""" Handy utility functions. """ from __future__ import annotations import asyncio import copy import functools import importlib import inspect import json import json.decoder import os import pkgutil import pprint import random import re import threading import time import traceback import typing import warnings from abc import ABC, abstractmethod from contextlib import contextmanager from io import BytesIO from numbers import Number from pathlib import Path from types import GeneratorType from typing import ( TYPE_CHECKING, Any, Callable, Iterator, Optional, TypeVar, ) import anyio import matplotlib import requests from gradio_client.serializing import Serializable from typing_extensions import ParamSpec import gradio from gradio.context import Context from gradio.strings import en if TYPE_CHECKING: # Only import for type checking (is False at runtime). from gradio.blocks import Block, BlockContext, Blocks from gradio.components import Component from gradio.routes import App JSON_PATH = os.path.join(os.path.dirname(gradio.__file__), "launches.json") P = ParamSpec("P") T = TypeVar("T") def get_package_version() -> str: try: package_json_data = ( pkgutil.get_data(__name__, "package.json").decode("utf-8").strip() # type: ignore ) package_data = json.loads(package_json_data) version = package_data.get("version", "") return version except Exception: return "" def safe_get_lock() -> asyncio.Lock: """Get asyncio.Lock() without fear of getting an Exception. Needed because in reload mode we import the Blocks object outside the main thread. """ try: asyncio.get_event_loop() return asyncio.Lock() except RuntimeError: return None # type: ignore class BaseReloader(ABC): @property @abstractmethod def running_app(self) -> App: pass def queue_changed(self, demo: Blocks): return ( hasattr(self.running_app.blocks, "_queue") and not hasattr(demo, "_queue") ) or ( not hasattr(self.running_app.blocks, "_queue") and hasattr(demo, "_queue") ) def swap_blocks(self, demo: Blocks): assert self.running_app.blocks # Copy over the blocks to get new components and events but # not a new queue if hasattr(self.running_app.blocks, "_queue"): self.running_app.blocks._queue.blocks_dependencies = demo.dependencies demo._queue = self.running_app.blocks._queue self.running_app.blocks = demo class SourceFileReloader(BaseReloader): def __init__( self, app: App, watch_dirs: list[str], watch_file: str, stop_event: threading.Event, change_event: threading.Event, demo_name: str = "demo", ) -> None: super().__init__() self.app = app self.watch_dirs = watch_dirs self.watch_file = watch_file self.stop_event = stop_event self.change_event = change_event self.demo_name = demo_name @property def running_app(self) -> App: return self.app def should_watch(self) -> bool: return not self.stop_event.is_set() def stop(self) -> None: self.stop_event.set() def alert_change(self): self.change_event.set() def swap_blocks(self, demo: Blocks): super().swap_blocks(demo) self.alert_change() def watchfn(reloader: SourceFileReloader): """Watch python files in a given module. get_changes is taken from uvicorn's default file watcher. """ # The thread running watchfn will be the thread reloading # the app. So we need to modify this thread_data attr here # so that subsequent calls to reload don't launch the app from gradio.reload import reload_thread reload_thread.running_reload = True def get_changes() -> Path | None: for file in iter_py_files(): try: mtime = file.stat().st_mtime except OSError: # pragma: nocover continue old_time = mtimes.get(file) if old_time is None: mtimes[file] = mtime continue elif mtime > old_time: return file return None def iter_py_files() -> Iterator[Path]: for reload_dir in reload_dirs: for path in list(reload_dir.rglob("*.py")): yield path.resolve() module = None reload_dirs = [Path(dir_) for dir_ in reloader.watch_dirs] mtimes = {} while reloader.should_watch(): import sys changed = get_changes() if changed: print(f"Changes detected in: {changed}") # To simulate a fresh reload, delete all module references from