# encoding: utf-8 """ """ import typing as _t from inspect import cleandoc as _cleandoc from frozendict import deepfreeze as _deepfreeze from comfy.comfy_types.node_typing import IO as _IO from . import _meta from .docstring_formatter import format_docstring as _format_docstring from .enums import DataTypes as _DataTypes from .funcs_common import _new_dict_with_updated_key, _T # A tiny optimization by reusing the same immutable dict: _input_types = _deepfreeze({ 'required': { 'name': (_IO.STRING, {'tooltip': ( "Name of the substring inserted into the dict. " "It must comprise only of latin letters, digits and underscores + it can't start with a digit." )}), 'cleanup': (_IO.BOOLEAN, {'default': True, 'label_on': 'leading/trailing spaces', 'label_off': 'no', 'tooltip': ( "When enabled, each line in the sub-string is stripped from any spaces at its start and end." )}), 'string': (_IO.STRING, {'multiline': True, 'tooltip': "The actual sub-string to add into the dict."}), }, 'optional': { 'dict': _DataTypes.input_dict(tooltip="An optional Format-Dictionary to extend/update."), }, # 'hidden': { # 'unique_id': 'UNIQUE_ID', # used for text display at the bottom of the node # }, }) class StringConstructorDictAddString: """Add/update a string to the Format-Dict.""" NODE_NAME = 'StringConstructorDictAddString' CATEGORY = _meta.category DESCRIPTION = _format_docstring(_cleandoc(__doc__)) FUNCTION = 'main' RETURN_TYPES = (str(_DataTypes.DICT), ) RETURN_NAMES = (_DataTypes.DICT.lower(), ) # OUTPUT_TOOLTIPS = tuple() @classmethod def INPUT_TYPES(cls): return _input_types @staticmethod def main( name: str, cleanup: bool, string: str, dict: _t.Dict[str, _T] = None, # unique_id: str = None, ) -> _t.Tuple[_t.Dict[str, _t.Union[_T, str]]]: """Update/append a string to the dict.""" string = '' if string is None else str(string) if cleanup: string = '\n'.join( x.strip() for x in string.splitlines() ) return (_new_dict_with_updated_key(dict, name, string), )