doc_chunks/

literal.rs

use crate::errors::*;
use crate::util::{self, sub_chars};
use crate::{Range, Span};

use fancy_regex::Regex;
use lazy_static::lazy_static;
use proc_macro2::LineColumn;

use std::fmt;

/// Determine if a `CommentVariant` is a documentation comment or not.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum CommentVariantCategory {
    /// Comment variant will end up in documentation.
    Doc,
    /// Comment variant is only visible in source code.
    Dev,
    /// It's a common mark file, and we actually don't know.
    CommonMark,
    /// Toml entries and such.
    Unmergable,
}

/// Track what kind of comment the literal is
#[derive(Debug, Clone, Hash, Eq, PartialEq)]
#[non_exhaustive]
pub enum CommentVariant {
    /// `///`
    TripleSlash,
    /// `//!`
    DoubleSlashEM,
    /// `/*!`
    SlashAsteriskEM,
    /// `/**`
    SlashAsteriskAsterisk,
    /// `/*`
    SlashAsterisk,
    /// `#[doc=` with actual prefix like `#[doc=` and the total length of `r###`
    /// etc. including `r` but without `"`
    MacroDocEqStr(String, usize),
    /// `#[doc= foo!(..)]`, content will be ignored, but allows clusters to not
    /// continue.
    MacroDocEqMacro,
    /// Commonmark File
    CommonMark,
    /// Developer line comment
    DoubleSlash,
    /// Developer block comment
    SlashStar,
    /// Unknown Variant
    Unknown,
    /// Toml entry
    TomlEntry,
}

impl Default for CommentVariant {
    fn default() -> Self {
        CommentVariant::Unknown
    }
}

impl CommentVariant {
    /// Obtain the comment variant category.
    pub fn category(&self) -> CommentVariantCategory {
        match self {
            Self::TripleSlash => CommentVariantCategory::Doc,
            Self::DoubleSlashEM => CommentVariantCategory::Doc,
            Self::MacroDocEqStr(_, _) => CommentVariantCategory::Doc,
            Self::MacroDocEqMacro => CommentVariantCategory::Doc,
            Self::SlashAsteriskEM => CommentVariantCategory::Doc,
            Self::SlashAsteriskAsterisk => CommentVariantCategory::Doc,
            Self::CommonMark => CommentVariantCategory::CommonMark,
            Self::TomlEntry => CommentVariantCategory::Unmergable,
            _ => CommentVariantCategory::Dev,
        }
    }
    /// Return the prefix string.
    ///
    /// Does not include whitespaces for `///` and `//!` variants!
    pub fn prefix_string(&self) -> String {
        match self {
            CommentVariant::TripleSlash => "///".into(),
            CommentVariant::DoubleSlashEM => "//!".into(),
            CommentVariant::MacroDocEqMacro => "".into(),
            CommentVariant::MacroDocEqStr(d, p) => {
                let raw = match p {
                    // TODO: make configureable if each line will start with #[doc ="
                    // TODO: but not here!
                    0 => "\"".to_owned(),
                    x => format!("r{}\"", "#".repeat(x.saturating_sub(1))),
                };
                format!(r#"{d}{raw}"#)
            }
            CommentVariant::CommonMark => "".to_string(),
            CommentVariant::DoubleSlash => "//".to_string(),
            CommentVariant::SlashStar => "/*".to_string(),
            CommentVariant::SlashAsterisk => "/*".to_string(),
            CommentVariant::SlashAsteriskEM => "/*!".to_string(),
            CommentVariant::SlashAsteriskAsterisk => "/**".to_string(),
            CommentVariant::TomlEntry => "".to_owned(),
            unhandled => {
                unreachable!("String representation for comment variant {unhandled:?} exists. qed")
            }
        }
    }
    /// Return length (in bytes) of comment prefix for each variant.
    ///
    /// By definition matches the length of `prefix_string`.
    pub fn prefix_len(&self) -> usize {
        match self {
            CommentVariant::TripleSlash | CommentVariant::DoubleSlashEM => 3,
            CommentVariant::MacroDocEqMacro => 0,
            CommentVariant::MacroDocEqStr(d, p) => d.len() + *p + 1,
            CommentVariant::SlashAsterisk => 2,
            CommentVariant::SlashAsteriskEM | CommentVariant::SlashAsteriskAsterisk => 3,
            _ => self.prefix_string().len(),
        }
    }

