rust/crates/ide/src/syntax_highlighting/inject.rs

//! "Recursive" Syntax highlighting for code in doctests and fixtures.

use std::mem;

use either::Either;
use hir::{HasAttrs, Semantics};
use ide_db::call_info::ActiveParameter;
use syntax::{
    ast::{self, AstNode, AttrsOwner, DocCommentsOwner},
    match_ast, AstToken, NodeOrToken, SyntaxNode, SyntaxToken, TextRange, TextSize,
};

use crate::{Analysis, HlMod, HlRange, HlTag, RootDatabase};

use super::{highlights::Highlights, injector::Injector};

pub(super) fn ra_fixture(
    hl: &mut Highlights,
    sema: &Semantics<RootDatabase>,
    literal: ast::String,
    expanded: SyntaxToken,
) -> Option<()> {
    let active_parameter = ActiveParameter::at_token(&sema, expanded)?;
    if !active_parameter.name.starts_with("ra_fixture") {
        return None;
    }
    let value = literal.value()?;

    if let Some(range) = literal.open_quote_text_range() {
        hl.add(HlRange { range, highlight: HlTag::StringLiteral.into(), binding_hash: None })
    }

    let mut inj = Injector::default();

    let mut text = &*value;
    let mut offset: TextSize = 0.into();

    while !text.is_empty() {
        let marker = "$0";
        let idx = text.find(marker).unwrap_or(text.len());
        let (chunk, next) = text.split_at(idx);
        inj.add(chunk, TextRange::at(offset, TextSize::of(chunk)));

        text = next;
        offset += TextSize::of(chunk);

        if let Some(next) = text.strip_prefix(marker) {
            if let Some(range) = literal.map_range_up(TextRange::at(offset, TextSize::of(marker))) {
                hl.add(HlRange { range, highlight: HlTag::Keyword.into(), binding_hash: None });
            }

            text = next;

            let marker_len = TextSize::of(marker);
            offset += marker_len;
        }
    }

    let (analysis, tmp_file_id) = Analysis::from_single_file(inj.text().to_string());

    for mut hl_range in analysis.highlight(tmp_file_id).unwrap() {
        for range in inj.map_range_up(hl_range.range) {
            if let Some(range) = literal.map_range_up(range) {
                hl_range.range = range;
                hl.add(hl_range);
            }
        }
    }

    if let Some(range) = literal.close_quote_text_range() {
        hl.add(HlRange { range, highlight: HlTag::StringLiteral.into(), binding_hash: None })
    }

    Some(())
}

const RUSTDOC_FENCE: &'static str = "```";
const RUSTDOC_FENCE_TOKENS: &[&'static str] = &[
    "",
    "rust",
    "should_panic",
    "ignore",
    "no_run",
    "compile_fail",
    "edition2015",
    "edition2018",
    "edition2021",
];

// Basically an owned dyn AttrsOwner without extra Boxing
struct AttrsOwnerNode {
    node: SyntaxNode,
}

impl AttrsOwnerNode {
    fn new<N: DocCommentsOwner>(node: N) -> Self {
        AttrsOwnerNode { node: node.syntax().clone() }
    }
}

impl AttrsOwner for AttrsOwnerNode {}
impl AstNode for AttrsOwnerNode {
    fn can_cast(_: syntax::SyntaxKind) -> bool
    where
        Self: Sized,
    {
        false
    }
    fn cast(_: SyntaxNode) -> Option<Self>
    where
        Self: Sized,
    {
        None
    }
    fn syntax(&self) -> &SyntaxNode {
        &self.node
    }
}

fn doc_attributes<'node>(
    sema: &Semantics<RootDatabase>,
    node: &'node SyntaxNode,
) -> Option<(AttrsOwnerNode, hir::Attrs)> {
    match_ast! {
        match node {
            ast::SourceFile(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Fn(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Struct(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Union(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::RecordField(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::TupleField(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Enum(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Variant(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Trait(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Module(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Static(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Const(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::TypeAlias(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::Impl(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            ast::MacroRules(it) => sema.to_def(&it).map(|def| (AttrsOwnerNode::new(it), def.attrs(sema.db))),
            // ast::MacroDef(it) => sema.to_def(&it).map(|def| (Box::new(it) as _, def.attrs(sema.db))),
            // ast::Use(it) => sema.to_def(&it).map(|def| (Box::new(it) as _, def.attrs(sema.db))),
            _ => return None
        }
    }
}

