rustc_demangle/

v0.rs

use core::convert::TryFrom;
use core::{char, fmt, iter, mem, str};

#[allow(unused_macros)]
macro_rules! write {
    ($($ignored:tt)*) => {
        compile_error!(
            "use `self.print(value)` or `fmt::Trait::fmt(&value, self.out)`, \
             instead of `write!(self.out, \"{...}\", value)`"
        )
    };
}

// Maximum recursion depth when parsing symbols before we just bail out saying
// "this symbol is invalid"
const MAX_DEPTH: u32 = 500;

/// Representation of a demangled symbol name.
pub struct Demangle<'a> {
    inner: &'a str,
}

#[derive(PartialEq, Eq, Debug)]
pub enum ParseError {
    /// Symbol doesn't match the expected `v0` grammar.
    Invalid,

    /// Parsing the symbol crossed the recursion limit (see `MAX_DEPTH`).
    RecursedTooDeep,
}

/// De-mangles a Rust symbol into a more readable version
///
/// This function will take a **mangled** symbol and return a value. When printed,
/// the de-mangled version will be written. If the symbol does not look like
/// a mangled symbol, the original value will be written instead.
pub fn demangle(s: &str) -> Result<(Demangle, &str), ParseError> {
    // First validate the symbol. If it doesn't look like anything we're
    // expecting, we just print it literally. Note that we must handle non-Rust
    // symbols because we could have any function in the backtrace.
    let inner;
    if s.len() > 2 && s.starts_with("_R") {
        inner = &s[2..];
    } else if s.len() > 1 && s.starts_with('R') {
        // On Windows, dbghelp strips leading underscores, so we accept "R..."
        // form too.
        inner = &s[1..];
    } else if s.len() > 3 && s.starts_with("__R") {
        // On OSX, symbols are prefixed with an extra _
        inner = &s[3..];
    } else {
        return Err(ParseError::Invalid);
    }

    // Paths always start with uppercase characters.
    match inner.as_bytes()[0] {
        b'A'..=b'Z' => {}
        _ => return Err(ParseError::Invalid),
    }

    // only work with ascii text
    if inner.bytes().any(|c| c & 0x80 != 0) {
        return Err(ParseError::Invalid);
    }

    // Verify that the symbol is indeed a valid path.
    let try_parse_path = |parser| {
        let mut dummy_printer = Printer {
            parser: Ok(parser),
            out: None,
            bound_lifetime_depth: 0,
        };
        dummy_printer
            .print_path(false)
            .expect("`fmt::Error`s should be impossible without a `fmt::Formatter`");
        dummy_printer.parser
    };
    let mut parser = Parser {
        sym: inner,
        next: 0,
        depth: 0,
    };
    parser = try_parse_path(parser)?;

    // Instantiating crate (paths always start with uppercase characters).
    if let Some(&(b'A'..=b'Z')) = parser.sym.as_bytes().get(parser.next) {
        parser = try_parse_path(parser)?;
    }

    Ok((Demangle { inner }, &parser.sym[parser.next..]))
}

impl<'s> fmt::Display for Demangle<'s> {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        let mut printer = Printer {
            parser: Ok(Parser {
                sym: self.inner,
                next: 0,
                depth: 0,
            }),
            out: Some(f),
            bound_lifetime_depth: 0,
        };
        printer.print_path(true)
    }
}

struct Ident<'s> {
    /// ASCII part of the identifier.
    ascii: &'s str,
    /// Punycode insertion codes for Unicode codepoints, if any.
    punycode: &'s str,
}

const SMALL_PUNYCODE_LEN: usize = 128;

impl<'s> Ident<'s> {
    /// Attempt to decode punycode on the stack (allocation-free),
    /// and pass the char slice to the closure, if successful.
    /// This supports up to `SMALL_PUNYCODE_LEN` characters.
    fn try_small_punycode_decode<F: FnOnce(&[char]) -> R, R>(&self, f: F) -> Option<R> {
        let mut out = ['\0'; SMALL_PUNYCODE_LEN];
        let mut out_len = 0;
        let r = self.punycode_decode(|i, c| {
            // Check there's space left for another character.
            out.get(out_len).ok_or(())?;

            // Move the characters after the insert position.
            let mut j = out_len;
            out_len += 1;

            while j > i {
                out[j] = out[j - 1];
                j -= 1;
            }

            // Insert the new character.
            out[i] = c;

            Ok(())
        });
        if r.is_ok() {
            Some(f(&out[..out_len]))
        } else {
            None
        }
    }

    /// Decode punycode as insertion positions and characters
    /// and pass them to the closure, which can return `Err(())`
    /// to stop the decoding process.
    fn punycode_decode<F: FnMut(usize, char) -> Result<(), ()>>(
        &self,
        mut insert: F,
    ) -> Result<(), ()> {
        let mut punycode_bytes = self.punycode.bytes().peekable();
        if punycode_bytes.peek().is_none() {
            return Err(());
        }

        let mut len = 0;

        // Populate initial output from ASCII fragment.
        for c in self.ascii.chars() {
            insert(len, c)?;
            len += 1;
        }

        // Punycode parameters and initial state.
        let base = 36;
        let t_min = 1;
        let t_max = 26;
        let skew = 38;
        let mut damp = 700;
        let mut bias = 72;
        let mut i: usize = 0;
        let mut n: usize = 0x80;

        loop {
            // Read one delta value.
            let mut delta: usize = 0;
            let mut w = 1;
            let mut k: usize = 0;
            loop {
                use core::cmp::{max, min};

                k += base;
                let t = min(max(k.saturating_sub(bias), t_min), t_max);

                let d = match punycode_bytes.next() {
                    Some(d @ b'a'..=b'z') => d - b'a',
                    Some(d @ b'0'..=b'9') => 26 + (d - b'0'),
                    _ => return Err(()),
                };
                let d = d as usize;
                delta = delta.checked_add(d.checked_mul(w).ok_or(())?).ok_or(())?;
                if d < t {
                    break;
                }
                w = w.checked_mul(base - t).ok_or(())?;
            }

            // Compute the new insert position and character.
            len += 1;
            i = i.checked_add(delta).ok_or(())?;
            n = n.checked_add(i / len).ok_or(())?;
            i %= len;

            let n_u32 = n as u32;
            let c = if n_u32 as usize == n {
                char::from_u32(n_u32).ok_or(())?
            } else {
                return Err(());
            };

            // Insert the new character and increment the insert position.
            insert(i, c)?;
            i += 1;

