GlitchTip/sentry/stacktraces/functions.py

from .platform import get_behavior_family_for_platform
import re


_windecl_hash = re.compile(r"^@?(.*?)@[0-9]+$")
_rust_hash = re.compile(r"::h[a-z0-9]{16}$")
_cpp_trailer_re = re.compile(r"(\bconst\b|&)$")
_rust_blanket_re = re.compile(r"^([A-Z] as )")
_lambda_re = re.compile(
    r"""(?x)
    # gcc
    (?:
        \{
            lambda\(.*?\)\#\d+
        \}
    ) |
    # msvc
    (?:
        \blambda_[a-f0-9]{32}\b
    ) |
    # clang
    (?:
        \$_\d+\b
    )
    """
)
_anon_namespace_re = re.compile(
    r"""(?x)
    \?A0x[a-f0-9]{8}::
    """
)

PAIRS = {"(": ")", "{": "}", "[": "]", "<": ">"}


def replace_enclosed_string(s, start, end, replacement=None):
    if start not in s:
        return s

    depth = 0

    rv = []
    pair_start = None
    for idx, char in enumerate(s):
        if char == start:
            if depth == 0:
                pair_start = idx
            depth += 1
        elif char == end:
            depth -= 1
            if depth == 0:
                if replacement is not None:
                    if callable(replacement):
                        rv.append(replacement(s[pair_start + 1 : idx], pair_start))
                    else:
                        rv.append(replacement)
        elif depth == 0:
            rv.append(char)

    return "".join(rv)


def split_func_tokens(s):
    buf = []
    rv = []
    stack = []
    end = 0

    for idx, char in enumerate(s):
        if char in PAIRS:
            stack.append(PAIRS[char])
        elif stack and char == stack[-1]:
            stack.pop()
            if not stack:
                buf.append(s[end : idx + 1])
                end = idx + 1
        elif not stack:
            if char.isspace():
                if buf:
                    rv.append(buf)
                buf = []
            else:
                buf.append(s[end : idx + 1])
            end = idx + 1

    if buf:
        rv.append(buf)

    return ["".join(x) for x in rv]


def trim_function_name(function, platform, normalize_lambdas=True):
    """Given a function value from the frame's function attribute this returns
    a trimmed version that can be stored in `function_name`.  This is only used
    if the client did not supply a value itself already.
    """
    if get_behavior_family_for_platform(platform) != "native":
        return function
    if function in ("<redacted>", "<unknown>"):
        return function

    original_function = function
    function = function.strip()

    # Ensure we don't operate on objc functions
    if function.startswith(("[", "+[", "-[")):
        return function

    # Chop off C++ trailers
    while True:
        match = _cpp_trailer_re.search(function)
        if match is None:
            break
        function = function[: match.start()].rstrip()

    # Because operator<< really screws with our balancing, so let's work
    # around that by replacing it with a character we do not observe in
    # `split_func_tokens` or `replace_enclosed_string`.
    function = (
        function.replace("operator<<", u"operator⟨⟨")
        .replace("operator<", u"operator⟨")
        .replace("operator()", u"operator◯")
        .replace(" -> ", u" ⟿ ")
        .replace("`anonymous namespace'", u"〔anonymousnamespace〕")
    )

    # normalize C++ lambdas.  This is necessary because different
    # compilers use different rules for now to name a lambda and they are
    # all quite inconsistent.  This does not give us perfect answers to
    # this problem but closer.  In particular msvc will call a lambda
    # something like `lambda_deadbeefeefffeeffeeff` whereas clang for
    # instance will name it `main::$_0` which will tell us in which outer
    # function it was declared.
    if normalize_lambdas:
        function = _lambda_re.sub("lambda", function)

    # Normalize MSVC anonymous namespaces from inline functions.  For inline
    # functions, the compiler inconsistently renders anonymous namespaces with
    # their hash.  For regular functions,  "`anonymous namespace'" is used.
    # The regular expression matches the trailing "::" to avoid accidental
    # replacement in mangled function names.
    if normalize_lambdas:
        function = _anon_namespace_re.sub(u"〔anonymousnamespace〕::", function)

    # Remove the arguments if there is one.
    def process_args(value, start):
        value = value.strip()
        if value in ("anonymous namespace", "operator"):
            return "(%s)" % value
        return ""

    function = replace_enclosed_string(function, "(", ")", process_args)

    # Resolve generic types, but special case rust which uses things like
    # <Foo as Bar>::baz to denote traits.
    def process_generics(value, start):
        # Special case for lambdas
        if value == "lambda" or _lambda_re.match(value):
            return "<%s>" % value

        if start > 0:
            return "<T>"

        # Rust special cases
        value = _rust_blanket_re.sub("", value)  # prefer trait for blanket impls
        value = replace_enclosed_string(value, "<", ">", process_generics)
        return value.split(" as ", 1)[0]

    function = replace_enclosed_string(function, "<", ">", process_generics)

    tokens = split_func_tokens(function)

    # MSVC demangles generic operator functions with a space between the
    # function name and the generics. Ensure that those two components both end
    # up in the function name.
    if len(tokens) > 1 and tokens[-1] == "<T>":
        tokens.pop()
        tokens[-1] += " <T>"

    # find the token which is the function name.  Since we chopped of C++
    # trailers there are only two cases we care about: the token left to
    # the -> return marker which is for instance used in Swift and if that
    # is not found, the last token in the last.
    #
    # ["unsigned", "int", "whatever"] -> whatever
    # ["@objc", "whatever", "->", "int"] -> whatever
    try:
        func_token = tokens[tokens.index(u"⟿") - 1]
    except ValueError:
        if tokens:
            func_token = tokens[-1]
        else:
            func_token = None

    if func_token:
        function = (
            func_token.replace(u"⟨", "<")
            .replace(u"◯", "()")
            .replace(u" ⟿ ", " -> ")
            .replace(u"〔anonymousnamespace〕", "`anonymous namespace'")
        )

    # This really should never happen
    else:
        function = original_function

    # trim off rust markers
    function = _rust_hash.sub("", function)

    # trim off windows decl markers
    return _windecl_hash.sub("\\1", function)


def get_function_name_for_frame(frame, platform=None):
    """Given a frame object or dictionary this returns the actual function
    name trimmed.
    """
    if hasattr(frame, "get_raw_data"):
        frame = frame.get_raw_data()

    # if there is a raw function, prioritize the function unchanged
    if frame.get("raw_function"):
        return frame.get("function")

    # otherwise trim the function on demand
    rv = frame.get("function")
    if rv:
        return trim_function_name(rv, frame.get("platform") or platform)