limu/

errors.rs

//! The errors of a diagram's text, and the edits that map a place in the text a parser reads
//! back to where its author wrote it, as Pele's src/errors.ts has them.

use std::cell::RefCell;
use std::fmt;
use std::rc::Rc;

use crate::detect::{Detected, DiagramType};
use crate::js::chars::{is_word, len16, trim_end};

/// What kind of error an [`Error`] is, as Pele's PeleErrorCode says.
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub enum ErrorCode {
    /// Text of no diagram type known, or of a type not supported.
    UnsupportedDiagram,
    /// Text that does not parse.
    Syntax,
    /// Text that parses but asks for what cannot be, as a shape of no such name.
    Semantic,
    /// Text past a limit, as of its length or of a flowchart's edges.
    Limit,
}

impl ErrorCode {
    /// The code as Pele writes it, as "unsupported-diagram".
    pub fn as_str(self) -> &'static str {
        match self {
            ErrorCode::UnsupportedDiagram => "unsupported-diagram",
            ErrorCode::Syntax => "syntax",
            ErrorCode::Semantic => "semantic",
            ErrorCode::Limit => "limit",
        }
    }
}

/// An error in a diagram's text, as Pele's PeleError is.
#[derive(Clone, Debug, PartialEq)]
#[non_exhaustive]
pub struct Error {
    /// What kind of error it is.
    pub code: ErrorCode,
    /// What is wrong, as Pele words it.
    pub message: String,
    /// The type of the diagram the error is in, where it is known.
    pub diagram: Option<DiagramType>,
    /// That type as Mermaid names it, as "flowchart-v2", where it is known.
    pub mermaid_type: Option<&'static str>,
    /// The line of the text the error is on, from 1, or 0 where it has no place.
    pub line: usize,
    /// The column of the line, from 1, counted in UTF-16 code units as JavaScript counts
    /// them, or 0 where the error has no place.
    pub column: usize,
    /// The line, and a caret under the column, or nothing where the error has no place.
    pub snippet: String,
}

impl Error {
    /// An error with no place.
    pub(crate) fn new(message: impl Into<String>, code: ErrorCode) -> Error {
        Error {
            code,
            message: message.into(),
            diagram: None,
            mermaid_type: None,
            line: 0,
            column: 0,
            snippet: String::new(),
        }
    }

    /// The error, of a diagram of a type found.
    pub(crate) fn of(mut self, found: Detected) -> Error {
        self.diagram = Some(found.diagram);
        self.mermaid_type = Some(found.mermaid_type);
        self
    }

    /// Gives the error the diagram's type where it has none, and writes the entity codes in
    /// its message and snippet as the text has them, not as the parser read them, as Pele's
    /// tagErrors does for the errors of a diagram's parser and of its layout.
    pub(crate) fn tagged(mut self, found: Detected) -> Error {
        self.diagram.get_or_insert(found.diagram);
        self.mermaid_type.get_or_insert(found.mermaid_type);
        self.message = written_codes(&self.message);
        self.snippet = written_codes(&self.snippet);
        self
    }
}

impl Error {
    /// The byte of src the error is at, where src is the text it was found in, or none where
    /// the error has no place, or src has no such line and column, or the column falls
    /// between the two code units of a character. It turns the column, which is counted in
    /// UTF-16 code units, into a byte for a host that slices src.
    pub fn byte_offset(&self, src: &str) -> Option<usize> {
        if self.line == 0 || self.column == 0 {
            return None;
        }
        let mut start = 0;
        for _ in 1..self.line {
            start += src[start..].find('\n')? + 1;
        }
        let line = &src[start..];
        let line = &line[..line.find('\n').unwrap_or(line.len())];
        let mut units = self.column - 1;
        for (i, c) in line.char_indices() {
            if units < c.len_utf16() {
                return (units == 0).then_some(start + i);
            }
            units -= c.len_utf16();
        }
        (units == 0).then_some(start + line.len())
    }
}

impl fmt::Display for Error {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.write_str(&self.message)
    }
}

impl std::error::Error for Error {}

/// Pele's Edit: the bytes from..to of a text replaced by text. Pele records the length of
/// the replacement in code units. An edit keeps the replacement itself, so that a place
/// inside it is counted in code units as Pele counts it.
#[derive(Clone, Debug)]
pub(crate) struct Edit {
    pub(crate) from: usize,
    pub(crate) to: usize,
    pub(crate) text: String,
}

impl Edit {
    /// An edit that takes the bytes from..to out.
    pub(crate) fn cut(from: usize, to: usize) -> Edit {
        Edit {
            from,
            to,
            text: String::new(),
        }
    }
}