sys.modules # for the modules in the package the change came from. dir_ = next(d for d in reload_dirs if is_in_or_equal(changed, d)) modules = list(sys.modules) for k in modules: v = sys.modules[k] sourcefile = getattr(v, "__file__", None) # Do not reload `reload.py` to keep thread data if ( sourcefile and dir_ == Path(inspect.getfile(gradio)).parent and sourcefile.endswith("reload.py") ): continue if sourcefile and is_in_or_equal(sourcefile, dir_): del sys.modules[k] try: module = importlib.import_module(reloader.watch_file) module = importlib.reload(module) except Exception as e: print( f"Reloading {reloader.watch_file} failed with the following exception: " ) traceback.print_exception(None, value=e, tb=None) mtimes = {} continue demo = getattr(module, reloader.demo_name) if reloader.queue_changed(demo): print( "Reloading failed. The new demo has a queue and the old one doesn't (or vice versa). " "Please launch your demo again" ) else: reloader.swap_blocks(demo) mtimes = {} def colab_check() -> bool: """ Check if interface is launching from Google Colab :return is_colab (bool): True or False """ is_colab = False try: # Check if running interactively using ipython. from IPython.core.getipython import get_ipython from_ipynb = get_ipython() if "google.colab" in str(from_ipynb): is_colab = True except (ImportError, NameError): pass return is_colab def kaggle_check() -> bool: return bool( os.environ.get("KAGGLE_KERNEL_RUN_TYPE") or os.environ.get("GFOOTBALL_DATA_DIR") ) def sagemaker_check() -> bool: try: import boto3 # type: ignore client = boto3.client("sts") response = client.get_caller_identity() return "sagemaker" in response["Arn"].lower() except Exception: return False def ipython_check() -> bool: """ Check if interface is launching from iPython (not colab) :return is_ipython (bool): True or False """ is_ipython = False try: # Check if running interactively using ipython. from IPython.core.getipython import get_ipython if get_ipython() is not None: is_ipython = True except (ImportError, NameError): pass return is_ipython def get_space() -> str | None: if os.getenv("SYSTEM") == "spaces": return os.getenv("SPACE_ID") return None def is_zero_gpu_space() -> bool: return os.getenv("SPACES_ZERO_GPU") == "true" def readme_to_html(article: str) -> str: try: response = requests.get(article, timeout=3) if response.status_code == requests.codes.ok: # pylint: disable=no-member article = response.text except requests.exceptions.RequestException: pass return article def show_tip(interface: gradio.Blocks) -> None: if interface.show_tips and random.random() < 1.5: tip: str = random.choice(en["TIPS"]) print(f"Tip: {tip}") def launch_counter() -> None: try: if not os.path.exists(JSON_PATH): launches = {"launches": 1} with open(JSON_PATH, "w+") as j: json.dump(launches, j) else: with open(JSON_PATH) as j: launches = json.load(j) launches["launches"] += 1 if launches["launches"] in [25, 50, 150, 500, 1000]: print(en["BETA_INVITE"]) with open(JSON_PATH, "w") as j: j.write(json.dumps(launches)) except Exception: pass def get_default_args(func: Callable) -> list[Any]: signature = inspect.signature(func) return [ v.default if v.default is not inspect.Parameter.empty else None for v in signature.parameters.values() ] def assert_configs_are_equivalent_besides_ids( config1: dict, config2: dict, root_keys: tuple = ("mode",) ): """Allows you to test if two different Blocks configs produce the same demo. Parameters: config1 (dict): nested dict with config from the first Blocks instance config2 (dict): nested dict with config from the second Blocks instance root_keys (Tuple): an interable consisting of which keys to test for equivalence at the root level of the config. By default, only "mode" is tested, so keys like "version" are ignored. """ config1 = copy.deepcopy(config1) config2 = copy.deepcopy(config2) pp = pprint.PrettyPrinter(indent=2) for key in root_keys: assert config1[key] == config2[key], f"Configs have different: {key}" assert len(config1["components"]) == len( config2["components"] ), "# of