diff options
| -rw-r--r-- | bot/exts/internal_eval/__init__.py | 10 | ||||
| -rw-r--r-- | bot/exts/internal_eval/_helpers.py | 249 | ||||
| -rw-r--r-- | bot/exts/internal_eval/_internal_eval.py | 176 |
3 files changed, 435 insertions, 0 deletions
diff --git a/bot/exts/internal_eval/__init__.py b/bot/exts/internal_eval/__init__.py new file mode 100644 index 00000000..695fa74d --- /dev/null +++ b/bot/exts/internal_eval/__init__.py @@ -0,0 +1,10 @@ +from bot.bot import Bot + + +def setup(bot: Bot) -> None: + """Set up the Internal Eval extension.""" + # Import the Cog at runtime to prevent side effects like defining + # RedisCache instances too early. + from ._internal_eval import InternalEval + + bot.add_cog(InternalEval(bot)) diff --git a/bot/exts/internal_eval/_helpers.py b/bot/exts/internal_eval/_helpers.py new file mode 100644 index 00000000..3a50b9f3 --- /dev/null +++ b/bot/exts/internal_eval/_helpers.py @@ -0,0 +1,249 @@ +import ast +import collections +import contextlib +import functools +import inspect +import io +import logging +import sys +import traceback +import types +import typing + + +log = logging.getLogger(__name__) + +# A type alias to annotate the tuples returned from `sys.exc_info()` +ExcInfo = typing.Tuple[typing.Type[Exception], Exception, types.TracebackType] +Namespace = typing.Dict[str, typing.Any] + +# This will be used as an coroutine function wrapper for the code +# to be evaluated. The wrapper contains one `pass` statement which +# will be replaced with `ast` with the code that we want to have +# evaluated. +# The function redirects output and captures exceptions that were +# raised in the code we evaluate. The latter is used to provide a +# meaningful traceback to the end user. +EVAL_WRAPPER = """ +async def _eval_wrapper_function(): + try: + with contextlib.redirect_stdout(_eval_context.stdout): + pass + if '_value_last_expression' in locals(): + if inspect.isawaitable(_value_last_expression): + _value_last_expression = await _value_last_expression + _eval_context._value_last_expression = _value_last_expression + else: + _eval_context._value_last_expression = None + except Exception: + _eval_context.exc_info = sys.exc_info() + finally: + _eval_context.locals = locals() +_eval_context.function = _eval_wrapper_function +""" +INTERNAL_EVAL_FRAMENAME = "<internal eval>" +EVAL_WRAPPER_FUNCTION_FRAMENAME = "_eval_wrapper_function" + + +def format_internal_eval_exception(exc_info: ExcInfo, code: str) -> str: + """Format an exception caught while evaluation code by inserting lines.""" + exc_type, exc_value, tb = exc_info + stack_summary = traceback.StackSummary.extract(traceback.walk_tb(tb)) + code = code.split("\n") + + output = ["Traceback (most recent call last):"] + for frame in stack_summary: + if frame.filename == INTERNAL_EVAL_FRAMENAME: + line = code[frame.lineno - 1].lstrip() + + if frame.name == EVAL_WRAPPER_FUNCTION_FRAMENAME: + name = INTERNAL_EVAL_FRAMENAME + else: + name = frame.name + else: + line = frame.line + name = frame.name + + output.append( + f' File "{frame.filename}", line {frame.lineno}, in {name}\n' + f" {line}" + ) + + output.extend(traceback.format_exception_only(exc_type, exc_value)) + return "\n".join(output) + + +class EvalContext: + """ + Represents the current `internal eval` context. + + The context remembers names set during earlier runs of `internal eval`. To + clear the context, use the `.internal clear` command. + """ + + def __init__(self, context_vars: Namespace, local_vars: Namespace) -> None: + self._locals = dict(local_vars) + self.context_vars = dict(context_vars) + + self.stdout = io.StringIO() + self._value_last_expression = None + self.exc_info = None + self.code = "" + self.function = None + self.eval_tree = None + + @property + def dependencies(self) -> typing.Dict[str, typing.Any]: + """ + Return a mapping of the