    /// Return suffix of different comment variants
    pub fn suffix_len(&self) -> usize {
        match self {
            CommentVariant::MacroDocEqStr(_, 0) => 2,
            CommentVariant::MacroDocEqStr(_, p) => p + 1,
            CommentVariant::SlashAsteriskAsterisk
            | CommentVariant::SlashAsteriskEM
            | CommentVariant::SlashAsterisk => 2,
            CommentVariant::MacroDocEqMacro => 0,
            _ => 0,
        }
    }

    /// Return string which will be appended to each line
    pub fn suffix_string(&self) -> String {
        match self {
            CommentVariant::MacroDocEqStr(_, p) if *p == 0 || *p == 1 => r#""]"#.to_string(),
            CommentVariant::MacroDocEqStr(_, p) => {
                r#"""#.to_string() + &"#".repeat(p.saturating_sub(1)) + "]"
            }
            CommentVariant::SlashAsteriskAsterisk
            | CommentVariant::SlashAsteriskEM
            | CommentVariant::SlashAsterisk => "*/".to_string(),
            _ => "".to_string(),
        }
    }
}

/// A literal with meta info where the first and list whitespace may be found.
#[derive(Clone)]
pub struct TrimmedLiteral {
    /// Track what kind of comment the literal is
    variant: CommentVariant,
    /// The span of rendered content, minus pre and post already applied.
    span: Span,
    /// the complete rendered string including post and pre.
    rendered: String,
    /// Literal prefix length.
    pre: usize,
    /// Literal postfix length.
    post: usize,
    /// Length of rendered **minus** `pre` and `post` in UTF-8 characters.
    len_in_chars: usize,
    len_in_bytes: usize,
}

impl std::cmp::PartialEq for TrimmedLiteral {
    fn eq(&self, other: &Self) -> bool {
        if self.rendered != other.rendered {
            return false;
        }
        if self.pre != other.pre {
            return false;
        }
        if self.post != other.post {
            return false;
        }
        if self.len() != other.len() {
            return false;
        }
        if self.span != other.span {
            return false;
        }
        if self.variant != other.variant {
            return false;
        }

        true
    }
}

impl std::cmp::Eq for TrimmedLiteral {}

impl std::hash::Hash for TrimmedLiteral {
    fn hash<H: std::hash::Hasher>(&self, hasher: &mut H) {
        self.variant.hash(hasher);
        self.rendered.hash(hasher);
        self.span.hash(hasher);
        self.pre.hash(hasher);
        self.post.hash(hasher);
        self.len_in_bytes.hash(hasher);
        self.len_in_chars.hash(hasher);
    }
}

/// Adjust the provided span by a number of `pre` and `post` characters.
fn trim_span(content: &str, span: &mut Span, pre: usize, post: usize) {
    span.start.column += pre;
    if span.end.column >= post {
        span.end.column -= post;
    } else {
        // look for the last character in the previous line
        let previous_line_length = content
            .chars()
            .rev()
            // assumes \n, we want to skip the first one from the back
            .skip(post + 1)
            .take_while(|c| *c != '\n')
            .count();
        span.end = LineColumn {
            line: span.end.line - 1,
            column: previous_line_length,
        };
    }
}