components are different" def assert_same_components(config1_id, config2_id): c1 = list(filter(lambda c: c["id"] == config1_id, config1["components"]))[0] c2 = list(filter(lambda c: c["id"] == config2_id, config2["components"]))[0] c1 = copy.deepcopy(c1) c1.pop("id") c2 = copy.deepcopy(c2) c2.pop("id") assert json.dumps(c1) == json.dumps( c2 ), f"{pp.pprint(c1)} does not match {pp.pprint(c2)}" def same_children_recursive(children1, chidren2): for child1, child2 in zip(children1, chidren2): assert_same_components(child1["id"], child2["id"]) if "children" in child1 or "children" in child2: same_children_recursive(child1["children"], child2["children"]) children1 = config1["layout"]["children"] children2 = config2["layout"]["children"] same_children_recursive(children1, children2) for d1, d2 in zip(config1["dependencies"], config2["dependencies"]): for t1, t2 in zip(d1.pop("targets"), d2.pop("targets")): assert_same_components(t1, t2) for i1, i2 in zip(d1.pop("inputs"), d2.pop("inputs")): assert_same_components(i1, i2) for o1, o2 in zip(d1.pop("outputs"), d2.pop("outputs")): assert_same_components(o1, o2) assert d1 == d2, f"{d1} does not match {d2}" return True def format_ner_list(input_string: str, ner_groups: list[dict[str, str | int]]): if len(ner_groups) == 0: return [(input_string, None)] output = [] end = 0 prev_end = 0 for group in ner_groups: entity, start, end = group["entity_group"], group["start"], group["end"] output.append((input_string[prev_end:start], None)) output.append((input_string[start:end], entity)) prev_end = end output.append((input_string[end:], None)) return output def delete_none(_dict: dict, skip_value: bool = False) -> dict: """ Delete keys whose values are None from a dictionary """ for key, value in list(_dict.items()): if skip_value and key == "value": continue elif value is None: del _dict[key] return _dict def resolve_singleton(_list: list[Any] | Any) -> Any: if len(_list) == 1: return _list[0] else: return _list def component_or_layout_class(cls_name: str) -> type[Component] | type[BlockContext]: """ Returns the component, template, or layout class with the given class name, or raises a ValueError if not found. Parameters: cls_name (str): lower-case string class name of a component Returns: cls: the component class """ import gradio.blocks import gradio.components import gradio.layouts import gradio.templates components = [ (name, cls) for name, cls in gradio.components.__dict__.items() if isinstance(cls, type) ] templates = [ (name, cls) for name, cls in gradio.templates.__dict__.items() if isinstance(cls, type) ] layouts = [ (name, cls) for name, cls in gradio.layouts.__dict__.items() if isinstance(cls, type) ] for name, cls in components + templates + layouts: if name.lower() == cls_name.replace("_", "") and ( issubclass(cls, gradio.components.Component) or issubclass(cls, gradio.blocks.BlockContext) ): return cls raise ValueError(f"No such component or layout: {cls_name}") def run_coro_in_background(func: Callable, *args, **kwargs): """ Runs coroutines in background. Warning, be careful to not use this function in other than FastAPI scope, because the event_loop has not started yet. You can use it in any scope reached by FastAPI app. correct scope examples: endpoints in routes, Blocks.process_api incorrect scope examples: Blocks.launch Use startup_events in routes.py if you need to run a coro in background in Blocks.launch(). Example: utils.run_coro_in_background(fn, *args, **kwargs) Args: func: *args: **kwargs: Returns: """ event_loop = asyncio.get_event_loop() return event_loop.create_task(func(*args, **kwargs)) def run_sync_iterator_async(iterator): """Helper for yielding StopAsyncIteration from sync iterators.""" try: return next(iterator) except StopIteration: # raise a ValueError here because co-routines can't raise StopIteration themselves raise StopAsyncIteration() from None class SyncToAsyncIterator: """Treat a synchronous iterator as async one.""" def __init__(self, iterator, limiter) -> None: self.iterator = iterator self.limiter = limiter def __aiter__(self): return self async def __anext__(self): return await anyio.to_thread.run_sync( run_sync_iterator_async, self.iterator, limiter=self.limiter ) async def async_iteration(iterator): # anext not introduced until 3.10 :( return await iterator.