dependencies for the wrapper function. + + By using a property descriptor, the mapping can't be accidentally + mutated during evaluation. This ensures the dependencies are always + available. + """ + return { + "print": functools.partial(print, file=self.stdout), + "contextlib": contextlib, + "inspect": inspect, + "sys": sys, + "_eval_context": self, + "_": self._value_last_expression, + } + + @property + def locals(self) -> typing.Dict[str, typing.Any]: + """Return a mapping of names->values needed for evaluation.""" + return {**collections.ChainMap(self.dependencies, self.context_vars, self._locals)} + + @locals.setter + def locals(self, locals_: typing.Dict[str, typing.Any]) -> None: + """Update the contextual mapping of names to values.""" + log.trace(f"Updating {self._locals} with {locals_}") + self._locals.update(locals_) + + def prepare_eval(self, code: str) -> typing.Optional[str]: + """Prepare an evaluation by processing the code and setting up the context.""" + self.code = code + + if not self.code: + log.debug("No code was attached to the evaluation command") + return "[No code detected]" + + try: + code_tree = ast.parse(code, filename=INTERNAL_EVAL_FRAMENAME) + except SyntaxError: + log.debug("Got a SyntaxError while parsing the eval code") + return "".join(traceback.format_exception(*sys.exc_info(), limit=0)) + + log.trace("Parsing the AST to see if there's a trailing expression we need to capture") + code_tree = CaptureLastExpression(code_tree).capture() + + log.trace("Wrapping the AST in the AST of the wrapper coroutine") + eval_tree = WrapEvalCodeTree(code_tree).wrap() + + self.eval_tree = eval_tree + return None + + async def run_eval(self) -> Namespace: + """Run the evaluation and return the updated locals.""" + log.trace("Compiling the AST to bytecode using `exec` mode") + compiled_code = compile(self.eval_tree, filename=INTERNAL_EVAL_FRAMENAME, mode="exec") + + log.trace("Executing the compiled code with the desired namespace environment") + exec(compiled_code, self.locals) # noqa: B102,S102 + + log.trace("Awaiting the created evaluation wrapper coroutine.") + await self.function() + + log.trace("Returning the updated captured locals.") + return self._locals + + def format_output(self) -> str: + """Format the output of the most recent evaluation.""" + output = [] + + log.trace(f"Getting output from stdout `{id(self.stdout)}`") + stdout_text = self.stdout.getvalue() + if stdout_text: + log.trace("Appending output captured from stdout/print") + output.append(stdout_text) + + if self._value_last_expression is not None: + log.trace("Appending the output of a captured trialing expression") + output.append(f"[Captured] {self._value_last_expression!r}") + + if self.exc_info: + log.trace("Appending exception information") + output.append(format_internal_eval_exception(self.exc_info, self.code)) + + log.trace(f"Generated output: {output!r}") + return "\n".join(output) or "[No output]" + + +class WrapEvalCodeTree(ast.NodeTransformer): + """Wraps the AST of eval code with the wrapper function.""" + + def __init__(self, eval_code_tree: ast.AST, *args, **kwargs) -> None: + super().__init__(*args, **kwargs) + self.eval_code_tree = eval_code_tree + + # To avoid mutable aliasing, parse the WRAPPER_FUNC for each wrapping + self.wrapper = ast.parse(EVAL_WRAPPER, filename=INTERNAL_EVAL_FRAMENAME) + + def wrap(self) -> ast.AST: + """Wrap the tree of the code by the tree of the wrapper function.""" + new_tree = self.visit(self.wrapper) + return ast.fix_missing_locations(new_tree) + + def visit_Pass(self, node: ast.Pass) -> typing.List[ast.AST]: # noqa: N802 + """ + Replace the `_ast.Pass` node in the wrapper function by the eval AST. + + This method works on the assumption that there's a single `pass` + statement in the wrapper function. + """ + return list(ast.iter_child_nodes(self.eval_code_tree)) + + +class CaptureLastExpression(ast.NodeTransformer): + """Captures the return value from a loose expression.""" + + def __init__(self, tree: ast.AST, *args, **kwargs) -> None: + super().