/// Injection of syntax highlighting of doctests.
pub(super) fn doc_comment(hl: &mut Highlights, sema: &Semantics<RootDatabase>, node: &SyntaxNode) {
    let (owner, attributes) = match doc_attributes(sema, node) {
        Some(it) => it,
        None => return,
    };

    if attributes.docs().map_or(true, |docs| !String::from(docs).contains(RUSTDOC_FENCE)) {
        return;
    }
    let attrs_source_map = attributes.source_map(&owner);

    let mut inj = Injector::default();
    inj.add_unmapped("fn doctest() {\n");

    let mut is_codeblock = false;
    let mut is_doctest = false;

    // Replace the original, line-spanning comment ranges by new, only comment-prefix
    // spanning comment ranges.
    let mut new_comments = Vec::new();
    let mut string;
    for attr in attributes.by_key("doc").attrs() {
        let src = attrs_source_map.source_of(&attr);
        let (line, range, prefix) = match &src {
            Either::Left(it) => {
                string = match find_doc_string_in_attr(attr, it) {
                    Some(it) => it,
                    None => continue,
                };
                let text_range = string.syntax().text_range();
                let text_range = TextRange::new(
                    text_range.start() + TextSize::from(1),
                    text_range.end() - TextSize::from(1),
                );
                let text = string.text();
                (&text[1..text.len() - 1], text_range, "")
            }
            Either::Right(comment) => {
                (comment.text(), comment.syntax().text_range(), comment.prefix())
            }
        };

        let mut pos = TextSize::from(prefix.len() as u32);
        let mut range_start = range.start();
        for line in line.split('\n') {
            let line_len = TextSize::from(line.len() as u32);
            let prev_range_start = {
                let next_range_start = range_start + line_len + TextSize::from(1);
                mem::replace(&mut range_start, next_range_start)
            };
            // only first line has the prefix so take it away for future iterations
            let mut pos = mem::take(&mut pos);

            match line.find(RUSTDOC_FENCE) {
                Some(idx) => {
                    is_codeblock = !is_codeblock;
                    // Check whether code is rust by inspecting fence guards
                    let guards = &line[idx + RUSTDOC_FENCE.len()..];
                    let is_rust =
                        guards.split(',').all(|sub| RUSTDOC_FENCE_TOKENS.contains(&sub.trim()));
                    is_doctest = is_codeblock && is_rust;
                    continue;
                }
                None if !is_doctest => continue,
                None => (),
            }

            // whitespace after comment is ignored
            if let Some(ws) = line[pos.into()..].chars().next().filter(|c| c.is_whitespace()) {
                pos += TextSize::of(ws);
            }
            // lines marked with `#` should be ignored in output, we skip the `#` char
            if line[pos.into()..].starts_with('#') {
                pos += TextSize::of('#');
            }

            new_comments.push(TextRange::at(prev_range_start, pos));
            inj.add(&line[pos.into()..], TextRange::new(pos, line_len) + prev_range_start);
            inj.add_unmapped("\n");
        }
    }
    inj.add_unmapped("\n}");

    let (analysis, tmp_file_id) = Analysis::from_single_file(inj.text().to_string());

    for h in analysis.with_db(|db| super::highlight(db, tmp_file_id, None, true)).unwrap() {
        for r in inj.map_range_up(h.range) {
            hl.add(HlRange {
                range: r,
                highlight: h.highlight | HlMod::Injected,
                binding_hash: h.binding_hash,
            });
        }
    }

    for range in new_comments {
        hl.add(HlRange {
            range,
            highlight: HlTag::Comment | HlMod::Documentation,
            binding_hash: None,
        });
    }
}

fn find_doc_string_in_attr(attr: &hir::Attr, it: &ast::Attr) -> Option<ast::String> {
    match it.literal() {
        // #[doc = lit]
        Some(lit) => match lit.kind() {
            ast::LiteralKind::String(it) => Some(it),
            _ => None,
        },
        // #[cfg_attr(..., doc = "", ...)]
        None => {
            // We gotta hunt the string token manually here
            let text = attr.string_value()?;
            // FIXME: We just pick the first string literal that has the same text as the doc attribute
            // This means technically we might highlight the wrong one
            it.syntax()
                .descendants_with_tokens()
                .filter_map(NodeOrToken::into_token)
                .filter_map(ast::String::cast)
                .find(|string| {
                    string.text().get(1..string.text().len() - 1).map_or(false, |it| it == text)
                })
        }
    }
}