/// Detect the comment variant based on the span based str content.
///
/// Became necessary, since the `proc_macro2::Span` does not distinguish between
/// `#[doc=".."]` and `/// ..` comment variants, and for one, and the span can't
/// cover both correctly.
fn detect_comment_variant(
    content: &str,
    rendered: &String,
    mut span: Span,
) -> Result<(CommentVariant, Span, usize, usize)> {
    let prefix_span = Span {
        start: crate::LineColumn {
            line: span.start.line,
            column: 0,
        },
        end: crate::LineColumn {
            line: span.start.line,
            column: span.start.column.saturating_sub(1),
        },
    };
    let prefix = util::load_span_from(content.as_bytes(), prefix_span)?
        .trim_start()
        .to_string();

    let (variant, span, pre, post) = if rendered.starts_with("///") || rendered.starts_with("//!") {
        let pre = 3; // `///`
        let post = 0; // trailing `\n` is already accounted for above

        span.start.column += pre;

        // must always be a single line
        assert_eq!(span.start.line, span.end.line);
        // if the line includes quotes, the rustc converts them internally
        // to `#[doc="content"]`, where - if `content` contains `"` will substitute
        // them as `\"` which will inflate the number columns.
        // Since we can not distinguish between orignally escaped, we simply
        // use the content read from source.

        let variant = if rendered.starts_with("///") {
            CommentVariant::TripleSlash
        } else {
            CommentVariant::DoubleSlashEM
        };

        (variant, span, pre, post)
    } else if rendered.starts_with("/*") && rendered.ends_with("*/") {
        let variant = if rendered.starts_with("/*!") {
            CommentVariant::SlashAsteriskEM
        } else if rendered.starts_with("/**") {
            CommentVariant::SlashAsteriskAsterisk
        } else {
            CommentVariant::SlashAsterisk
        };

        let pre = variant.prefix_len();
        let post = variant.suffix_len();

        #[cfg(debug_assertions)]
        let orig = span;

        trim_span(rendered, &mut span, pre, post);

        #[cfg(debug_assertions)]
        {
            let raw = util::load_span_from(&mut content.as_bytes(), orig)?;
            let adjusted = util::load_span_from(&mut content.as_bytes(), span)?;

            // we know pre and post only consist of single byte characters
            // so `.len()` is way faster here yet correct.
            assert_eq!(adjusted.len() + pre + post, raw.len());
        }

        (variant, span, pre, post)
    } else {
        // pre and post are for the rendered content
        // not necessarily for the span

        //^r(#+?)"(?:.*\s*)+(?=(?:"\1))("\1)$
        lazy_static! {
            static ref BOUNDED_RAW_STR: Regex =
                Regex::new(r##"^(r(#*)")(?:.*\s*)+?(?=(?:"\2))("\2)\s*\]?\s*$"##)
                    .expect("BOUNEDED_RAW_STR regex compiles");
            static ref BOUNDED_STR: Regex = Regex::new(r##"^"(?:.(?!"\\"))*?"*\s*\]?\s*"$"##)
                .expect("BOUNEDED_STR regex compiles");
        };

        let (pre, post) =
            if let Some(captures) = BOUNDED_RAW_STR.captures(rendered.as_str()).ok().flatten() {
                log::trace!("raw str: >{}<", rendered.as_str());
                let pre = if let Some(prefix) = captures.get(1) {
                    log::trace!("raw str pre: >{}<", prefix.as_str());
                    prefix.as_str().len()
                } else {
                    return Err(Error::Span(
                        "Should have a raw str pre match with a capture group".to_string(),
                    ));
                };
                let post = if let Some(suffix) = captures.get(captures.len() - 1) {
                    log::trace!("raw str post: >{}<", suffix.as_str());
                    suffix.as_str().len()
                } else {
                    return Err(Error::Span(
                        "Should have a raw str post match with a capture group".to_string(),
                    ));
                };

                // r####" must match "####
                debug_assert_eq!(pre, post + 1);

                (pre, post)
            } else if let Some(_captures) = BOUNDED_STR.captures(rendered.as_str()).ok().flatten() {
                // r####" must match "####
                let pre = 1;
                let post = 1;
                debug_assert_eq!('"', rendered.as_bytes()[0_usize] as char);
                debug_assert_eq!('"', rendered.as_bytes()[rendered.len() - 1_usize] as char);
                (pre, post)
            } else {
                return Err(Error::Span(format!("Regex should match >{rendered}<")));
            };

        span.start.column += pre;
        span.end.column = span.end.column.saturating_sub(post);