__init__(*args, **kwargs) + self.tree = tree + self.last_node = list(ast.iter_child_nodes(tree))[-1] + + def visit_Expr(self, node: ast.Expr) -> typing.Union[ast.Expr, ast.Assign]: # noqa: N802 + """ + Replace the Expr node that is last child node of Module with an assignment. + + We use an assignment to capture the value of the last node, if it's a loose + Expr node. Normally, the value of an Expr node is lost, meaning we don't get + the output of such a last "loose" expression. By assigning it a name, we can + retrieve it for our output. + """ + if node is not self.last_node: + return node + + log.trace("Found a trailing last expression in the evaluation code") + + log.trace("Creating assignment statement with trailing expression as the right-hand side") + right_hand_side = list(ast.iter_child_nodes(node))[0] + + assignment = ast.Assign( + targets=[ast.Name(id='_value_last_expression', ctx=ast.Store())], + value=right_hand_side, + lineno=node.lineno, + col_offset=0, + ) + ast.fix_missing_locations(assignment) + return assignment + + def capture(self) -> ast.AST: + """Capture the value of the last expression with an assignment.""" + if not isinstance(self.last_node, ast.Expr): + # We only have to replace a node if the very last node is an Expr node + return self.tree + + new_tree = self.visit(self.tree) + return ast.fix_missing_locations(new_tree) diff --git a/bot/exts/internal_eval/_internal_eval.py b/bot/exts/internal_eval/_internal_eval.py new file mode 100644 index 00000000..757a2a1e --- /dev/null +++ b/bot/exts/internal_eval/_internal_eval.py @@ -0,0 +1,176 @@ +import logging +import re +import textwrap +import typing + +import discord +from discord.ext import commands + +from bot.bot import Bot +from bot.constants import Roles +from bot.utils.decorators import with_role +from bot.utils.extensions import invoke_help_command +from ._helpers import EvalContext + +__all__ = ["InternalEval"] + +log = logging.getLogger(__name__) + +FORMATTED_CODE_REGEX = re.compile( + r"(?P<delim>(?P<block>```)|``?)" # code delimiter: 1-3 backticks; (?P=block) only matches if it's a block + r"(?(block)(?:(?P<lang>[a-z]+)\n)?)" # if we're in a block, match optional language (only letters plus newline) + r"(?:[ \t]*\n)*" # any blank (empty or tabs/spaces only) lines before the code + r"(?P<code>.*?)" # extract all code inside the markup + r"\s*" # any more whitespace before the end of the code markup + r"(?P=delim)", # match the exact same delimiter from the start again + re.DOTALL | re.IGNORECASE # "." also matches newlines, case insensitive +) + +RAW_CODE_REGEX = re.compile( + r"^(?:[ \t]*\n)*" # any blank (empty or tabs/spaces only) lines before the code + r"(?P<code>.*?)" # extract all the rest as code + r"\s*$", # any trailing whitespace until the end of the string + re.DOTALL # "." also matches newlines +) + + +class InternalEval(commands.Cog): + """Top secret code evaluation for admins and owners.""" + + def __init__(self, bot: Bot): + self.bot = bot + self.locals = {} + + @staticmethod + def shorten_output( + output: str, + max_length: int = 1900, + placeholder: str = "\n[output truncated]" + ) -> str: + """ + Shorten the `output` so it's shorter than `max_length`. + + There are three tactics for this, tried in the following order: + - Shorten the output on a line-by-line basis + - Shorten the output on any whitespace character + - Shorten the output solely on character count + """ + max_length = max_length - len(placeholder) + + shortened_output = [] + char_count = 0 + for line in output.split("\n"): + if char_count + len(line) > max_length: + break + shortened_output.append(line) + char_count += len(line) + 