
        if isinstance(inp, logicalOperation):
            res = inp[:, :, 0]
        else:
            res = (inp[:, :, 0] != 0)

        for band in range(1, inp.shape[-1]):
            if isinstance(inp, logicalOperation):
                res = res | inp[:, :, band]
            else:
                res = res | (inp[:, :, band] != 0)

    # Checking that at least one image is True
    else:
        if isinstance(inputs[0], logicalOperation):
            res = inputs[0]
        else:
            res = (inputs[0] != 0)
        for inp in inputs[1:]:
            if isinstance(inp, logicalOperation):
                res = res | inp
            else:
                res = res | (inp != 0)

    return res

def clip(a, a_min, a_max):
    """Clip values of image in a range of values.

    Args:
        a: input raster, can be filepath or any pyotb object
        a_min: minimum value of the range
        a_max: maximum value of the range

    Returns:
        raster whose values are clipped in the range

    """
    if isinstance(a, str):
        a = Input(a)

    res = where(a <= a_min, a_min,
                where(a >= a_max, a_max, a))
    return res

def define_processing_area(*args, window_rule='intersection', pixel_size_rule='minimal', interpolator='nn',
                           reference_window_input=None, reference_pixel_size_input=None):
    """Given several inputs, this function handles the potential resampling and cropping to same extent.

    WARNING: Not fully implemented / tested

    Args:
        *args: list of raster inputs. Can be str (filepath) or pyotb objects
        window_rule: Can be 'intersection', 'union', 'same_as_input', 'specify' (Default value = 'intersection')
        pixel_size_rule: Can be 'minimal', 'maximal', 'same_as_input', 'specify' (Default value = 'minimal')
        interpolator: Can be 'bco', 'nn', 'linear' (Default value = 'nn')
        reference_window_input: Required if window_rule = 'same_as_input' (Default value = None)
        reference_pixel_size_input: Required if pixel_size_rule = 'same_as_input' (Default value = None)

    Returns:
        list of in-memory pyotb objects with all the same resolution, shape and extent

    """
    # Flatten all args into one list
    inputs = []
    for arg in args:
        if isinstance(arg, (list, tuple)):
            inputs.extend(arg)
        else:
            inputs.append(arg)

    # Getting metadatas of inputs
    metadatas = {}
    for inp in inputs:
        if isinstance(inp, str):  # this is for filepaths
            metadata = Input(inp).GetImageMetaData('out')
        elif isinstance(inp, otbObject):
            metadata = inp.GetImageMetaData(inp.output_param)
        else:
            raise TypeError(f"Wrong input : {inp}")
        metadatas[inp] = metadata

    # Get a metadata of an arbitrary image. This is just to compare later with other images
    any_metadata = next(iter(metadatas.values()))

    # Checking if all images have the same projection
    if not all(metadata['ProjectionRef'] == any_metadata['ProjectionRef']
               for metadata in metadatas.values()):
        logger.warning('All images may not have the same CRS, which might cause unpredictable results')

    # Handling different spatial footprints
    # TODO: there seems to have a bug, ImageMetaData is not updated when running an app,
    #  cf https://gitlab.orfeo-toolbox.org/orfeotoolbox/otb/-/issues/2234. Should we use ImageOrigin instead?
    if not all(metadata['UpperLeftCorner'] == any_metadata['UpperLeftCorner']
               and metadata['LowerRightCorner'] == any_metadata['LowerRightCorner']
               for metadata in metadatas.values()):
        # Retrieving the bounding box that will be common for all inputs
        if window_rule == 'intersection':
            # The coordinates depend on the orientation of the axis of projection
            if any_metadata['GeoTransform'][1] >= 0:
                ulx = max(metadata['UpperLeftCorner'][0] for metadata in metadatas.values())
                lrx = min(metadata['LowerRightCorner'][0] for metadata in metadatas.values())
            else:
                ulx = min(metadata['UpperLeftCorner'][0] for metadata in metadatas.values())
                lrx = max(metadata['LowerRightCorner'][0] for metadata in metadatas.values())
            if any_metadata['GeoTransform'][-1] >= 0:
                lry = min(metadata['LowerRightCorner'][1] for metadata in metadatas.values())
                uly = max(metadata['UpperLeftCorner'][1] for metadata in metadatas.values())
            else:
                lry = max(metadata['LowerRightCorner'][1] for metadata in metadatas.values())
                uly = min(metadata['UpperLeftCorner'][1] for metadata in metadatas.values())

