use std::{borrow::Cow, cell::OnceCell, fmt, io, path::Path, time};

use {
    bstr::ByteVec,
    grep_matcher::{Captures, LineTerminator, Match, Matcher},
    grep_searcher::{
        LineIter, Searcher, SinkContext, SinkContextKind, SinkError, SinkMatch,
    },
};

use crate::{hyperlink::HyperlinkPath, MAX_LOOK_AHEAD};

/// A type for handling replacements while amortizing allocation.
pub(crate) struct Replacer<M: Matcher> {
    space: Option<Space<M>>,
}

struct Space<M: Matcher> {
    /// The place to store capture locations.
    caps: M::Captures,
    /// The place to write a replacement to.
    dst: Vec<u8>,
    /// The place to store match offsets in terms of `dst`.
    matches: Vec<Match>,
}

impl<M: Matcher> fmt::Debug for Replacer<M> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let (dst, matches) = self.replacement().unwrap_or((&[], &[]));
        f.debug_struct("Replacer")
            .field("dst", &dst)
            .field("matches", &matches)
            .finish()
    }
}

impl<M: Matcher> Replacer<M> {
    /// Create a new replacer for use with a particular matcher.
    ///
    /// This constructor does not allocate. Instead, space for dealing with
    /// replacements is allocated lazily only when needed.
    pub(crate) fn new() -> Replacer<M> {
        Replacer { space: None }
    }

    /// Executes a replacement on the given haystack string by replacing all
    /// matches with the given replacement. To access the result of the
    /// replacement, use the `replacement` method.
    ///
    /// This can fail if the underlying matcher reports an error.
    pub(crate) fn replace_all<'a>(
        &'a mut self,
        searcher: &Searcher,
        matcher: &M,
        mut haystack: &[u8],
        range: std::ops::Range<usize>,
        replacement: &[u8],
    ) -> io::Result<()> {
        // See the giant comment in 'find_iter_at_in_context' below for why we
        // do this dance.
        let is_multi_line = searcher.multi_line_with_matcher(&matcher);
        if is_multi_line {
            if haystack[range.end..].len() >= MAX_LOOK_AHEAD {
                haystack = &haystack[..range.end + MAX_LOOK_AHEAD];
            }
        } else {
            // When searching a single line, we should remove the line
            // terminator. Otherwise, it's possible for the regex (via
            // look-around) to observe the line terminator and not match
            // because of it.
            let mut m = Match::new(0, range.end);
            trim_line_terminator(searcher, haystack, &mut m);
            haystack = &haystack[..m.end()];
        }
        {
            let &mut Space { ref mut dst, ref mut caps, ref mut matches } =
                self.allocate(matcher)?;
            dst.clear();
            matches.clear();

            replace_with_captures_in_context(
                matcher,
                haystack,
                range.clone(),
                caps,
                dst,
                |caps, dst| {
                    let start = dst.len();
                    caps.interpolate(
                        |name| matcher.capture_index(name),
                        haystack,
                        replacement,
                        dst,
                    );
                    let end = dst.len();
                    matches.push(Match::new(start, end));
                    true
                },
            )
            .map_err(io::Error::error_message)?;
        }
        Ok(())
    }

    /// Return the result of the prior replacement and the match offsets for
    /// all replacement occurrences within the returned replacement buffer.
    ///
    /// If no replacement has occurred then `None` is returned.
    pub(crate) fn replacement<'a>(
        &'a self,
    ) -> Option<(&'a [u8], &'a [Match])> {
        match self.space {
            None => None,
            Some(ref space) => {
                if space.matches.is_empty() {
                    None
                } else {
                    Some((&space.dst, &space.matches))
                }
            }
        }
    }

    /// Clear space used for performing a replacement.
    ///
    /// Subsequent calls to `replacement` after calling `clear` (but before
    /// executing another replacement) will always return `None`.
    pub(crate) fn clear(&mut self) {
        if let Some(ref mut space) = self.space {
            space.dst.clear();
            space.matches.clear();
        }
    }

