turbopack_core/issue/

mod.rs

pub mod analyze;
pub mod code_gen;
pub mod module;
pub mod resolve;

use std::{
    cmp::min,
    fmt::{Display, Formatter},
};

use anyhow::{Result, bail};
use async_trait::async_trait;
use auto_hash_map::AutoSet;
use bincode::{Decode, Encode};
use serde::{Deserialize, Serialize};
use turbo_esregex::EsRegex;
use turbo_rcstr::{RcStr, rcstr};
use turbo_tasks::{
    CollectiblesSource, NonLocalValue, OperationVc, RawVc, ReadRef, ResolvedVc, TryFlatJoinIterExt,
    TryJoinIterExt, Upcast, ValueDefault, ValueToString, ValueToStringRef, Vc, emit,
    trace::TraceRawVcs,
};
use turbo_tasks_fs::{
    FileContent, FileLine, FileLinesContent, FileSystem, FileSystemPath, glob::Glob,
    json::UnparsableJson,
};
use turbo_tasks_hash::{DeterministicHash, Xxh3Hash64Hasher};

use crate::{
    asset::{Asset, AssetContent},
    condition::ContextCondition,
    generated_code_source::GeneratedCodeSource,
    ident::{AssetIdent, Layer},
    source::Source,
    source_map::{GenerateSourceMap, SourceMap, TokenWithSource},
    source_pos::SourcePos,
};

#[turbo_tasks::value(shared, task_input)]
#[derive(PartialOrd, Ord, Copy, Clone, Hash, Debug, DeterministicHash, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
pub enum IssueSeverity {
    Bug,
    Fatal,
    Error,
    Warning,
    Hint,
    Note,
    Suggestion,
    Info,
}

impl IssueSeverity {
    pub fn as_str(&self) -> &'static str {
        match self {
            IssueSeverity::Bug => "bug",
            IssueSeverity::Fatal => "fatal",
            IssueSeverity::Error => "error",
            IssueSeverity::Warning => "warning",
            IssueSeverity::Hint => "hint",
            IssueSeverity::Note => "note",
            IssueSeverity::Suggestion => "suggestion",
            IssueSeverity::Info => "info",
        }
    }

    pub fn as_help_str(&self) -> &'static str {
        match self {
            IssueSeverity::Bug => "bug in implementation",
            IssueSeverity::Fatal => "unrecoverable problem",
            IssueSeverity::Error => "problem that cause a broken result",
            IssueSeverity::Warning => "problem should be addressed in short term",
            IssueSeverity::Hint => "idea for improvement",
            IssueSeverity::Note => "detail that is worth mentioning",
            IssueSeverity::Suggestion => "change proposal for improvement",
            IssueSeverity::Info => "detail that is worth telling",
        }
    }
}

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

/// Represents a section of structured styled text. This can be interpreted and
/// rendered by various UIs as appropriate, e.g. HTML for display on the web,
/// ANSI sequences in TTYs.
#[derive(Clone, Debug, PartialOrd, Ord, DeterministicHash, Serialize)]
#[turbo_tasks::value(shared)]
pub enum StyledString {
    /// Multiple [StyledString]s concatenated into a single line. Each item is
    /// considered as inline element. Items might contain line breaks, which
    /// would be considered as soft line breaks.
    Line(Vec<StyledString>),
    /// Multiple [StyledString]s stacked vertically. They are considered as
    /// block elements, just like the top level [StyledString].
    Stack(Vec<StyledString>),
    /// Some prose text.
    Text(RcStr),
    /// Code snippet.
    // TODO add language to support syntax highlighting
    Code(RcStr),
    /// Some important text.
    Strong(RcStr),
}

impl StyledString {
    pub fn to_unstyled_string(&self) -> String {
        match self {
            StyledString::Line(items) => items
                .iter()
                .map(|item| item.to_unstyled_string())
                .collect::<Vec<_>>()
                .join(""),
            StyledString::Stack(items) => items
                .iter()
                .map(|item| item.to_unstyled_string())
                .collect::<Vec<_>>()
                .join("\n"),
            StyledString::Text(s) | StyledString::Code(s) | StyledString::Strong(s) => {
                s.to_string()
            }
        }
    }
}