            // If there are no more deltas, decoding is complete.
            if punycode_bytes.peek().is_none() {
                return Ok(());
            }

            // Perform bias adaptation.
            delta /= damp;
            damp = 2;

            delta += delta / len;
            let mut k = 0;
            while delta > ((base - t_min) * t_max) / 2 {
                delta /= base - t_min;
                k += base;
            }
            bias = k + ((base - t_min + 1) * delta) / (delta + skew);
        }
    }
}

impl<'s> fmt::Display for Ident<'s> {
    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
        self.try_small_punycode_decode(|chars| {
            for &c in chars {
                c.fmt(f)?;
            }
            Ok(())
        })
        .unwrap_or_else(|| {
            if !self.punycode.is_empty() {
                f.write_str("punycode{")?;

                // Reconstruct a standard Punycode encoding,
                // by using `-` as the separator.
                if !self.ascii.is_empty() {
                    f.write_str(self.ascii)?;
                    f.write_str("-")?;
                }
                f.write_str(self.punycode)?;

                f.write_str("}")
            } else {
                f.write_str(self.ascii)
            }
        })
    }
}

/// Sequence of lowercase hexadecimal nibbles (`0-9a-f`), used by leaf consts.
struct HexNibbles<'s> {
    nibbles: &'s str,
}

impl<'s> HexNibbles<'s> {
    /// Decode an integer value (with the "most significant nibble" first),
    /// returning `None` if it can't fit in an `u64`.
    // FIXME(eddyb) should this "just" use `u128` instead?
    fn try_parse_uint(&self) -> Option<u64> {
        let nibbles = self.nibbles.trim_start_matches("0");

        if nibbles.len() > 16 {
            return None;
        }

        let mut v = 0;
        for nibble in nibbles.chars() {
            v = (v << 4) | (nibble.to_digit(16).unwrap() as u64);
        }
        Some(v)
    }

    /// Decode a UTF-8 byte sequence (with each byte using a pair of nibbles)
    /// into individual `char`s, returning `None` for invalid UTF-8.
    fn try_parse_str_chars(&self) -> Option<impl Iterator<Item = char> + 's> {
        if self.nibbles.len() % 2 != 0 {
            return None;
        }

        // FIXME(eddyb) use `array_chunks` instead, when that becomes stable.
        let mut bytes = self
            .nibbles
            .as_bytes()
            .chunks_exact(2)
            .map(|slice| match slice {
                [a, b] => [a, b],
                _ => unreachable!(),
            })
            .map(|[&hi, &lo]| {
                let half = |nibble: u8| (nibble as char).to_digit(16).unwrap() as u8;
                (half(hi) << 4) | half(lo)
            });

        let chars = iter::from_fn(move || {
            // As long as there are any bytes left, there's at least one more
            // UTF-8-encoded `char` to decode (or the possibility of error).
            bytes.next().map(|first_byte| -> Result<char, ()> {
                // FIXME(eddyb) this `enum` and `fn` should be somewhere in `core`.
                enum Utf8FirstByteError {
                    ContinuationByte,
                    TooLong,
                }
                fn utf8_len_from_first_byte(byte: u8) -> Result<usize, Utf8FirstByteError> {
                    match byte {
                        0x00..=0x7f => Ok(1),
                        0x80..=0xbf => Err(Utf8FirstByteError::ContinuationByte),
                        0xc0..=0xdf => Ok(2),
                        0xe0..=0xef => Ok(3),
                        0xf0..=0xf7 => Ok(4),
                        0xf8..=0xff => Err(Utf8FirstByteError::TooLong),
                    }
                }

                // Collect the appropriate amount of bytes (up to 4), according
                // to the UTF-8 length implied by the first byte.
                let utf8_len = utf8_len_from_first_byte(first_byte).map_err(|_| ())?;
                let utf8 = &mut [first_byte, 0, 0, 0][..utf8_len];
                for i in 1..utf8_len {
                    utf8[i] = bytes.next().ok_or(())?;
                }

                // Fully validate the UTF-8 sequence.
                let s = str::from_utf8(utf8).map_err(|_| ())?;

                // Since we included exactly one UTF-8 sequence, and validation
                // succeeded, `str::chars` should return exactly one `char`.
                let mut chars = s.chars();
                match (chars.next(), chars.next()) {
                    (Some(c), None) => Ok(c),
                    _ => unreachable!(
                        "str::from_utf8({:?}) = {:?} was expected to have 1 char, \
                         but {} chars were found",
                        utf8,
                        s,
                        s.chars().count()
                    ),
                }
            })
        });

        // HACK(eddyb) doing a separate validation iteration like this might be
        // wasteful, but it's easier to avoid starting to print a string literal
        // in the first place, than to abort it mid-string.
        if chars.clone().any(|r| r.is_err()) {
            None
        } else {
            Some(chars.map(Result::unwrap))
        }
    }
}

fn basic_type(tag: u8) -> Option<&'static str> {
    Some(match tag {
        b'b' => "bool",
        b'c' => "char",
        b'e' => "str",
        b'u' => "()",
        b'a' => "i8",
        b's' => "i16",
        b'l' => "i32",
        b'x' => "i64",
        b'n' => "i128",
        b'i' => "isize",
        b'h' => "u8",
        b't' => "u16",
        b'm' => "u32",
        b'y' => "u64",
        b'o' => "u128",
        b'j' => "usize",
        b'f' => "f32",
        b'd' => "f64",
        b'z' => "!",
        b'p' => "_",
        b'v' => "...",