    /// Allocate space for replacements when used with the given matcher and
    /// return a mutable reference to that space.
    ///
    /// This can fail if allocating space for capture locations from the given
    /// matcher fails.
    fn allocate(&mut self, matcher: &M) -> io::Result<&mut Space<M>> {
        if self.space.is_none() {
            let caps =
                matcher.new_captures().map_err(io::Error::error_message)?;
            self.space = Some(Space { caps, dst: vec![], matches: vec![] });
        }
        Ok(self.space.as_mut().unwrap())
    }
}

/// A simple layer of abstraction over either a match or a contextual line
/// reported by the searcher.
///
/// In particular, this provides an API that unions the `SinkMatch` and
/// `SinkContext` types while also exposing a list of all individual match
/// locations.
///
/// While this serves as a convenient mechanism to abstract over `SinkMatch`
/// and `SinkContext`, this also provides a way to abstract over replacements.
/// Namely, after a replacement, a `Sunk` value can be constructed using the
/// results of the replacement instead of the bytes reported directly by the
/// searcher.
#[derive(Debug)]
pub(crate) struct Sunk<'a> {
    bytes: &'a [u8],
    absolute_byte_offset: u64,
    line_number: Option<u64>,
    context_kind: Option<&'a SinkContextKind>,
    matches: &'a [Match],
    original_matches: &'a [Match],
}

impl<'a> Sunk<'a> {
    #[inline]
    pub(crate) fn empty() -> Sunk<'static> {
        Sunk {
            bytes: &[],
            absolute_byte_offset: 0,
            line_number: None,
            context_kind: None,
            matches: &[],
            original_matches: &[],
        }
    }

    #[inline]
    pub(crate) fn from_sink_match(
        sunk: &'a SinkMatch<'a>,
        original_matches: &'a [Match],
        replacement: Option<(&'a [u8], &'a [Match])>,
    ) -> Sunk<'a> {
        let (bytes, matches) =
            replacement.unwrap_or_else(|| (sunk.bytes(), original_matches));
        Sunk {
            bytes,
            absolute_byte_offset: sunk.absolute_byte_offset(),
            line_number: sunk.line_number(),
            context_kind: None,
            matches,
            original_matches,
        }
    }

    #[inline]
    pub(crate) fn from_sink_context(
        sunk: &'a SinkContext<'a>,
        original_matches: &'a [Match],
        replacement: Option<(&'a [u8], &'a [Match])>,
    ) -> Sunk<'a> {
        let (bytes, matches) =
            replacement.unwrap_or_else(|| (sunk.bytes(), original_matches));
        Sunk {
            bytes,
            absolute_byte_offset: sunk.absolute_byte_offset(),
            line_number: sunk.line_number(),
            context_kind: Some(sunk.kind()),
            matches,
            original_matches,
        }
    }

    #[inline]
    pub(crate) fn context_kind(&self) -> Option<&'a SinkContextKind> {
        self.context_kind
    }

    #[inline]
    pub(crate) fn bytes(&self) -> &'a [u8] {
        self.bytes
    }

    #[inline]
    pub(crate) fn matches(&self) -> &'a [Match] {
        self.matches
    }

    #[inline]
    pub(crate) fn original_matches(&self) -> &'a [Match] {
        self.original_matches
    }

    #[inline]
    pub(crate) fn lines(&self, line_term: u8) -> LineIter<'a> {
        LineIter::new(line_term, self.bytes())
    }

    #[inline]
    pub(crate) fn absolute_byte_offset(&self) -> u64 {
        self.absolute_byte_offset
    }

    #[inline]
    pub(crate) fn line_number(&self) -> Option<u64> {
        self.line_number
    }
}