#[async_trait]
#[turbo_tasks::value_trait]
pub trait Issue {
    /// Severity allows the user to filter out unimportant issues, with Bug
    /// being the highest priority and Info being the lowest.
    fn severity(&self) -> IssueSeverity {
        IssueSeverity::Error
    }

    /// The file path that generated the issue, displayed to the user as message
    /// header.
    async fn file_path(&self) -> Result<FileSystemPath>;

    /// The stage of the compilation process at which the issue occurred. This
    /// is used to sort issues.
    fn stage(&self) -> IssueStage;

    /// The issue title should be descriptive of the issue, but should be a
    /// single line. This is displayed to the user directly under the issue
    /// header.
    async fn title(&self) -> Result<StyledString>;

    /// A more verbose message of the issue, appropriate for providing multiline
    /// information of the issue.
    async fn description(&self) -> Result<Option<StyledString>> {
        Ok(None)
    }

    /// Full details of the issue, appropriate for providing debug level
    /// information. Only displayed if the user explicitly asks for detailed
    /// messages (not to be confused with severity).
    async fn detail(&self) -> Result<Option<StyledString>> {
        Ok(None)
    }

    /// A link to relevant documentation of the issue. Only displayed in console
    /// if the user explicitly asks for detailed messages.
    fn documentation_link(&self) -> RcStr {
        rcstr!("")
    }

    /// The source location that caused the issue. Eg, for a parsing error it
    /// should point at the offending character. Displayed to the user alongside
    /// the title/description.
    fn source(&self) -> Option<IssueSource> {
        None
    }

    /// Additional source locations related to this issue (e.g., generated code
    /// from a loader). Each source includes a description and location.
    /// These are displayed alongside the primary source to give users full
    /// context about the error.
    async fn additional_sources(&self) -> Result<Vec<AdditionalIssueSource>> {
        Ok(vec![])
    }
}

// A collectible trait that allows traces to be computed for a given module.
#[turbo_tasks::value_trait]
pub trait ImportTracer {
    #[turbo_tasks::function]
    fn get_traces(self: Vc<Self>, path: FileSystemPath) -> Vc<ImportTraces>;
}

#[turbo_tasks::value]
#[derive(Debug)]
pub struct DelegatingImportTracer {
    delegates: AutoSet<ResolvedVc<Box<dyn ImportTracer>>>,
}

impl DelegatingImportTracer {
    async fn get_traces(&self, path: FileSystemPath) -> Result<Vec<ImportTrace>> {
        Ok(self
            .delegates
            .iter()
            .map(|d| d.get_traces(path.clone()))
            .try_join()
            .await?
            .iter()
            .flat_map(|v| v.0.iter().cloned())
            .collect())
    }
}

pub type ImportTrace = Vec<ReadRef<AssetIdent>>;

#[turbo_tasks::value(shared)]
pub struct ImportTraces(pub Vec<ImportTrace>);

#[turbo_tasks::value_impl]
impl ValueDefault for ImportTraces {
    #[turbo_tasks::function]
    fn value_default() -> Vc<Self> {
        Self::cell(ImportTraces(vec![]))
    }
}

pub trait IssueExt {
    fn emit(self);
}

impl<T> IssueExt for ResolvedVc<T>
where
    T: Upcast<Box<dyn Issue>>,
{
    fn emit(self) {
        emit(ResolvedVc::upcast_non_strict::<Box<dyn Issue>>(self));
    }
}

#[turbo_tasks::value(transparent)]
pub struct Issues(Vec<ResolvedVc<Box<dyn Issue>>>);