        _ => return None,
    })
}

struct Parser<'s> {
    sym: &'s str,
    next: usize,
    depth: u32,
}

impl<'s> Parser<'s> {
    fn push_depth(&mut self) -> Result<(), ParseError> {
        self.depth += 1;
        if self.depth > MAX_DEPTH {
            Err(ParseError::RecursedTooDeep)
        } else {
            Ok(())
        }
    }

    fn pop_depth(&mut self) {
        self.depth -= 1;
    }

    fn peek(&self) -> Option<u8> {
        self.sym.as_bytes().get(self.next).cloned()
    }

    fn eat(&mut self, b: u8) -> bool {
        if self.peek() == Some(b) {
            self.next += 1;
            true
        } else {
            false
        }
    }

    fn next(&mut self) -> Result<u8, ParseError> {
        let b = self.peek().ok_or(ParseError::Invalid)?;
        self.next += 1;
        Ok(b)
    }

    fn hex_nibbles(&mut self) -> Result<HexNibbles<'s>, ParseError> {
        let start = self.next;
        loop {
            match self.next()? {
                b'0'..=b'9' | b'a'..=b'f' => {}
                b'_' => break,
                _ => return Err(ParseError::Invalid),
            }
        }
        Ok(HexNibbles {
            nibbles: &self.sym[start..self.next - 1],
        })
    }

    fn digit_10(&mut self) -> Result<u8, ParseError> {
        let d = match self.peek() {
            Some(d @ b'0'..=b'9') => d - b'0',
            _ => return Err(ParseError::Invalid),
        };
        self.next += 1;
        Ok(d)
    }

    fn digit_62(&mut self) -> Result<u8, ParseError> {
        let d = match self.peek() {
            Some(d @ b'0'..=b'9') => d - b'0',
            Some(d @ b'a'..=b'z') => 10 + (d - b'a'),
            Some(d @ b'A'..=b'Z') => 10 + 26 + (d - b'A'),
            _ => return Err(ParseError::Invalid),
        };
        self.next += 1;
        Ok(d)
    }

    fn integer_62(&mut self) -> Result<u64, ParseError> {
        if self.eat(b'_') {
            return Ok(0);
        }

        let mut x: u64 = 0;
        while !self.eat(b'_') {
            let d = self.digit_62()? as u64;
            x = x.checked_mul(62).ok_or(ParseError::Invalid)?;
            x = x.checked_add(d).ok_or(ParseError::Invalid)?;
        }
        x.checked_add(1).ok_or(ParseError::Invalid)
    }

    fn opt_integer_62(&mut self, tag: u8) -> Result<u64, ParseError> {
        if !self.eat(tag) {
            return Ok(0);
        }
        self.integer_62()?.checked_add(1).ok_or(ParseError::Invalid)
    }

    fn disambiguator(&mut self) -> Result<u64, ParseError> {
        self.opt_integer_62(b's')
    }

    fn namespace(&mut self) -> Result<Option<char>, ParseError> {
        match self.next()? {
            // Special namespaces, like closures and shims.
            ns @ b'A'..=b'Z' => Ok(Some(ns as char)),

            // Implementation-specific/unspecified namespaces.
            b'a'..=b'z' => Ok(None),

            _ => Err(ParseError::Invalid),
        }
    }

    fn backref(&mut self) -> Result<Parser<'s>, ParseError> {
        let s_start = self.next - 1;
        let i = self.integer_62()?;
        if i >= s_start as u64 {
            return Err(ParseError::Invalid);
        }
        let mut new_parser = Parser {
            sym: self.sym,
            next: i as usize,
            depth: self.depth,
        };
        new_parser.push_depth()?;
        Ok(new_parser)
    }

    fn ident(&mut self) -> Result<Ident<'s>, ParseError> {
        let is_punycode = self.eat(b'u');
        let mut len = self.digit_10()? as usize;
        if len != 0 {
            while let Ok(d) = self.digit_10() {
                len = len.checked_mul(10).ok_or(ParseError::Invalid)?;
                len = len.checked_add(d as usize).ok_or(ParseError::Invalid)?;
            }
        }

        // Skip past the optional `_` separator.
        self.eat(b'_');

        let start = self.next;
        self.next = self.next.checked_add(len).ok_or(ParseError::Invalid)?;
        if self.next > self.sym.len() {
            return Err(ParseError::Invalid);
        }

        let ident = &self.sym[start..self.next];

        if is_punycode {
            let ident = match ident.bytes().rposition(|b| b == b'_') {
                Some(i) => Ident {
                    ascii: &ident[..i],
                    punycode: &ident[i + 1..],
                },
                None => Ident {
                    ascii: "",
                    punycode: ident,
                },
            };
            if ident.punycode.is_empty() {
                return Err(ParseError::Invalid);
            }
            Ok(ident)
        } else {
            Ok(Ident {
                ascii: ident,
                punycode: "",
            })
        }
    }
}

struct Printer<'a, 'b: 'a, 's> {
    /// The input parser to demangle from, or `Err` if any (parse) error was
    /// encountered (in order to disallow further likely-incorrect demangling).
    ///
    /// See also the documentation on the `invalid!` and `parse!` macros below.
    parser: Result<Parser<'s>, ParseError>,

    /// The output formatter to demangle to, or `None` while skipping printing.
    out: Option<&'a mut fmt::Formatter<'b>>,

    /// Cumulative number of lifetimes bound by `for<...>` binders ('G'),
    /// anywhere "around" the current entity (e.g. type) being demangled.
    /// This value is not tracked while skipping printing, as it'd be unused.
    ///
    /// See also the documentation on the `Printer::in_binder` method.
    bound_lifetime_depth: u32,
}

impl ParseError {
    /// Snippet to print when the error is initially encountered.
    fn message(&self) -> &str {
        match self {
            ParseError::Invalid => "{invalid syntax}",
            ParseError::RecursedTooDeep => "{recursion limit reached}",
        }
    }
}

/// Mark the parser as errored (with `ParseError::Invalid`), print the
/// appropriate message (see `ParseError::message`) and return early.
macro_rules! invalid {
    ($printer:ident) => {{
        let err = ParseError::Invalid;
        $printer.print(err.message())?;
        $printer.parser = Err(err);
        return Ok(());
    }};
}

/// Call a parser method (if the parser hasn't errored yet),
/// and mark the parser as errored if it returns `Err`.
///
/// If the parser errored, before or now, this returns early,
/// from the current function, after printing either:
/// * for a new error, the appropriate message (see `ParseError::message`)
/// * for an earlier error, only `?` -  this allows callers to keep printing
///   the approximate syntax of the path/type/const, despite having errors,
///   e.g. `Vec<[(A, ?); ?]>` instead of `Vec<[(A, ?`
macro_rules! parse {
    ($printer:ident, $method:ident $(($($arg:expr),*))*) => {
        match $printer.parser {
            Ok(ref mut parser) => match parser.$method($($($arg),*)*) {
                Ok(x) => x,
                Err(err) => {
                    $printer.print(err.message())?;
                    $printer.parser = Err(err);
                    return Ok(());
                }
            }
            Err(_) => return $printer.print("?"),
        }
    };
}

impl<'a, 'b, 's> Printer<'a, 'b, 's> {
    /// Eat the given character from the parser,
    /// returning `false` if the parser errored.
    fn eat(&mut self, b: u8) -> bool {
        self.parser.as_mut().map(|p| p.eat(b)) == Ok(true)
    }