1 # account for (possible) line ending + + if shortened_output: + shortened_output.append(placeholder) + return "\n".join(shortened_output) + + shortened_output = textwrap.shorten(output, width=max_length, placeholder=placeholder) + + if shortened_output.strip() == placeholder.strip(): + # `textwrap` was unable to find whitespace to shorten on, so it has + # reduced the output to just the placeholder. Let's shorten based on + # characters instead. + shortened_output = output[:max_length] + placeholder + + return shortened_output + + async def _upload_output(self, output: str) -> typing.Optional[str]: + """Upload `internal eval` output to our pastebin and return the url.""" + try: + async with self.bot.http_session.post( + "https://paste.pythondiscord.com/documents", data=output, raise_for_status=True + ) as resp: + data = await resp.json() + + if "key" in data: + return f"https://paste.pythondiscord.com/{data['key']}" + except Exception: + # 400 (Bad Request) means there are too many characters + log.exception("Failed to upload `internal eval` output to paste service!") + + async def _send_output(self, ctx: commands.Context, output: str) -> None: + """Send the `internal eval` output to the command invocation context.""" + upload_message = "" + if len(output) >= 1980: + # The output is too long, let's truncate it for in-channel output and + # upload the complete output to the paste service. + url = await self._upload_output(output) + + if url: + upload_message = f"\nFull output here: {url}" + else: + upload_message = "\n:warning: Failed to upload full output!" + + output = self.shorten_output(output) + + await ctx.send(f"```py\n{output}\n```{upload_message}") + + async def _eval(self, ctx: commands.Context, code: str) -> None: + """Evaluate the `code` in the current evaluation context.""" + context_vars = { + "message": ctx.message, + "author": ctx.message.author, + "channel": ctx.channel, + "guild": ctx.guild, + "ctx": ctx, + "self": self, + "bot": self.bot, + "discord": discord, + } + + eval_context = EvalContext(context_vars, self.locals) + + log.trace("Preparing the evaluation by parsing the AST of the code") + error = eval_context.prepare_eval(code) + + if error: + log.trace("The code can't be evaluated due to an error") + await ctx.send(f"```py\n{error}\n```") + return + + log.trace("Evaluate the AST we've generated for the evaluation") + new_locals = await eval_context.run_eval() + + log.trace("Updating locals with those set during evaluation") + self.locals.update(new_locals) + + log.trace("Sending the formatted output back to the context") + await self._send_output(ctx, eval_context.format_output()) + + @commands.group(name='internal', aliases=('int',)) + @with_role(Roles.admin) + async def internal_group(self, ctx: commands.Context) -> None: + """Internal commands. Top secret!""" + if not ctx.invoked_subcommand: + await invoke_help_command(ctx) + + @internal_group.command(name='eval', aliases=('e',)) + @with_role(Roles.admin) + async def eval(self, ctx: commands.Context, *, code: str) -> None: + """Run eval in a REPL-like format.""" + if match := list(FORMATTED_CODE_REGEX.finditer(code)): + blocks = [block for block in match if block.group("block")] + + if len(blocks) > 1: + code = '\n'.join(block.group("code") for block in blocks) + else: + match = match[0] if len(blocks) == 0 else blocks[0] + code, block, lang, delim = match.group("code", "block", "lang", "delim") + + else: + code = RAW_CODE_REGEX.fullmatch(code).group("code") + + code = textwrap.dedent(code) + await self._eval(ctx, code) + + @internal_group.command(name='reset', aliases=("clear", "exit", "r", "c")) + @with_role(Roles.admin) + async def reset(self, ctx: commands.Context) -> None: + """Reset the context and locals of the eval session.""" + self.locals = {} + await ctx.send("The evaluation context was reset.") |