/// The edits of the passes over a text, each pass's in the order they were made, or none
/// where nothing is recorded.
pub(crate) type Steps<'a> = Option<&'a mut Vec<Vec<Edit>>>;

/// Adds the edits of a pass to steps, if they are recorded. A pass adds its edits once it is
/// done, where Pele's recording adds them as it begins. No pass adds while another is under
/// way, so the steps are in the same order.
pub(crate) fn push(steps: &mut Steps, edits: Vec<Edit>) {
    if let Some(steps) = steps {
        steps.push(edits);
    }
}

/// Maps byte at of the text edits made back to the text they were made in, as Pele's
/// stepInput does. A place inside a replacement maps to as many code units into what it
/// replaced, or to the last of those.
fn step_input(edits: &[Edit], at: usize) -> usize {
    let mut shift: isize = 0;
    for e in edits {
        let start = (e.from as isize - shift) as usize;
        if at < start {
            break;
        }
        if at < start + e.text.len() {
            return e.from + len16(&e.text[..at - start]).min(e.to - e.from - 1);
        }
        shift += e.to as isize - e.from as isize - e.text.len() as isize;
    }
    (at as isize + shift) as usize
}

/// Maps byte at of the text the last of steps made back to the text the first was made in,
/// as Pele's inputOffset does.
pub(crate) fn input_offset(steps: &[Vec<Edit>], at: usize) -> usize {
    steps
        .iter()
        .rev()
        .fold(at, |at, edits| step_input(edits, at))
}

/// Writes each entity code that Pele's encodeEntities hid behind placeholders, as `fl°°35¶ß`
/// or `fl°quot¶ß`, back as its author wrote it, `#35;` or `#quot;`, as Pele's writtenCodes
/// does with `text.replace(/fl°°?(\w+)¶ß/g, '#$1;')`.
pub(crate) fn written_codes(text: &str) -> String {
    const OPEN: &str = "fl°";
    const CLOSE: &str = "¶ß";
    if !text.contains(CLOSE) {
        return text.to_string();
    }
    let mut out = String::new();
    let (mut last, mut i) = (0, 0);
    while let Some(n) = text[i..].find(OPEN) {
        let at = i + n;
        let mut j = at + OPEN.len();
        if text[j..].starts_with('°') {
            j += '°'.len_utf8();
        }
        let k = j + text[j..]
            .bytes()
            .take_while(|&c| is_word(i32::from(c)))
            .count();
        if k > j && text[k..].starts_with(CLOSE) {
            out.push_str(&text[last..at]);
            out.push('#');
            out.push_str(&text[j..k]);
            out.push(';');
            last = k + CLOSE.len();
            i = last;
            continue;
        }
        // With no word after `fl°°`, the expression backtracks to take the second `°` as no
        // part of it, and no word starts there either.
        i = at + OPEN.len();
    }
    out.push_str(&text[last..]);
    out
}

/// A text a diagram's parser is reading, and what finds a byte of it in the text its author
/// wrote, as Pele's Origin has them.
pub(crate) struct Origin {
    pub(crate) parsed: Rc<str>,
    pub(crate) find: Box<dyn Fn(usize) -> (String, usize)>,
}

thread_local! {
    /// The origin of the text being parsed on this thread. Pele keeps one origin in a variable
    /// while its parser runs, and a thread runs one parser at a time.
    static ORIGIN: RefCell<Option<Rc<Origin>>> = const { RefCell::new(None) };
}

/// Runs run with the errors it locates in next's text reporting where they are in the text
/// its author wrote, as Pele's locatingIn does.
pub(crate) fn locating_in<T>(next: Origin, run: impl FnOnce() -> T) -> T {
    /// Puts the outer origin back, however run ends.
    struct Restore(Option<Rc<Origin>>);
    impl Drop for Restore {
        fn drop(&mut self) {
            let outer = self.0.take();
            ORIGIN.with(|o| *o.borrow_mut() = outer);
        }
    }
    let outer = ORIGIN.with(|o| o.borrow_mut().replace(Rc::new(next)));
    let _restore = Restore(outer);
    run()
}