    /// Skip printing (i.e. `self.out` will be `None`) for the duration of the
    /// given closure. This should not change parsing behavior, only disable the
    /// output, but there may be optimizations (such as not traversing backrefs).
    fn skipping_printing<F>(&mut self, f: F)
    where
        F: FnOnce(&mut Self) -> fmt::Result,
    {
        let orig_out = self.out.take();
        f(self).expect("`fmt::Error`s should be impossible without a `fmt::Formatter`");
        self.out = orig_out;
    }

    /// Print the target of a backref, using the given closure.
    /// When printing is being skipped, the backref will only be parsed,
    /// ignoring the backref's target completely.
    fn print_backref<F>(&mut self, f: F) -> fmt::Result
    where
        F: FnOnce(&mut Self) -> fmt::Result,
    {
        let backref_parser = parse!(self, backref);

        if self.out.is_none() {
            return Ok(());
        }

        let orig_parser = mem::replace(&mut self.parser, Ok(backref_parser));
        let r = f(self);
        self.parser = orig_parser;
        r
    }

    fn pop_depth(&mut self) {
        if let Ok(ref mut parser) = self.parser {
            parser.pop_depth();
        }
    }

    /// Output the given value to `self.out` (using `fmt::Display` formatting),
    /// if printing isn't being skipped.
    fn print(&mut self, x: impl fmt::Display) -> fmt::Result {
        if let Some(out) = &mut self.out {
            fmt::Display::fmt(&x, out)?;
        }
        Ok(())
    }

    /// Output the given `char`s (escaped using `char::escape_debug`), with the
    /// whole sequence wrapped in quotes, for either a `char` or `&str` literal,
    /// if printing isn't being skipped.
    fn print_quoted_escaped_chars(
        &mut self,
        quote: char,
        chars: impl Iterator<Item = char>,
    ) -> fmt::Result {
        if let Some(out) = &mut self.out {
            use core::fmt::Write;

            out.write_char(quote)?;
            for c in chars {
                // Special-case not escaping a single/double quote, when
                // inside the opposite kind of quote.
                if matches!((quote, c), ('\'', '"') | ('"', '\'')) {
                    out.write_char(c)?;
                    continue;
                }

                for escaped in c.escape_debug() {
                    out.write_char(escaped)?;
                }
            }
            out.write_char(quote)?;
        }
        Ok(())
    }

    /// Print the lifetime according to the previously decoded index.
    /// An index of `0` always refers to `'_`, but starting with `1`,
    /// indices refer to late-bound lifetimes introduced by a binder.
    fn print_lifetime_from_index(&mut self, lt: u64) -> fmt::Result {
        // Bound lifetimes aren't tracked when skipping printing.
        if self.out.is_none() {
            return Ok(());
        }

        self.print("'")?;
        if lt == 0 {
            return self.print("_");
        }
        match (self.bound_lifetime_depth as u64).checked_sub(lt) {
            Some(depth) => {
                // Try to print lifetimes alphabetically first.
                if depth < 26 {
                    let c = (b'a' + depth as u8) as char;
                    self.print(c)
                } else {
                    // Use `'_123` after running out of letters.
                    self.print("_")?;
                    self.print(depth)
                }
            }
            None => invalid!(self),
        }
    }

    /// Optionally enter a binder ('G') for late-bound lifetimes,
    /// printing e.g. `for<'a, 'b> ` before calling the closure,
    /// and make those lifetimes visible to it (via depth level).
    fn in_binder<F>(&mut self, f: F) -> fmt::Result
    where
        F: FnOnce(&mut Self) -> fmt::Result,
    {
        let bound_lifetimes = parse!(self, opt_integer_62(b'G'));

        // Don't track bound lifetimes when skipping printing.
        if self.out.is_none() {
            return f(self);
        }

        if bound_lifetimes > 0 {
            self.print("for<")?;
            for i in 0..bound_lifetimes {
                if i > 0 {
                    self.print(", ")?;
                }
                self.bound_lifetime_depth += 1;
                self.print_lifetime_from_index(1)?;
            }
            self.print("> ")?;
        }

        let r = f(self);

        // Restore `bound_lifetime_depth` to the previous value.
        self.bound_lifetime_depth -= bound_lifetimes as u32;

        r
    }

    /// Print list elements using the given closure and separator,
    /// until the end of the list ('E') is found, or the parser errors.
    /// Returns the number of elements printed.
    fn print_sep_list<F>(&mut self, f: F, sep: &str) -> Result<usize, fmt::Error>
    where
        F: Fn(&mut Self) -> fmt::Result,
    {
        let mut i = 0;
        while self.parser.is_ok() && !self.eat(b'E') {
            if i > 0 {
                self.print(sep)?;
            }
            f(self)?;
            i += 1;
        }
        Ok(i)
    }

    fn print_path(&mut self, in_value: bool) -> fmt::Result {
        parse!(self, push_depth);

        let tag = parse!(self, next);
        match tag {
            b'C' => {
                let dis = parse!(self, disambiguator);
                let name = parse!(self, ident);

                self.print(name)?;
                if let Some(out) = &mut self.out {
                    if !out.alternate() && dis != 0 {
                        out.write_str("[")?;
                        fmt::LowerHex::fmt(&dis, out)?;
                        out.write_str("]")?;
                    }
                }
            }
            b'N' => {
                let ns = parse!(self, namespace);

                self.print_path(in_value)?;

                // HACK(eddyb) if the parser is already marked as having errored,
                // `parse!` below will print a `?` without its preceding `::`
                // (because printing the `::` is skipped in certain conditions,
                // i.e. a lowercase namespace with an empty identifier),
                // so in order to get `::?`, the `::` has to be printed here.
                if self.parser.is_err() {
                    self.print("::")?;
                }

                let dis = parse!(self, disambiguator);
                let name = parse!(self, ident);

                match ns {
                    // Special namespaces, like closures and shims.
                    Some(ns) => {
                        self.print("::{")?;
                        match ns {
                            'C' => self.print("closure")?,
                            'S' => self.print("shim")?,
                            _ => self.print(ns)?,
                        }
                        if !name.ascii.is_empty() || !name.punycode.is_empty() {
                            self.print(":")?;
                            self.print(name)?;
                        }
                        self.print("#")?;
                        self.print(dis)?;
                        self.print("}")?;
                    }