        elif window_rule == 'same_as_input':
            ulx = metadatas[reference_window_input]['UpperLeftCorner'][0]
            lrx = metadatas[reference_window_input]['LowerRightCorner'][0]
            lry = metadatas[reference_window_input]['LowerRightCorner'][1]
            uly = metadatas[reference_window_input]['UpperLeftCorner'][1]
        elif window_rule == 'specify':
            pass
            # TODO : it is when the user explicitly specifies the bounding box -> add some arguments in the function
        elif window_rule == 'union':
            pass
            # TODO : it is when the user wants the final bounding box to be the union of all bounding box
            #  It should replace any 'outside' pixel by some NoData -> add `fillvalue` argument in the function

        logger.info('Cropping all images to extent Upper Left (%s, %s), Lower Right (%s, %s)', ulx, uly, lrx, lry)

        # Applying this bounding box to all inputs
        new_inputs = []
        for inp in inputs:
            try:
                params = {
                    'in': inp, 'mode': 'extent', 'mode.extent.unit': 'phy',
                    'mode.extent.ulx': ulx, 'mode.extent.uly': lry,  # bug in OTB <= 7.3 :
                    'mode.extent.lrx': lrx, 'mode.extent.lry': uly,  # ULY/LRY are inverted
                }
                new_input = App('ExtractROI', params)
                # TODO: OTB 7.4 fixes this bug, how to handle different versions of OTB?
                new_inputs.append(new_input)
                # Potentially update the reference inputs for later resampling
                if str(inp) == str(reference_pixel_size_input):  # we use comparison of string because calling '=='
                    # on pyotb objects implicitly calls BandMathX application, which is not desirable
                    reference_pixel_size_input = new_input
            except RuntimeError as e:
                logger.error('Cannot define the processing area for input %s: %s', inp, e)
                raise
        inputs = new_inputs

        # Update metadatas
        metadatas = {input: input.GetImageMetaData('out') for input in inputs}

    # Get a metadata of an arbitrary image. This is just to compare later with other images
    any_metadata = next(iter(metadatas.values()))

    # Handling different pixel sizes
    if not all(metadata['GeoTransform'][1] == any_metadata['GeoTransform'][1]
               and metadata['GeoTransform'][5] == any_metadata['GeoTransform'][5]
               for metadata in metadatas.values()):
        # Retrieving the pixel size that will be common for all inputs
        if pixel_size_rule == 'minimal':
            # selecting the input with the smallest x pixel size
            reference_input = min(metadatas, key=lambda x: metadatas[x]['GeoTransform'][1])
        if pixel_size_rule == 'maximal':
            # selecting the input with the highest x pixel size
            reference_input = max(metadatas, key=lambda x: metadatas[x]['GeoTransform'][1])
        elif pixel_size_rule == 'same_as_input':
            reference_input = reference_pixel_size_input
        elif pixel_size_rule == 'specify':
            pass
            # TODO : when the user explicitly specify the pixel size -> add argument inside the function
        pixel_size = metadatas[reference_input]['GeoTransform'][1]
        logger.info('Resampling all inputs to resolution: %s', pixel_size)

        # Perform resampling on inputs that do not comply with the target pixel size
        new_inputs = []
        for inp in inputs:
            if metadatas[inp]['GeoTransform'][1] != pixel_size:
                superimposed = App('Superimpose', inr=reference_input, inm=inp, interpolator=interpolator)
                new_inputs.append(superimposed)
            else:
                new_inputs.append(inp)
        inputs = new_inputs