/// A simple encapsulation of a file path used by a printer.
///
/// This represents any transforms that we might want to perform on the path,
/// such as converting it to valid UTF-8 and/or replacing its separator with
/// something else. This allows us to amortize work if we are printing the
/// file path for every match.
///
/// In the common case, no transformation is needed, which lets us avoid
/// the allocation. Typically, only Windows requires a transform, since
/// it's fraught to access the raw bytes of a path directly and first need
/// to lossily convert to UTF-8. Windows is also typically where the path
/// separator replacement is used, e.g., in cygwin environments to use `/`
/// instead of `\`.
///
/// Users of this type are expected to construct it from a normal `Path`
/// found in the standard library. It can then be written to any `io::Write`
/// implementation using the `as_bytes` method. This achieves platform
/// portability with a small cost: on Windows, paths that are not valid UTF-16
/// will not roundtrip correctly.
#[derive(Clone, Debug)]
pub(crate) struct PrinterPath<'a> {
    // On Unix, we can re-materialize a `Path` from our `Cow<'a, [u8]>` with
    // zero cost, so there's no point in storing it. At time of writing,
    // OsStr::as_os_str_bytes (and its corresponding constructor) are not
    // stable yet. Those would let us achieve the same end portably. (As long
    // as we keep our UTF-8 requirement on Windows.)
    #[cfg(not(unix))]
    path: &'a Path,
    bytes: Cow<'a, [u8]>,
    hyperlink: OnceCell<Option<HyperlinkPath>>,
}

impl<'a> PrinterPath<'a> {
    /// Create a new path suitable for printing.
    pub(crate) fn new(path: &'a Path) -> PrinterPath<'a> {
        PrinterPath {
            #[cfg(not(unix))]
            path,
            // N.B. This is zero-cost on Unix and requires at least a UTF-8
            // check on Windows. This doesn't allocate on Windows unless the
            // path is invalid UTF-8 (which is exceptionally rare).
            bytes: Vec::from_path_lossy(path),
            hyperlink: OnceCell::new(),
        }
    }

    /// Set the separator on this path.
    ///
    /// When set, `PrinterPath::as_bytes` will return the path provided but
    /// with its separator replaced with the one given.
    pub(crate) fn with_separator(
        mut self,
        sep: Option<u8>,
    ) -> PrinterPath<'a> {
        /// Replace the path separator in this path with the given separator
        /// and do it in place. On Windows, both `/` and `\` are treated as
        /// path separators that are both replaced by `new_sep`. In all other
        /// environments, only `/` is treated as a path separator.
        fn replace_separator(bytes: &[u8], sep: u8) -> Vec<u8> {
            let mut bytes = bytes.to_vec();
            for b in bytes.iter_mut() {
                if *b == b'/' || (cfg!(windows) && *b == b'\\') {
                    *b = sep;
                }
            }
            bytes
        }
        let Some(sep) = sep else { return self };
        self.bytes = Cow::Owned(replace_separator(self.as_bytes(), sep));
        self
    }

    /// Return the raw bytes for this path.
    pub(crate) fn as_bytes(&self) -> &[u8] {
        &self.bytes
    }

    /// Return this path as a hyperlink.
    ///
    /// Note that a hyperlink may not be able to be created from a path.
    /// Namely, computing the hyperlink may require touching the file system
    /// (e.g., for path canonicalization) and that can fail. This failure is
    /// silent but is logged.
    pub(crate) fn as_hyperlink(&self) -> Option<&HyperlinkPath> {
        self.hyperlink
            .get_or_init(|| HyperlinkPath::from_path(self.as_path()))
            .as_ref()
    }

    /// Return this path as an actual `Path` type.
    pub(crate) fn as_path(&self) -> &Path {
        #[cfg(unix)]
        fn imp<'p>(p: &'p PrinterPath<'_>) -> &'p Path {
            use std::{ffi::OsStr, os::unix::ffi::OsStrExt};
            Path::new(OsStr::from_bytes(p.as_bytes()))
        }
        #[cfg(not(unix))]
        fn imp<'p>(p: &'p PrinterPath<'_>) -> &'p Path {
            p.path
        }
        imp(self)
    }
}