                    // Implementation-specific/unspecified namespaces.
                    None => {
                        if !name.ascii.is_empty() || !name.punycode.is_empty() {
                            self.print("::")?;
                            self.print(name)?;
                        }
                    }
                }
            }
            b'M' | b'X' | b'Y' => {
                if tag != b'Y' {
                    // Ignore the `impl`'s own path.
                    parse!(self, disambiguator);
                    self.skipping_printing(|this| this.print_path(false));
                }

                self.print("<")?;
                self.print_type()?;
                if tag != b'M' {
                    self.print(" as ")?;
                    self.print_path(false)?;
                }
                self.print(">")?;
            }
            b'I' => {
                self.print_path(in_value)?;
                if in_value {
                    self.print("::")?;
                }
                self.print("<")?;
                self.print_sep_list(Self::print_generic_arg, ", ")?;
                self.print(">")?;
            }
            b'B' => {
                self.print_backref(|this| this.print_path(in_value))?;
            }
            _ => invalid!(self),
        }

        self.pop_depth();
        Ok(())
    }

    fn print_generic_arg(&mut self) -> fmt::Result {
        if self.eat(b'L') {
            let lt = parse!(self, integer_62);
            self.print_lifetime_from_index(lt)
        } else if self.eat(b'K') {
            self.print_const(false)
        } else {
            self.print_type()
        }
    }

    fn print_type(&mut self) -> fmt::Result {
        // The splat feature is unstable and its mangling is subject to change
        // FIXME(splat):
        // - for efficiency we might want to use a letter that can't occur in any type, rather than
        //   taking an unused letter
        // - splat isn't implemented for legacy mangling
        if self.eat(b'w') {
            self.print("#[splat] ")?;
        }
        let tag = parse!(self, next);

        if let Some(ty) = basic_type(tag) {
            return self.print(ty);
        }

        parse!(self, push_depth);

        match tag {
            b'R' | b'Q' => {
                self.print("&")?;
                if self.eat(b'L') {
                    let lt = parse!(self, integer_62);
                    if lt != 0 {
                        self.print_lifetime_from_index(lt)?;
                        self.print(" ")?;
                    }
                }
                if tag != b'R' {
                    self.print("mut ")?;
                }
                self.print_type()?;
            }

            b'P' | b'O' => {
                self.print("*")?;
                if tag != b'P' {
                    self.print("mut ")?;
                } else {
                    self.print("const ")?;
                }
                self.print_type()?;
            }

            b'A' | b'S' => {
                self.print("[")?;
                self.print_type()?;
                if tag == b'A' {
                    self.print("; ")?;
                    self.print_const(true)?;
                }
                self.print("]")?;
            }
            b'T' => {
                self.print("(")?;
                let count = self.print_sep_list(Self::print_type, ", ")?;
                if count == 1 {
                    self.print(",")?;
                }
                self.print(")")?;
            }
            b'F' => self.in_binder(|this| {
                let is_unsafe = this.eat(b'U');
                let abi = if this.eat(b'K') {
                    if this.eat(b'C') {
                        Some("C")
                    } else {
                        let abi = parse!(this, ident);
                        if abi.ascii.is_empty() || !abi.punycode.is_empty() {
                            invalid!(this);
                        }
                        Some(abi.ascii)
                    }
                } else {
                    None
                };

                if is_unsafe {
                    this.print("unsafe ")?;
                }

                if let Some(abi) = abi {
                    this.print("extern \"")?;

                    // If the ABI had any `-`, they were replaced with `_`,
                    // so the parts between `_` have to be re-joined with `-`.
                    let mut parts = abi.split('_');
                    this.print(parts.next().unwrap())?;
                    for part in parts {
                        this.print("-")?;
                        this.print(part)?;
                    }

                    this.print("\" ")?;
                }

                this.print("fn(")?;
                this.print_sep_list(Self::print_type, ", ")?;
                this.print(")")?;

                if this.eat(b'u') {
                    // Skip printing the return type if it's 'u', i.e. `()`.
                } else {
                    this.print(" -> ")?;
                    this.print_type()?;
                }

                Ok(())
            })?,
            b'D' => {
                self.print("dyn ")?;
                self.in_binder(|this| {
                    this.print_sep_list(Self::print_dyn_trait, " + ")?;
                    Ok(())
                })?;

                if !self.eat(b'L') {
                    invalid!(self);
                }
                let lt = parse!(self, integer_62);
                if lt != 0 {
                    self.print(" + ")?;
                    self.print_lifetime_from_index(lt)?;
                }
            }
            b'B' => {
                self.print_backref(Self::print_type)?;
            }
            b'W' => {
                self.print_type()?;
                self.print(" is ")?;
                self.print_pat()?;
            }
            _ => {
                // Go back to the tag, so `print_path` also sees it.
                let _ = self.parser.as_mut().map(|p| p.next -= 1);
                self.print_path(false)?;
            }
        }

        self.pop_depth();
        Ok(())
    }

    /// A trait in a trait object may have some "existential projections"
    /// (i.e. associated type bindings) after it, which should be printed
    /// in the `<...>` of the trait, e.g. `dyn Trait<T, U, Assoc=X>`.
    /// To this end, this method will keep the `<...>` of an 'I' path
    /// open, by omitting the `>`, and return `Ok(true)` in that case.
    fn print_path_maybe_open_generics(&mut self) -> Result<bool, fmt::Error> {
        if self.eat(b'B') {
            // NOTE(eddyb) the closure may not run if printing is being skipped,
            // but in that case the returned boolean doesn't matter.
            let mut open = false;
            self.print_backref(|this| {
                open = this.print_path_maybe_open_generics()?;
                Ok(())
            })?;
            Ok(open)
        } else if self.eat(b'I') {
            self.print_path(false)?;
            self.print("<")?;
            self.print_sep_list(Self::print_generic_arg, ", ")?;
            Ok(true)
        } else {
            self.print_path(false)?;
            Ok(false)
        }
    }

    fn print_dyn_trait(&mut self) -> fmt::Result {
        let mut open = self.print_path_maybe_open_generics()?;

        while self.eat(b'p') {
            if !open {
                self.print("<")?;
                open = true;
            } else {
                self.print(", ")?;
            }

            let name = parse!(self, ident);
            self.print(name)?;
            self.print(" = ")?;
            if self.eat(b'K') {
                self.print_const(false)
            } else {
                self.print_type()
            }?;
        }

        if open {
            self.print(">")?;
        }

        Ok(())
    }

    fn print_pat(&mut self) -> fmt::Result {
        let tag = parse!(self, next);

        match tag {
            b'R' => {
                self.print_const(false)?;
                self.print("..=")?;
                self.print_const(false)?;
            }
            b'O' => {
                parse!(self, push_depth);
                self.print_pat()?;
                while !self.eat(b'E') {
                    // May have reached the end of the string,
                    // avoid going into an endless loop.
                    if self.parser.is_err() {
                        invalid!(self)
                    }
                    self.print(" | ")?;
                    self.print_pat()?;
                }
                self.pop_depth();
            }
            b'N' => self.print("!null")?,
            _ => invalid!(self),
        }