        (
            CommentVariant::MacroDocEqStr(prefix, pre.saturating_sub(1)),
            span,
            pre,
            post,
        )
    };
    Ok((variant, span, pre, post))
}

impl TrimmedLiteral {
    /// Create an empty comment.
    ///
    /// Prime use case is for `#[doc = foo!()]` cases.
    pub(crate) fn new_empty(
        _content: impl AsRef<str>,
        span: Span,
        variant: CommentVariant,
    ) -> Self {
        Self {
            // Track what kind of comment the literal is
            variant,
            span,
            // .
            rendered: String::new(),
            pre: 0,
            post: 0,
            len_in_chars: 0,
            len_in_bytes: 0,
        }
    }

    pub(crate) fn load_from(content: &str, mut span: Span) -> Result<Self> {
        // let rendered = literal.to_string();
        // produces pretty unusable garabage, since it modifies the content of `///`
        // comments which could contain " which will be escaped
        // and therefor cause the `span()` to yield something that does
        // not align with the rendered literal at all and there are too
        // many pitfalls to sanitize all cases, so reading given span
        // from the file again, and then determining its type is way safer.

        // It's unclear why the trailing `]` character is part of the given span, it shout not be part
        // of it, but the span we obtain from literal seems to be wrong, adding one trailing char.

        // Either cut off `]` or `\n` - we don't need either.
        span.end.column = span.end.column.saturating_sub(1);

        // If the line ending has more than one character, we have to account
        // for that. Otherwise cut of the last character of the ending such that
        // we can't properly detect them anymore.
        if crate::util::extract_delimiter(content)
            .unwrap_or("\n")
            .len()
            > 1
        {
            log::trace!(target: "documentation", "Found two character line ending like CRLF");
            span.end.column += 1;
        }

        let rendered = util::load_span_from(content.as_bytes(), span)?;

        // TODO cache the offsets for faster processing and avoiding repeated O(n) ops
        // let byteoffset2char = rendered.char_indices().enumerate().collect::<indexmap::IndexMap<_usize, (_usize, char)>>();
        // let rendered_len = byteoffset2char.len();

        let rendered_len = rendered.chars().count();

        log::trace!("extracted from source: >{rendered}< @ {span:?}");
        let (variant, span, pre, post) = detect_comment_variant(content, &rendered, span)?;

        let len_in_chars = rendered_len.saturating_sub(post + pre);

        if let Some(span_len) = span.one_line_len() {
            if log::log_enabled!(log::Level::Trace) {
                let extracted =
                    sub_chars(rendered.as_str(), pre..rendered_len.saturating_sub(post));
                log::trace!(target: "quirks", "{span:?} {pre}||{post} for \n extracted: >{extracted}<\n rendered:  >{rendered}<");
                assert_eq!(len_in_chars, span_len);
            }
        }

        let len_in_bytes = rendered.len().saturating_sub(post + pre);
        let trimmed_literal = Self {
            variant,
            len_in_chars,
            len_in_bytes,
            rendered,
            span,
            pre,
            post,
        };
        Ok(trimmed_literal)
    }
}

impl TrimmedLiteral {
    /// Creates a new (single line) literal from the variant, the content, the
    /// size of the pre & post and the line/column on which it starts. Fails if
    /// provided with multiline content (i.e. if the content contains a
    /// line-break).
    pub fn from(
        variant: CommentVariant,
        content: &str,
        pre: usize,
        post: usize,
        line: usize,
        column: usize,
    ) -> std::result::Result<TrimmedLiteral, String> {
        let content_chars_len = content.chars().count();
        let mut span = Span {
            start: LineColumn { line, column },
            end: LineColumn {
                line,
                column: column + content_chars_len,
            },
        };

        trim_span(content, &mut span, pre, post + 1);