/// A type that provides "nicer" Display and Serialize impls for
/// std::time::Duration. The serialization format should actually be compatible
/// with the Deserialize impl for std::time::Duration, since this type only
/// adds new fields.
#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
pub(crate) struct NiceDuration(pub time::Duration);

impl fmt::Display for NiceDuration {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "{:0.6}s", self.fractional_seconds())
    }
}

impl NiceDuration {
    /// Returns the number of seconds in this duration in fraction form.
    /// The number to the left of the decimal point is the number of seconds,
    /// and the number to the right is the number of milliseconds.
    fn fractional_seconds(&self) -> f64 {
        let fractional = (self.0.subsec_nanos() as f64) / 1_000_000_000.0;
        self.0.as_secs() as f64 + fractional
    }
}

#[cfg(feature = "serde")]
impl serde::Serialize for NiceDuration {
    fn serialize<S: serde::Serializer>(
        &self,
        ser: S,
    ) -> Result<S::Ok, S::Error> {
        use serde::ser::SerializeStruct;

        let mut state = ser.serialize_struct("Duration", 3)?;
        state.serialize_field("secs", &self.0.as_secs())?;
        state.serialize_field("nanos", &self.0.subsec_nanos())?;
        state.serialize_field("human", &format!("{}", self))?;
        state.end()
    }
}

/// A simple formatter for converting `u64` values to ASCII byte strings.
///
/// This avoids going through the formatting machinery which seems to
/// substantially slow things down.
///
/// The `itoa` crate does the same thing as this formatter, but is a bit
/// faster. We roll our own which is a bit slower, but gets us enough of a win
/// to be satisfied with and with pure safe code.
#[derive(Debug)]
pub(crate) struct DecimalFormatter {
    buf: [u8; Self::MAX_U64_LEN],
    start: usize,
}

impl DecimalFormatter {
    /// Discovered via `u64::MAX.to_string().len()`.
    const MAX_U64_LEN: usize = 20;

    /// Create a new decimal formatter for the given 64-bit unsigned integer.
    pub(crate) fn new(mut n: u64) -> DecimalFormatter {
        let mut buf = [0; Self::MAX_U64_LEN];
        let mut i = buf.len();
        loop {
            i -= 1;

            let digit = u8::try_from(n % 10).unwrap();
            n /= 10;
            buf[i] = b'0' + digit;
            if n == 0 {
                break;
            }
        }
        DecimalFormatter { buf, start: i }
    }

    /// Return the decimal formatted as an ASCII byte string.
    pub(crate) fn as_bytes(&self) -> &[u8] {
        &self.buf[self.start..]
    }
}

/// Trim prefix ASCII spaces from the given slice and return the corresponding
/// range.
///
/// This stops trimming a prefix as soon as it sees non-whitespace or a line
/// terminator.
pub(crate) fn trim_ascii_prefix(
    line_term: LineTerminator,
    slice: &[u8],
    range: Match,
) -> Match {
    fn is_space(b: u8) -> bool {
        match b {
            b'\t' | b'\n' | b'\x0B' | b'\x0C' | b'\r' | b' ' => true,
            _ => false,
        }
    }

    let count = slice[range]
        .iter()
        .take_while(|&&b| -> bool {
            is_space(b) && !line_term.as_bytes().contains(&b)
        })
        .count();
    range.with_start(range.start() + count)
}