        Ok(())
    }

    fn print_const(&mut self, in_value: bool) -> fmt::Result {
        let tag = parse!(self, next);

        parse!(self, push_depth);

        // Only literals (and the names of `const` generic parameters, but they
        // don't get mangled at all), can appear in generic argument position
        // without any disambiguation, all other expressions require braces.
        // To avoid duplicating the mapping between `tag` and what syntax gets
        // used (especially any special-casing), every case that needs braces
        // has to call `open_brace(self)?` (and the closing brace is automatic).
        let mut opened_brace = false;
        let mut open_brace_if_outside_expr = |this: &mut Self| {
            // If this expression is nested in another, braces aren't required.
            if in_value {
                return Ok(());
            }

            opened_brace = true;
            this.print("{")
        };

        match tag {
            b'p' => self.print("_")?,

            // Primitive leaves with hex-encoded values (see `basic_type`).
            b'h' | b't' | b'm' | b'y' | b'o' | b'j' => self.print_const_uint(tag)?,
            b'a' | b's' | b'l' | b'x' | b'n' | b'i' => {
                if self.eat(b'n') {
                    self.print("-")?;
                }

                self.print_const_uint(tag)?;
            }
            b'b' => match parse!(self, hex_nibbles).try_parse_uint() {
                Some(0) => self.print("false")?,
                Some(1) => self.print("true")?,
                _ => invalid!(self),
            },
            b'c' => {
                let valid_char = parse!(self, hex_nibbles)
                    .try_parse_uint()
                    .and_then(|v| u32::try_from(v).ok())
                    .and_then(char::from_u32);
                match valid_char {
                    Some(c) => self.print_quoted_escaped_chars('\'', iter::once(c))?,
                    None => invalid!(self),
                }
            }
            b'e' => {
                // NOTE(eddyb) a string literal `"..."` has type `&str`, so
                // to get back the type `str`, `*"..."` syntax is needed
                // (even if that may not be valid in Rust itself).
                open_brace_if_outside_expr(self)?;
                self.print("*")?;

                self.print_const_str_literal()?;
            }

            b'R' | b'Q' => {
                // NOTE(eddyb) this prints `"..."` instead of `&*"..."`, which
                // is what `Re..._` would imply (see comment for `str` above).
                if tag == b'R' && self.eat(b'e') {
                    self.print_const_str_literal()?;
                } else {
                    open_brace_if_outside_expr(self)?;
                    self.print("&")?;
                    if tag != b'R' {
                        self.print("mut ")?;
                    }
                    self.print_const(true)?;
                }
            }
            b'A' => {
                open_brace_if_outside_expr(self)?;
                self.print("[")?;
                self.print_sep_list(|this| this.print_const(true), ", ")?;
                self.print("]")?;
            }
            b'T' => {
                open_brace_if_outside_expr(self)?;
                self.print("(")?;
                let count = self.print_sep_list(|this| this.print_const(true), ", ")?;
                if count == 1 {
                    self.print(",")?;
                }
                self.print(")")?;
            }
            b'V' => {
                open_brace_if_outside_expr(self)?;
                self.print_path(true)?;
                match parse!(self, next) {
                    b'U' => {}
                    b'T' => {
                        self.print("(")?;
                        self.print_sep_list(|this| this.print_const(true), ", ")?;
                        self.print(")")?;
                    }
                    b'S' => {
                        self.print(" { ")?;
                        self.print_sep_list(
                            |this| {
                                parse!(this, disambiguator);
                                let name = parse!(this, ident);
                                this.print(name)?;
                                this.print(": ")?;
                                this.print_const(true)
                            },
                            ", ",
                        )?;
                        self.print(" }")?;
                    }
                    _ => invalid!(self),
                }
            }
            b'B' => {
                self.print_backref(|this| this.print_const(in_value))?;
            }
            _ => invalid!(self),
        }

        if opened_brace {
            self.print("}")?;
        }

        self.pop_depth();
        Ok(())
    }

    fn print_const_uint(&mut self, ty_tag: u8) -> fmt::Result {
        let hex = parse!(self, hex_nibbles);

        match hex.try_parse_uint() {
            Some(v) => self.print(v)?,

            // Print anything that doesn't fit in `u64` verbatim.
            None => {
                self.print("0x")?;
                self.print(hex.nibbles)?;
            }
        }

        if let Some(out) = &mut self.out {
            if !out.alternate() {
                let ty = basic_type(ty_tag).unwrap();
                self.print(ty)?;
            }
        }

        Ok(())
    }

    fn print_const_str_literal(&mut self) -> fmt::Result {
        match parse!(self, hex_nibbles).try_parse_str_chars() {
            Some(chars) => self.print_quoted_escaped_chars('"', chars),
            None => invalid!(self),
        }
    }
}

#[cfg(test)]
mod tests {
    use std::prelude::v1::*;

    macro_rules! t {
        ($a:expr, $b:expr) => {{
            assert_eq!(format!("{}", ::demangle($a)), $b);
        }};
    }
    macro_rules! t_nohash {
        ($a:expr, $b:expr) => {{
            assert_eq!(format!("{:#}", ::demangle($a)), $b);
        }};
    }
    macro_rules! t_nohash_type {
        ($a:expr, $b:expr) => {
            t_nohash!(concat!("_RMC0", $a), concat!("<", $b, ">"))
        };
    }
    macro_rules! t_const {
        ($mangled:expr, $value:expr) => {
            t_nohash!(
                concat!("_RIC0K", $mangled, "E"),
                concat!("::<", $value, ">")
            )
        };
    }
    macro_rules! t_const_suffixed {
        ($mangled:expr, $value:expr, $value_ty_suffix:expr) => {{
            t_const!($mangled, $value);
            t!(
                concat!("_RIC0K", $mangled, "E"),
                concat!("::<", $value, $value_ty_suffix, ">")
            );
        }};
    }