__anext__() @contextmanager def set_directory(path: Path | str): """Context manager that sets the working directory to the given path.""" origin = Path().absolute() try: os.chdir(path) yield finally: os.chdir(origin) def sanitize_value_for_csv(value: str | Number) -> str | Number: """ Sanitizes a value that is being written to a CSV file to prevent CSV injection attacks. Reference: https://owasp.org/www-community/attacks/CSV_Injection """ if isinstance(value, Number): return value unsafe_prefixes = ["=", "+", "-", "@", "\t", "\n"] unsafe_sequences = [",=", ",+", ",-", ",@", ",\t", ",\n"] if any(value.startswith(prefix) for prefix in unsafe_prefixes) or any( sequence in value for sequence in unsafe_sequences ): value = f"'{value}" return value def sanitize_list_for_csv(values: list[Any]) -> list[Any]: """ Sanitizes a list of values (or a list of list of values) that is being written to a CSV file to prevent CSV injection attacks. """ sanitized_values = [] for value in values: if isinstance(value, list): sanitized_value = [sanitize_value_for_csv(v) for v in value] sanitized_values.append(sanitized_value) else: sanitized_value = sanitize_value_for_csv(value) sanitized_values.append(sanitized_value) return sanitized_values def append_unique_suffix(name: str, list_of_names: list[str]): """Appends a numerical suffix to `name` so that it does not appear in `list_of_names`.""" set_of_names: set[str] = set(list_of_names) # for O(1) lookup if name not in set_of_names: return name else: suffix_counter = 1 new_name = f"{name}_{suffix_counter}" while new_name in set_of_names: suffix_counter += 1 new_name = f"{name}_{suffix_counter}" return new_name def validate_url(possible_url: str) -> bool: headers = {"User-Agent": "gradio (https://gradio.app/; team@gradio.app)"} try: head_request = requests.head(possible_url, headers=headers) # some URLs, such as AWS S3 presigned URLs, return a 405 or a 403 for HEAD requests if head_request.status_code == 405 or head_request.status_code == 403: return requests.get(possible_url, headers=headers).ok return head_request.ok except Exception: return False def is_update(val): return isinstance(val, dict) and "update" in val.get("__type__", "") def get_continuous_fn(fn: Callable, every: float) -> Callable: def continuous_fn(*args): while True: output = fn(*args) if isinstance(output, GeneratorType): yield from output else: yield output time.sleep(every) return continuous_fn def function_wrapper( f, before_fn=None, before_args=None, after_fn=None, after_args=None ): before_args = [] if before_args is None else before_args after_args = [] if after_args is None else after_args if inspect.isasyncgenfunction(f): @functools.wraps(f) async def asyncgen_wrapper(*args, **kwargs): if before_fn: before_fn(*before_args) async for response in f(*args, **kwargs): yield response if after_fn: after_fn(*after_args) return asyncgen_wrapper elif asyncio.iscoroutinefunction(f): @functools.wraps(f) async def async_wrapper(*args, **kwargs): if before_fn: before_fn(*before_args) response = await f(*args, **kwargs) if after_fn: after_fn(*after_args) return response return async_wrapper elif inspect.isgeneratorfunction(f): @functools.wraps(f) def gen_wrapper(*args, **kwargs): if before_fn: before_fn(*before_args) yield from f(*args, **kwargs) if after_fn: after_fn(*after_args) return gen_wrapper else: @functools.wraps(f) def wrapper(*args, **kwargs): if before_fn: before_fn(*before_args) response = f(*args, **kwargs) if after_fn: after_fn(*after_args) return response return wrapper def get_function_with_locals(fn: Callable, blocks: Blocks, event_id: str | None): def before_fn(blocks, event_id): from gradio.context import thread_data thread_data.blocks = blocks thread_data.event_id = event_id return