pub(crate) fn find_iter_at_in_context<M, F>(
    searcher: &Searcher,
    matcher: M,
    mut bytes: &[u8],
    range: std::ops::Range<usize>,
    mut matched: F,
) -> io::Result<()>
where
    M: Matcher,
    F: FnMut(Match) -> bool,
{
    // This strange dance is to account for the possibility of look-ahead in
    // the regex. The problem here is that mat.bytes() doesn't include the
    // lines beyond the match boundaries in mulit-line mode, which means that
    // when we try to rediscover the full set of matches here, the regex may no
    // longer match if it required some look-ahead beyond the matching lines.
    //
    // PCRE2 (and the grep-matcher interfaces) has no way of specifying an end
    // bound of the search. So we kludge it and let the regex engine search the
    // rest of the buffer... But to avoid things getting too crazy, we cap the
    // buffer.
    //
    // If it weren't for multi-line mode, then none of this would be needed.
    // Alternatively, if we refactored the grep interfaces to pass along the
    // full set of matches (if available) from the searcher, then that might
    // also help here. But that winds up paying an upfront unavoidable cost for
    // the case where matches don't need to be counted. So then you'd have to
    // introduce a way to pass along matches conditionally, only when needed.
    // Yikes.
    //
    // Maybe the bigger picture thing here is that the searcher should be
    // responsible for finding matches when necessary, and the printer
    // shouldn't be involved in this business in the first place. Sigh. Live
    // and learn. Abstraction boundaries are hard.
    let is_multi_line = searcher.multi_line_with_matcher(&matcher);
    if is_multi_line {
        if bytes[range.end..].len() >= MAX_LOOK_AHEAD {
            bytes = &bytes[..range.end + MAX_LOOK_AHEAD];
        }
    } else {
        // When searching a single line, we should remove the line terminator.
        // Otherwise, it's possible for the regex (via look-around) to observe
        // the line terminator and not match because of it.
        let mut m = Match::new(0, range.end);
        trim_line_terminator(searcher, bytes, &mut m);
        bytes = &bytes[..m.end()];
    }
    matcher
        .find_iter_at(bytes, range.start, |m| {
            if m.start() >= range.end {
                return false;
            }
            matched(m)
        })
        .map_err(io::Error::error_message)
}

/// Given a buf and some bounds, if there is a line terminator at the end of
/// the given bounds in buf, then the bounds are trimmed to remove the line
/// terminator.
pub(crate) fn trim_line_terminator(
    searcher: &Searcher,
    buf: &[u8],
    line: &mut Match,
) {
    let lineterm = searcher.line_terminator();
    if lineterm.is_suffix(&buf[*line]) {
        let mut end = line.end() - 1;
        if lineterm.is_crlf() && end > 0 && buf.get(end - 1) == Some(&b'\r') {
            end -= 1;
        }
        *line = line.with_end(end);
    }
}

/// Like `Matcher::replace_with_captures_at`, but accepts an end bound.
///
/// See also: `find_iter_at_in_context` for why we need this.
fn replace_with_captures_in_context<M, F>(
    matcher: M,
    bytes: &[u8],
    range: std::ops::Range<usize>,
    caps: &mut M::Captures,
    dst: &mut Vec<u8>,
    mut append: F,
) -> Result<(), M::Error>
where
    M: Matcher,
    F: FnMut(&M::Captures, &mut Vec<u8>) -> bool,
{
    let mut last_match = range.start;
    matcher.captures_iter_at(bytes, range.start, caps, |caps| {
        let m = caps.get(0).unwrap();
        if m.start() >= range.end {
            return false;
        }
        dst.extend(&bytes[last_match..m.start()]);
        last_match = m.end();
        append(caps, dst)
    })?;
    let end = std::cmp::min(bytes.len(), range.end);
    dst.extend(&bytes[last_match..end]);
    Ok(())
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn custom_decimal_format() {
        let fmt = |n: u64| {
            let bytes = DecimalFormatter::new(n).as_bytes().to_vec();
            String::from_utf8(bytes).unwrap()
        };
        let std = |n: u64| n.to_string();

        let ints = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 100, 123, u64::MAX];
        for n in ints {
            assert_eq!(std(n), fmt(n));
        }
    }
}