    #[test]
    fn demangle_crate_with_leading_digit() {
        t_nohash!("_RNvC6_123foo3bar", "123foo::bar");
    }

    #[test]
    fn demangle_crate_with_zero_disambiguator() {
        t!("_RC4f128", "f128");
        t_nohash!("_RC4f128", "f128");
    }

    #[test]
    fn demangle_utf8_idents() {
        t_nohash!(
            "_RNqCs4fqI2P2rA04_11utf8_identsu30____7hkackfecea1cbdathfdh9hlq6y",
            "utf8_idents::საჭმელად_გემრიელი_სადილი"
        );
    }

    #[test]
    fn demangle_closure() {
        t_nohash!(
            "_RNCNCNgCs6DXkGYLi8lr_2cc5spawn00B5_",
            "cc::spawn::{closure#0}::{closure#0}"
        );
        t_nohash!(
            "_RNCINkXs25_NgCsbmNqQUJIY6D_4core5sliceINyB9_4IterhENuNgNoBb_4iter8iterator8Iterator9rpositionNCNgNpB9_6memchr7memrchrs_0E0Bb_",
            "<core::slice::Iter<u8> as core::iter::iterator::Iterator>::rposition::<core::slice::memchr::memrchr::{closure#1}>::{closure#0}"
        );
    }

    #[test]
    fn demangle_dyn_trait() {
        t_nohash!(
            "_RINbNbCskIICzLVDPPb_5alloc5alloc8box_freeDINbNiB4_5boxed5FnBoxuEp6OutputuEL_ECs1iopQbuBiw2_3std",
            "alloc::alloc::box_free::<dyn alloc::boxed::FnBox<(), Output = ()>>"
        );
    }

    #[test]
    fn demangle_dyn_trait_assoc_const_binding() {
        t_nohash_type!("DNtC5krate5Traitp1NKj0_EL_", "dyn krate::Trait<N = 0>");
    }

    #[test]
    fn demangle_pat_ty() {
        t_nohash_type!("WmRm1_m9_", "u32 is 1..=9");
        t_nohash_type!("WmORm1_m2_Rm5_m6_E", "u32 is 1..=2 | 5..=6");
        assert!(::v0::demangle("_RMC0WmORm1_m2_Rm5_m6_").is_err());
    }

    #[test]
    fn demangle_splat() {
        t_nohash!(
            "_RNvMNtCs5CcWRJzwAYz_4core6optionINtB2_6OptionFwTlEEuE6unwrapCsiksvdpJ6Jnj_14splat_mangling",
            "<core::option::Option<fn(#[splat] (i32,))>>::unwrap"
        );
        t_nohash!(
            "_RMNvCsiksvdpJ6Jnj_14splat_mangling4mainINtB0_4TypeFwThmEEuE",
            "<splat_mangling::main::Type<fn(#[splat] (u8, u32))>>"
        );
        t_nohash!(
            "_RMs0_NvCsiksvdpJ6Jnj_14splat_mangling4mainINtB3_4TypeFwTThmEEEuE",
            "<splat_mangling::main::Type<fn(#[splat] ((u8, u32),))>>"
        );
        t_nohash!(
            "_RMs2_NvCsiksvdpJ6Jnj_14splat_mangling4mainINtB3_4TypePFwTmaEEuE",
            "<splat_mangling::main::Type<*const fn(#[splat] (u32, i8))>>"
        );
        t_nohash!(
            "_RMs4_NvCsiksvdpJ6Jnj_14splat_mangling4mainINtB3_4TypeFwTmaEdEuE",
            "<splat_mangling::main::Type<fn(#[splat] (u32, i8), f64)>>"
        );
    }

    #[test]
    fn demangle_const_generics_preview() {
        // NOTE(eddyb) this was hand-written, before rustc had working
        // const generics support (but the mangling format did include them).
        t_nohash_type!(
            "INtC8arrayvec8ArrayVechKj7b_E",
            "arrayvec::ArrayVec<u8, 123>"
        );
        t_const_suffixed!("j7b_", "123", "usize");
    }

    #[test]
    fn demangle_min_const_generics() {
        t_const!("p", "_");
        t_const_suffixed!("hb_", "11", "u8");
        t_const_suffixed!("off00ff00ff00ff00ff_", "0xff00ff00ff00ff00ff", "u128");
        t_const_suffixed!("s98_", "152", "i16");
        t_const_suffixed!("anb_", "-11", "i8");
        t_const!("b0_", "false");
        t_const!("b1_", "true");
        t_const!("c76_", "'v'");
        t_const!("c22_", r#"'"'"#);
        t_const!("ca_", "'\\n'");
        t_const!("c2202_", "'∂'");
    }

    #[test]
    fn demangle_const_str() {
        t_const!("e616263_", "{*\"abc\"}");
        t_const!("e27_", r#"{*"'"}"#);
        t_const!("e090a_", "{*\"\\t\\n\"}");
        t_const!("ee28882c3bc_", "{*\"∂ü\"}");
        t_const!(
            "ee183a1e18390e183ade1839be18394e1839ae18390e183935fe18392e18394e1839b\
              e183a0e18398e18394e1839ae183985fe183a1e18390e18393e18398e1839ae18398_",
            "{*\"საჭმელად_გემრიელი_სადილი\"}"
        );
        t_const!(
            "ef09f908af09fa688f09fa686f09f90ae20c2a720f09f90b6f09f9192e298\
              95f09f94a520c2a720f09fa7a1f09f929bf09f929af09f9299f09f929c_",
            "{*\"🐊🦈🦆🐮 § 🐶👒☕🔥 § 🧡💛💚💙💜\"}"
        );
    }

    // NOTE(eddyb) this uses the same strings as `demangle_const_str` and should
    // be kept in sync with it - while a macro could be used to generate both
    // `str` and `&str` tests, from a single list of strings, this seems clearer.
    #[test]
    fn demangle_const_ref_str() {
        t_const!("Re616263_", "\"abc\"");
        t_const!("Re27_", r#""'""#);
        t_const!("Re090a_", "\"\\t\\n\"");
        t_const!("Ree28882c3bc_", "\"∂ü\"");
        t_const!(
            "Ree183a1e18390e183ade1839be18394e1839ae18390e183935fe18392e18394e1839b\
               e183a0e18398e18394e1839ae183985fe183a1e18390e18393e18398e1839ae18398_",
            "\"საჭმელად_გემრიელი_სადილი\""
        );
        t_const!(
            "Ref09f908af09fa688f09fa686f09f90ae20c2a720f09f90b6f09f9192e298\
               95f09f94a520c2a720f09fa7a1f09f929bf09f929af09f9299f09f929c_",
            "\"🐊🦈🦆🐮 § 🐶👒☕🔥 § 🧡💛💚💙💜\""
        );
    }