function_wrapper(fn, before_fn=before_fn, before_args=(blocks, event_id)) async def cancel_tasks(task_ids: set[str]): matching_tasks = [ task for task in asyncio.all_tasks() if task.get_name() in task_ids ] for task in matching_tasks: task.cancel() await asyncio.gather(*matching_tasks, return_exceptions=True) def set_task_name(task, session_hash: str, fn_index: int, batch: bool): if not batch: task.set_name(f"{session_hash}_{fn_index}") def get_cancel_function( dependencies: list[dict[str, Any]] ) -> tuple[Callable, list[int]]: fn_to_comp = {} for dep in dependencies: if Context.root_block: fn_index = next( i for i, d in enumerate(Context.root_block.dependencies) if d == dep ) fn_to_comp[fn_index] = [ Context.root_block.blocks[o] for o in dep["outputs"] ] async def cancel(session_hash: str) -> None: task_ids = {f"{session_hash}_{fn}" for fn in fn_to_comp} await cancel_tasks(task_ids) return ( cancel, list(fn_to_comp.keys()), ) def get_type_hints(fn): # Importing gradio with the canonical abbreviation. Used in typing._eval_type. import gradio as gr # noqa: F401 from gradio import OAuthProfile, Request # noqa: F401 if inspect.isfunction(fn) or inspect.ismethod(fn): pass elif callable(fn): fn = fn.__call__ else: return {} try: return typing.get_type_hints(fn) except TypeError: # On Python 3.9 or earlier, get_type_hints throws a TypeError if the function # has a type annotation that include "|". We resort to parsing the signature # manually using inspect.signature. type_hints = {} sig = inspect.signature(fn) for name, param in sig.parameters.items(): if param.annotation is inspect.Parameter.empty: continue if param.annotation == "gr.OAuthProfile | None": # Special case: we want to inject the OAuthProfile value even on Python 3.9 type_hints[name] = Optional[OAuthProfile] if "|" in str(param.annotation): continue # To convert the string annotation to a class, we use the # internal typing._eval_type function. This is not ideal, but # it's the only way to do it without eval-ing the string. # Since the API is internal, it may change in the future. try: type_hints[name] = typing._eval_type( # type: ignore typing.ForwardRef(param.annotation), globals(), locals() ) except (NameError, TypeError): pass return type_hints def is_special_typed_parameter(name, parameter_types): from gradio.helpers import EventData from gradio.oauth import OAuthProfile from gradio.routes import Request """Checks if parameter has a type hint designating it as a gr.Request, gr.EventData or gr.OAuthProfile.""" hint = parameter_types.get(name) if not hint: return False is_request = hint == Request is_oauth_arg = hint in (OAuthProfile, Optional[OAuthProfile]) is_event_data = inspect.isclass(hint) and issubclass(hint, EventData) return is_request or is_event_data or is_oauth_arg def check_function_inputs_match(fn: Callable, inputs: list, inputs_as_dict: bool): """ Checks if the input component set matches the function Returns: None if valid, a string error message if mismatch """ signature = inspect.signature(fn) parameter_types = get_type_hints(fn) min_args = 0 max_args = 0 infinity = -1 for name, param in signature.parameters.items(): has_default = param.default != param.empty if param.kind in [param.POSITIONAL_ONLY, param.POSITIONAL_OR_KEYWORD]: if not is_special_typed_parameter(name, parameter_types): if not has_default: min_args += 1 max_args += 1 elif param.kind == param.VAR_POSITIONAL: max_args = infinity elif param.kind == param.KEYWORD_ONLY and not has_default: return f"Keyword-only args must have default values for function {fn}" arg_count = 1 if inputs_as_dict else len(inputs) if min_args == max_args and max_args != arg_count: warnings.warn( f"Expected {max_args} arguments for function {fn}, received {arg_count}." ) if arg_count < min_args: warnings.warn( f"Expected at least {min_args} arguments for function {fn}, received {arg_count}." ) if max_args != infinity and arg_count > max_args: warnings.warn( f"Expected maximum {max_args} arguments for function {fn}, received {arg_count}." ) def concurrency_count_warning(queue: Callable[P, T]) -> Callable[P, T]: @functools.wraps(queue) def wrapper(*args: P.args, **kwargs: P.kwargs) -> T: _self, *positional = args if is_zero_gpu_space() and ( len(positional) >= 1 or "concurrency_count" in kwargs ): warnings.warn( "Queue concurrency_count on ZeroGPU Spaces cannot be overridden " "and is always equal to Block's max_threads. " "Consider setting max_threads value on the Block instead" ) return queue(*args, **kwargs) return wrapper class TupleNoPrint(tuple): # To remove printing function return in notebook def __repr__(self): return "" def __str__(self): return "" class MatplotlibBackendMananger: def __enter__(self): self._original_backend = matplotlib.get_backend() matplotlib.use("agg") def __exit__(self, exc_type, exc_val, exc_tb): matplotlib.use(self._original_backend) def tex2svg(formula, *args): with MatplotlibBackendMananger(): import matplotlib.pyplot as plt fontsize = 20 dpi = 300 plt.rc("mathtext", fontset="cm") fig = plt.figure(figsize=(0.01, 0.01)) fig.text(0, 0, rf"${formula}$", fontsize=fontsize) output = BytesIO() fig.savefig( output, dpi=dpi, transparent=True, format="svg", bbox_inches="tight", pad_inches=0.0, ) plt.close(fig) output.seek(0) xml_code = output.read().decode("utf-8") svg_start = xml_code.index(".*<\/metadata>", "", svg_code, flags=re.DOTALL) svg_code = re.sub(r' width="[^"]+"', "", svg_code) height_match = re.search(r'height="([\d.]+)pt"', svg_code) if height_match: height = float(height_match.group(1)) new_height = height / fontsize # conversion from pt to em svg_code = re.sub( r'height="[\d.]+pt"', f'height="{new_height}em"', svg_code ) copy_code = f"{formula}" return f"{copy_code}{svg_code}" def abspath(path: str | Path) -> Path: """Returns absolute path of a str or Path path, but does not resolve symlinks.""" path = Path(path) if path.is_absolute(): return path # recursively check if there is a symlink within the path is_symlink = path.is_symlink() or any( parent.is_symlink() for parent in path.parents ) if is_symlink or path == path.resolve(): # in case path couldn't be resolved return Path.cwd() / path else: return path.resolve() def is_in_or_equal(path_1: str | Path, path_2: str | Path): """ True if path_1 is a descendant (i.e. located within) path_2 or if the paths are the same, returns False otherwise. Parameters: path_1: str or Path (should be a file) path_2: str or Path (can be a file or directory) """ path_1, path_2 = abspath(path_1), abspath(path_2) try: if str(path_1.relative_to(path_2)).startswith(".."): # prevent path traversal return False except ValueError: return False return True def get_serializer_name(block: Block) -> str | None: if not hasattr(block, "serialize"): return None def get_class_that_defined_method(meth: Callable): # Adapted from: https://stackoverflow.com/a/25959545/5209347 if isinstance(meth, functools.partial): return get_class_that_defined_method(meth.func) if inspect.ismethod(meth) or ( inspect.isbuiltin(meth) and getattr(meth, "__self__", None) is not None and getattr(meth.__self__, "__class__", None) ): for cls in inspect.getmro(meth.__self__.__class__): # Find the first serializer defined in gradio_client that if issubclass(cls, Serializable) and "gradio_client" in cls.__module__: return cls if meth.__name__ in cls.__dict__: return cls meth = getattr(meth, "__func__", meth) # fallback to __qualname__ parsing if inspect.isfunction(meth): cls = getattr( inspect.getmodule(meth), meth.__qualname__.split(".", 1)[0].rsplit(".", 1)[0], None, ) if isinstance(cls, type): return cls return getattr(meth, "__objclass__", None) cls = get_class_that_defined_method(block.serialize) # type: ignore if cls: return cls.__name__ HTML_TAG_RE = re.compile("<.*?>") def remove_html_tags(raw_html: str | None) -> str: return re.sub(HTML_TAG_RE, "", raw_html or "") def find_user_stack_level() -> int: """ Find the first stack frame not inside Gradio. """ frame = inspect.currentframe() n = 0 while frame: fname = inspect.getfile(frame) if "/gradio/" not in fname.replace(os.sep, "/"): break frame = frame.f_back n += 1 return n