        # Update metadatas
        metadatas = {inp: inp.GetImageMetaData('out') for inp in inputs}

    # Final superimposition to be sure to have the exact same image sizes
    # Getting the sizes of images
    image_sizes = {}
    for inp in inputs:
        if isinstance(inp, str):
            inp = Input(inp)
        image_sizes[inp] = inp.shape[:2]

    # Selecting the most frequent image size. It will be used as reference.
    most_common_image_size, _ = Counter(image_sizes.values()).most_common(1)[0]
    same_size_images = [inp for inp, image_size in image_sizes.items() if image_size == most_common_image_size]

    # Superimposition for images that do not have the same size as the others
    new_inputs = []
    for inp in inputs:
        if image_sizes[inp] != most_common_image_size:
            superimposed = App('Superimpose', inr=same_size_images[0], inm=inp, interpolator=interpolator)
            new_inputs.append(superimposed)
        else:
            new_inputs.append(inp)
    inputs = new_inputs

    return inputs

def run_tf_function(func):
    """This function enables using a function that calls some TF operations, with pyotb object as inputs.

    For example, you can write a function that uses TF operations like this :
        ```python
        @run_tf_function
        def multiply(input1, input2):
            import tensorflow as tf
            return tf.multiply(input1, input2)

        # Then you can use the function like this :
        result = multiply(pyotb_object1, pyotb_object1)  # this is a pyotb object
        ```

    Args:
        func: function taking one or several inputs and returning *one* output

    Returns:
        wrapper: a function that returns a pyotb object

    """
    try:
        from .apps import TensorflowModelServe
    except ImportError:
        logger.error('Could not run Tensorflow function: failed to import TensorflowModelServe.'
                     'Check that you have OTBTF configured (https://github.com/remicres/otbtf#how-to-install)')
        raise

    def get_tf_pycmd(output_dir, channels, scalar_inputs):
        """Create a string containing all python instructions necessary to create and save the Keras model.

        Args:
            output_dir: directory under which to save the model
            channels: list of raster channels (int). Contain `None` entries for non-raster inputs
            scalar_inputs: list of scalars (int/float). Contain `None` entries for non-scalar inputs

        Returns:
            the whole string code for function definition + model saving

        """
        # Getting the string definition of the tf function (e.g. "def multiply(x1, x2):...")
        # TODO: maybe not entirely foolproof, maybe we should use dill instead? but it would add a dependency
        func_def_str = inspect.getsource(func)
        func_name = func.__name__

        create_and_save_model_str = func_def_str

        # Adding the instructions to create the model and save it to output dir
        create_and_save_model_str += textwrap.dedent(f"""
            import tensorflow as tf

            model_inputs = []
            tf_inputs = []
            for channel, scalar_input in zip({channels}, {scalar_inputs}):
                if channel:
                    input = tf.keras.Input((None, None, channel))
                    tf_inputs.append(input)
                    model_inputs.append(input)
                else:
                    if isinstance(scalar_input, int):  # TF doesn't like mixing float and int
                        scalar_input = float(scalar_input)
                    tf_inputs.append(scalar_input)

            output = {func_name}(*tf_inputs)

            # Create and save the .pb model
            model = tf.keras.Model(inputs=model_inputs, outputs=output)
            model.save("{output_dir}")
            """)

        return create_and_save_model_str

    def wrapper(*inputs, tmp_dir='/tmp'):
        """For the user point of view, this function simply applies some TensorFlow operations to some rasters.

        Implicitly, it saves a .pb model that describe the TF operations, then creates an OTB ModelServe application
        that applies this .pb model to the inputs.