        Ok(TrimmedLiteral {
            variant,
            span,
            rendered: content.to_string(),
            pre,
            post,
            len_in_chars: content_chars_len - pre - post,
            len_in_bytes: content.len() - pre - post,
        })
    }
}

impl TrimmedLiteral {
    /// Represent the rendered content as `str`.
    ///
    /// Does not contain `pre` and `post` characters.
    pub fn as_str(&self) -> &str {
        &self.rendered.as_str()[self.pre..(self.pre + self.len_in_bytes)]
    }

    /// The prefix characters.
    pub fn prefix(&self) -> &str {
        &self.rendered.as_str()[..self.pre]
    }

    /// The suffix characters.
    pub fn suffix(&self) -> &str {
        &self.rendered.as_str()[(self.pre + self.len_in_bytes)..]
    }

    /// Full representation including `prefix` and `postfix` characters.
    pub fn as_untrimmed_str(&self) -> &str {
        self.rendered.as_str()
    }

    /// Length in characters, excluding `pre` and `post`.
    pub fn len_in_chars(&self) -> usize {
        self.len_in_chars
    }

    /// Length in bytes, excluding `pre` and `post`.
    pub fn len(&self) -> usize {
        self.len_in_bytes
    }

    pub fn is_empty(&self) -> bool {
        self.len_in_bytes == 0
    }

    /// Obtain the number of characters in `pre()`.
    ///
    /// Since all pre characters are ASCII, this is equivalent to the number of
    /// bytes in `pre()`.
    pub fn pre(&self) -> usize {
        self.pre
    }

    /// Obtain the number of characters in `post()`.
    ///
    /// Since all pre characters are ASCII, this is equivalent to the number of
    /// bytes in `post()`.
    pub fn post(&self) -> usize {
        self.post
    }

    /// The span that is covered by this literal.
    ///
    /// Covers only the content, no marker or helper characters.
    pub fn span(&self) -> Span {
        self.span
    }

    /// Access the characters via an iterator.
    pub fn chars(&self) -> impl Iterator<Item = char> + '_ {
        self.as_str().chars()
    }

    /// The string variant type, see [`CommentVariant`](self::CommentVariant)
    /// for details.
    pub fn variant(&self) -> CommentVariant {
        self.variant.clone()
    }

    /// Display helper, mostly used for debug investigations
    #[allow(unused)]
    pub(crate) fn display(&self, highlight: Range) -> TrimmedLiteralDisplay<'_> {
        TrimmedLiteralDisplay::from((self, highlight))
    }
}

impl fmt::Debug for TrimmedLiteral {
    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
        use console::Style;

        let pick = Style::new().on_black().underlined().dim().cyan();
        let cutoff = Style::new().on_black().bold().dim().yellow();

        write!(
            formatter,
            "{}{}{}",
            cutoff.apply_to(&self.prefix()),
            pick.apply_to(&self.as_str()),
            cutoff.apply_to(&self.suffix()),
        )
    }
}

/// A display style wrapper for a trimmed literal.
///
/// Allows better display of coverage results without code duplication.
///
/// Consists of literal reference and a relative range to the start of the
/// literal.
#[derive(Debug, Clone)]
pub struct TrimmedLiteralDisplay<'a>(pub &'a TrimmedLiteral, pub Range);

impl<'a, R> From<(R, Range)> for TrimmedLiteralDisplay<'a>
where
    R: Into<&'a TrimmedLiteral>,
{
    fn from(tuple: (R, Range)) -> Self {
        let tuple0 = tuple.0.into();
        Self(tuple0, tuple.1)
    }
}

impl<'a> From<TrimmedLiteralDisplay<'a>> for (&'a TrimmedLiteral, Range) {
    fn from(val: TrimmedLiteralDisplay<'a>) -> Self {
        (val.0, val.1)
    }
}

impl<'a> fmt::Display for TrimmedLiteralDisplay<'a> {
    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
        use console::Style;