/// Finds byte offset of src in the text its author wrote, as Pele's written does: where src
/// is the text being parsed, the text as written and the byte in it, and otherwise src and
/// offset as they are.
pub(crate) fn written(src: &str, offset: usize) -> (String, usize) {
    match ORIGIN.with(|o| o.borrow().clone()) {
        Some(origin) if *origin.parsed == *src => (origin.find)(offset),
        _ => (src.to_string(), offset),
    }
}

/// A place in a text: its line and column, counted from 1, the column in UTF-16 code units,
/// and its line with a caret under the column.
#[derive(Clone, Debug, PartialEq, Eq)]
pub(crate) struct Location {
    pub(crate) line: usize,
    pub(crate) column: usize,
    pub(crate) snippet: String,
}

/// The place of byte offset in src, as Pele's locate finds that of a code unit. An offset at
/// or past the end of what src holds but white space is put after its last character, and a
/// place in the text being parsed is found in the text its author wrote.
pub(crate) fn locate(src: &str, offset: usize) -> Location {
    // Keep errors at the end on the last line that holds anything.
    let end = trim_end(src).len();
    let (src, mut offset) = if end > 0 && offset >= end {
        // Pele steps back one code unit, to the last character or the low half of it, and on
        // past it. limu steps back to the start of the character.
        let last = src[..end].chars().next_back().map_or(1, char::len_utf8);
        let (src, at) = written(src, end - last);
        let size = src[at.min(src.len())..]
            .chars()
            .next()
            .map_or(0, char::len_utf8);
        (src, at + size)
    } else {
        written(src, offset)
    };
    offset = offset.min(src.len());
    while !src.is_char_boundary(offset) {
        offset -= 1;
    }
    let line_start = src[..offset].rfind('\n').map_or(0, |i| i + 1);
    let line = 1 + src[..offset].matches('\n').count();
    let line_end = src[offset..].find('\n').map_or(src.len(), |i| offset + i);
    let column = len16(&src[line_start..offset]) + 1;
    let snippet = format!(
        "{}\n{}^",
        &src[line_start..line_end],
        "-".repeat(column - 1)
    );
    Location {
        line,
        column,
        snippet,
    }
}

/// The error of text that does not parse at byte offset, worded as Jison's parsers word it,
/// as Pele's syntaxError makes it. A lexical error is of text no token starts with.
pub(crate) fn syntax_error(
    diagram: DiagramType,
    src: &str,
    offset: usize,
    detail: &str,
    lexical: bool,
) -> Error {
    let loc = locate(src, offset);
    let message = if lexical {
        format!(
            "Lexical error on line {}. Unrecognized text.\n{}",
            loc.line, loc.snippet
        )
    } else {
        format!(
            "Parse error on line {}:\n{}\n{detail}",
            loc.line, loc.snippet
        )
    };
    Error {
        code: ErrorCode::Syntax,
        message,
        diagram: Some(diagram),
        mermaid_type: None,
        line: loc.line,
        column: loc.column,
        snippet: loc.snippet,
    }
}

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

    #[test]
    fn finds_the_byte_of_a_place() {
        let src = "graph TD\n  a😀b --> 中\n\nx\n";
        for (at, _) in src.char_indices().chain([(src.len(), ' ')]) {
            let loc = locate_in(src, at);
            let e = Error {
                line: loc.line,
                column: loc.column,
                ..Error::new("", ErrorCode::Syntax)
            };
            assert_eq!(e.byte_offset(src), Some(at), "{at}");
        }
        let at = |line, column| {
            Error {
                line,
                column,
                ..Error::new("", ErrorCode::Syntax)
            }
            .byte_offset(src)
        };
        // After the emoji, inside it, at a line's end and past it, on no line, and with no
        // place.
        assert_eq!(at(2, 6), Some(16));
        assert_eq!(at(2, 5), None);
        assert_eq!(at(1, 9), Some(8));
        assert_eq!(at(1, 10), None);
        assert_eq!(at(6, 1), None);
        assert_eq!(at(0, 0), None);
    }

    /// The place of a byte, found as locate finds it but without its rules for the end of
    /// the text.
    fn locate_in(src: &str, offset: usize) -> Location {
        let line_start = src[..offset].rfind('\n').map_or(0, |i| i + 1);
        Location {
            line: 1 + src[..offset].matches('\n').count(),
            column: len16(&src[line_start..offset]) + 1,
            snippet: String::new(),
        }
    }
}