    #[test]
    fn demangle_const_ref() {
        t_const!("Rp", "{&_}");
        t_const!("Rh7b_", "{&123}");
        t_const!("Rb0_", "{&false}");
        t_const!("Rc58_", "{&'X'}");
        t_const!("RRRh0_", "{&&&0}");
        t_const!("RRRe_", "{&&\"\"}");
        t_const!("QAE", "{&mut []}");
    }

    #[test]
    fn demangle_const_array() {
        t_const!("AE", "{[]}");
        t_const!("Aj0_E", "{[0]}");
        t_const!("Ah1_h2_h3_E", "{[1, 2, 3]}");
        t_const!("ARe61_Re62_Re63_E", "{[\"a\", \"b\", \"c\"]}");
        t_const!("AAh1_h2_EAh3_h4_EE", "{[[1, 2], [3, 4]]}");
    }

    #[test]
    fn demangle_const_tuple() {
        t_const!("TE", "{()}");
        t_const!("Tj0_E", "{(0,)}");
        t_const!("Th1_b0_E", "{(1, false)}");
        t_const!(
            "TRe616263_c78_RAh1_h2_h3_EE",
            "{(\"abc\", 'x', &[1, 2, 3])}"
        );
    }

    #[test]
    fn demangle_const_adt() {
        t_const!(
            "VNvINtNtC4core6option6OptionjE4NoneU",
            "{core::option::Option::<usize>::None}"
        );
        t_const!(
            "VNvINtNtC4core6option6OptionjE4SomeTj0_E",
            "{core::option::Option::<usize>::Some(0)}"
        );
        t_const!(
            "VNtC3foo3BarS1sRe616263_2chc78_5sliceRAh1_h2_h3_EE",
            "{foo::Bar { s: \"abc\", ch: 'x', slice: &[1, 2, 3] }}"
        );
    }

    #[test]
    fn demangle_exponential_explosion() {
        // NOTE(eddyb) because of the prefix added by `t_nohash_type!` is
        // 3 bytes long, `B2_` refers to the start of the type, not `B_`.
        // 6 backrefs (`B8_E` through `B3_E`) result in 2^6 = 64 copies of `_`.
        // Also, because the `p` (`_`) type is after all of the starts of the
        // backrefs, it can be replaced with any other type, independently.
        t_nohash_type!(
            concat!("TTTTTT", "p", "B8_E", "B7_E", "B6_E", "B5_E", "B4_E", "B3_E"),
            "((((((_, _), (_, _)), ((_, _), (_, _))), (((_, _), (_, _)), ((_, _), (_, _)))), \
             ((((_, _), (_, _)), ((_, _), (_, _))), (((_, _), (_, _)), ((_, _), (_, _))))), \
             (((((_, _), (_, _)), ((_, _), (_, _))), (((_, _), (_, _)), ((_, _), (_, _)))), \
             ((((_, _), (_, _)), ((_, _), (_, _))), (((_, _), (_, _)), ((_, _), (_, _))))))"
        );
    }

    #[test]
    fn demangle_thinlto() {
        t_nohash!("_RC3foo.llvm.9D1C9369", "foo");
        t_nohash!("_RC3foo.llvm.9D1C9369@@16", "foo");
        t_nohash!("_RNvC9backtrace3foo.llvm.A5310EB9", "backtrace::foo");
    }

    #[test]
    fn demangle_extra_suffix() {
        // From alexcrichton/rustc-demangle#27:
        t_nohash!(
            "_RNvNtNtNtNtCs92dm3009vxr_4rand4rngs7adapter9reseeding4fork23FORK_HANDLER_REGISTERED.0.0",
            "rand::rngs::adapter::reseeding::fork::FORK_HANDLER_REGISTERED.0.0"
        );
    }

    #[test]
    fn demangling_limits() {
        // Stress tests found via fuzzing.

        for sym in include_str!("v0-large-test-symbols/early-recursion-limit")
            .lines()
            .filter(|line| !line.is_empty() && !line.starts_with('#'))
        {
            assert_eq!(
                super::demangle(sym).map(|_| ()),
                Err(super::ParseError::RecursedTooDeep)
            );
        }

        assert_contains!(
            ::demangle(
                "RIC20tRYIMYNRYFG05_EB5_B_B6_RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRR\
        RRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRRB_E",
            )
            .to_string(),
            "{recursion limit reached}"
        );
    }

    #[test]
    fn recursion_limit_leaks() {
        // NOTE(eddyb) this test checks that both paths and types support the
        // recursion limit correctly, i.e. matching `push_depth` and `pop_depth`,
        // and don't leak "recursion levels" and trip the limit.
        // The test inputs are generated on the fly, using a repeated pattern,
        // as hardcoding the actual strings would be too verbose.
        // Also, `MAX_DEPTH` can be directly used, instead of assuming its value.
        for &(sym_leaf, expected_leaf) in &[("p", "_"), ("Rp", "&_"), ("C1x", "x")] {
            let mut sym = format!("_RIC0p");
            let mut expected = format!("::<_");
            for _ in 0..(super::MAX_DEPTH * 2) {
                sym.push_str(sym_leaf);
                expected.push_str(", ");
                expected.push_str(expected_leaf);
            }
            sym.push('E');
            expected.push('>');

            t_nohash!(&sym, expected);
        }
    }

    #[test]
    fn recursion_limit_backref_free_bypass() {
        // NOTE(eddyb) this test checks that long symbols cannot bypass the
        // recursion limit by not using backrefs, and cause a stack overflow.

        // This value was chosen to be high enough that stack overflows were
        // observed even with `cargo test --release`.
        let depth = 100_000;

        // In order to hide the long mangling from the initial "shallow" parse,
        // it's nested in an identifier (crate name), preceding its use.
        let mut sym = format!("_RIC{}", depth);
        let backref_start = sym.len() - 2;
        for _ in 0..depth {
            sym.push('R');
        }

        // Write a backref to just after the length of the identifier.
        sym.push('B');
        sym.push(char::from_digit((backref_start - 1) as u32, 36).unwrap());
        sym.push('_');

        // Close the `I` at the start.
        sym.push('E');

        assert_contains!(::demangle(&sym).to_string(), "{recursion limit reached}");
    }
}