        // part that is hidden by the trimmed literal, but still present in the actual literal
        let cutoff = Style::new().on_black().bold().underlined().yellow();
        // the contextual characters not covered by range `self.1`
        let context = Style::new().on_black().bold().cyan();
        // highlight the mistake
        let highlight = Style::new().on_black().bold().underlined().red().italic();
        // a special style for any errors, to visualize out of bounds access
        let oob = Style::new().blink().bold().on_yellow().red();

        // simplify
        let literal = self.0;
        let start = self.1.start;
        let end = self.1.end;

        assert!(start <= end);

        // content without quote characters
        let data = literal.as_str();

        // colour the preceding quote character
        // and the context preceding the highlight
        let (pre, ctx1) = if start > literal.pre() {
            (
                // ok since that is ascii, so it's single bytes
                cutoff.apply_to(&data[..literal.pre()]).to_string(),
                {
                    let s = sub_chars(data, literal.pre()..start);
                    context.apply_to(s.as_str()).to_string()
                },
            )
        } else if start <= literal.len_in_chars() {
            let s = sub_chars(data, 0..start);
            (cutoff.apply_to(s.as_str()).to_string(), String::new())
        } else {
            (String::new(), "!!!".to_owned())
        };
        // highlight the given range
        let highlight = if end >= literal.len_in_chars() {
            let s = sub_chars(data, start..literal.len_in_chars());
            oob.apply_to(s.as_str()).to_string()
        } else {
            let s = sub_chars(data, start..end);
            highlight.apply_to(s.as_str()).to_string()
        };
        // color trailing context if any as well as the closing quote character
        let post_idx = literal.pre() + literal.len_in_chars();
        let (ctx2, post) = if post_idx > end {
            let s_ctx = sub_chars(data, end..post_idx);
            let s_cutoff = sub_chars(data, post_idx..literal.len_in_chars());
            (
                context.apply_to(s_ctx.as_str()).to_string(),
                cutoff.apply_to(s_cutoff.as_str()).to_string(),
            )
        } else if end < literal.len_in_chars() {
            let s = sub_chars(
                data,
                end..(literal.len_in_chars() + literal.pre() + literal.post()),
            );
            (String::new(), cutoff.apply_to(s.as_str()).to_string())
        } else {
            (String::new(), oob.apply_to("!!!").to_string())
        };

        write!(formatter, "{pre}{ctx1}{highlight}{ctx2}{post}")
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::testcase::annotated_literals_raw;
    use assert_matches::assert_matches;

    #[test]
    fn variant_detect() {
        let content = r###"#[doc=r"foo"]"###.to_owned();
        let rendered = r##"r"foo""##.to_owned();
        assert_matches!(
        detect_comment_variant(content.as_str(), &rendered, Span{
            start: LineColumn {
                line: 1,
                column: 6,
            },
            end: LineColumn {
                line: 1,
                column: 12 + 1,
            },
        }), Ok((CommentVariant::MacroDocEqStr(prefix, n_pounds), _, _, _)) => {
            assert_eq!(n_pounds, 1);
            assert_eq!(prefix, "#[doc=");
        });
    }

    macro_rules! block_comment_test {
        ($name:ident, $content:literal) => {
            #[test]
            fn $name() {
                const CONTENT: &str = $content;
                let mut literals = annotated_literals_raw(CONTENT);
                let literal = literals.next().unwrap();
                assert!(literals.next().is_none());

                let tl = TrimmedLiteral::load_from(CONTENT, Span::from(literal.span())).unwrap();
                assert!(CONTENT.starts_with(tl.prefix()));
                assert!(CONTENT.ends_with(tl.suffix()));
                assert_eq!(
                    CONTENT
                        .chars()
                        .skip(tl.pre())
                        .take(tl.len_in_chars())
                        .collect::<String>(),
                    tl.as_str().to_owned()
                )
            }
        };
    }

    block_comment_test!(trimmed_oneline_doc, "/** dooc */");
    block_comment_test!(trimmed_oneline_mod, "/*! dooc */");

    block_comment_test!(
        trimmed_multi_doc,
        "/**
mood
*/"
    );
    block_comment_test!(
        trimmed_multi_mod,
        "/*!
mood
*/"
    );
}