        Args:
            *inputs: a list of pyotb objects, filepaths or int/float numbers
            tmp_dir: directory where temporary models can be written (Default value = '/tmp')

        Returns:
            a pyotb object, output of TensorFlowModelServe

        """
        # Get infos about the inputs
        channels = []
        scalar_inputs = []
        raster_inputs = []
        for inp in inputs:
            try:
                # this is for raster input
                channel = get_nbchannels(inp)
                channels.append(channel)
                scalar_inputs.append(None)
                raster_inputs.append(inp)
            except TypeError:
                # this is for other inputs (float, int)
                channels.append(None)
                scalar_inputs.append(inp)

        # Create and save the model. This is executed **inside an independent process** because (as of 2022-03),
        # tensorflow python library and OTBTF are incompatible
        out_savedmodel = os.path.join(tmp_dir, f'tmp_otbtf_model_{uuid.uuid4()}')
        pycmd = get_tf_pycmd(out_savedmodel, channels, scalar_inputs)
        cmd_args = [sys.executable, "-c", pycmd]
        try:
            import subprocess
            subprocess.run(cmd_args, env=os.environ, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True)
        except subprocess.SubprocessError:
            logger.debug("Failed to call subprocess")
        if not os.path.isdir(out_savedmodel):
            logger.info("Failed to save the model")

        # Initialize the OTBTF model serving application
        model_serve = TensorflowModelServe({'model.dir': out_savedmodel, 'optim.disabletiling': 'on',
                                            'model.fullyconv': 'on'}, n_sources=len(raster_inputs), frozen=True)
        # Set parameters and execute
        for i, inp in enumerate(raster_inputs):
            model_serve.set_parameters({f'source{i + 1}.il': [inp]})
        model_serve.execute()
        # TODO: handle the deletion of the temporary model ?

        return model_serve

    return wrapper

def where(cond, x, y):
    """Functionally similar to numpy.where. Where cond is True (!=0), returns x. Else returns y.

    Args:
        cond: condition, must be a raster (filepath, App, Operation...). If cond is monoband whereas x or y are
              multiband, cond channels are expanded to match x & y ones.
        x: value if cond is True. Can be float, int, App, filepath, Operation...
        y: value if cond is False. Can be float, int, App, filepath, Operation...

    Returns:
        an output where pixels are x if cond is True, else y

    """
    # Checking the number of bands of rasters. Several cases :
    # - if cond is monoband, x and y can be multibands. Then cond will adapt to match x and y nb of bands
    # - if cond is multiband, x and y must have the same nb of bands if they are rasters.
    x_nb_channels, y_nb_channels = None, None
    if not isinstance(x, (int, float)):
        x_nb_channels = get_nbchannels(x)
    if not isinstance(y, (int, float)):
        y_nb_channels = get_nbchannels(y)

    if x_nb_channels and y_nb_channels:
        if x_nb_channels != y_nb_channels:
            raise ValueError('X and Y images do not have the same number of bands. '
                             f'X has {x_nb_channels} bands whereas Y has {y_nb_channels} bands')

    x_or_y_nb_channels = x_nb_channels if x_nb_channels else y_nb_channels
    cond_nb_channels = get_nbchannels(cond)

    # Get the number of bands of the result
    if x_or_y_nb_channels:  # if X or Y is a raster
        out_nb_channels = x_or_y_nb_channels
    else:  # if only cond is a raster
        out_nb_channels = cond_nb_channels

    if cond_nb_channels != 1 and x_or_y_nb_channels and cond_nb_channels != x_or_y_nb_channels:
        raise ValueError('Condition and X&Y do not have the same number of bands. Condition has '
                         f'{cond_nb_channels} bands whereas X&Y have {x_or_y_nb_channels} bands')

    # If needed, duplicate the single band binary mask to multiband to match the dimensions of x & y
    if cond_nb_channels == 1 and x_or_y_nb_channels and x_or_y_nb_channels != 1:
        logger.info('The condition has one channel whereas X/Y has/have %s channels. Expanding number'
                    ' of channels of condition to match the number of channels of X/Y', x_or_y_nb_channels)

    operation = Operation('?', cond, x, y, nb_bands=out_nb_channels)

    return operation