/// A pattern that can match by exact string, glob, or regex.
#[derive(Clone, Debug, PartialEq, Eq, TraceRawVcs, NonLocalValue, Encode, Decode)]
pub enum IgnoreIssuePattern {
    /// The value must exactly equal the pattern string.
    ExactString(RcStr),
    /// The pattern is treated as a glob (uses turbo-tasks-fs glob matching).
    Glob(Glob),
    /// The pattern is a regular expression (supports ES-style patterns via `EsRegex`).
    Regex(EsRegex),
}

impl IgnoreIssuePattern {
    /// Test whether the pattern matches the given value.
    pub fn matches(&self, value: &str) -> bool {
        match self {
            IgnoreIssuePattern::ExactString(s) => value == s.as_str(),
            IgnoreIssuePattern::Glob(glob) => glob.matches(value),
            IgnoreIssuePattern::Regex(regex) => regex.is_match(value),
        }
    }
}

/// A rule describing an issue to ignore. `path` is mandatory;
/// `title` and `description` are optional additional filters.
#[derive(Clone, Debug, PartialEq, Eq, TraceRawVcs, NonLocalValue, Encode, Decode)]
pub struct IgnoreIssue {
    /// File-path pattern (mandatory).
    pub path: IgnoreIssuePattern,
    /// Title pattern (optional).
    pub title: Option<IgnoreIssuePattern>,
    /// Description pattern (optional).
    pub description: Option<IgnoreIssuePattern>,
}

#[turbo_tasks::value(shared)]
pub struct IssueFilter {
    /// The minimum severity for issues
    severity: IssueSeverity,
    /// The minimum severity for issues in node_modules
    foreign_severity: IssueSeverity,
    /// Issues matching any of these rules are ignored (dropped from results).
    ignore_rules: Box<[IgnoreIssue]>,
}

impl IssueFilter {
    /// A filter that lets everything through.
    pub fn everything() -> Self {
        IssueFilter {
            severity: IssueSeverity::Info,
            foreign_severity: IssueSeverity::Info,
            ignore_rules: Box::from([]),
        }
    }

    /// Construct a filter with the standard warning/foreign-error severities.
    pub fn warnings_and_foreign_errors() -> Self {
        IssueFilter {
            severity: IssueSeverity::Warning,
            foreign_severity: IssueSeverity::Error,
            ignore_rules: Box::from([]),
        }
    }

    /// Set the ignore rules for this filter.
    pub fn with_ignore_rules(mut self, rules: Box<[IgnoreIssue]>) -> Self {
        self.ignore_rules = rules;
        self
    }

    /// Returns true if the issue is allowed by this filter.
    pub async fn matches(&self, issue: ResolvedVc<Box<dyn Issue>>) -> Result<bool> {
        Ok(self.matches_all_fast_path()
            || self
                .matches_ref_slow_path(&*issue.into_trait_ref().await?)
                .await?)
    }

    pub async fn matches_ref(&self, issue: &dyn Issue) -> Result<bool> {
        Ok(self.matches_all_fast_path() || self.matches_ref_slow_path(issue).await?)
    }

    fn matches_all_fast_path(&self) -> bool {
        self.severity == IssueSeverity::Info
            && self.foreign_severity == IssueSeverity::Info
            && self.ignore_rules.is_empty()
    }

    async fn matches_ref_slow_path(&self, issue: &dyn Issue) -> Result<bool> {
        // Fetch the file path once — it's used by both severity and ignore-rule
        // checks.
        let file_path = issue.file_path().await?;

        // Check severity first — this is cheap and avoids fetching
        // title/description for issues that would be filtered out anyway.
        let severity = issue.severity();
        // NOTE: Lower severities are _more_ severe
        let severity_allowed = if severity <= self.severity || severity <= self.foreign_severity {
            // we need to check the path to see if it is foreign or not.  Only await the
            // path if it might possibly matter
            if severity <= self.severity && severity <= self.foreign_severity {
                // it matches no matter where the path is
                true
            } else if ContextCondition::InNodeModules.matches(&file_path) {
                severity <= self.foreign_severity
            } else {
                severity <= self.severity
            }
        } else {
            // it is too low severity to match either way
            false
        };

        if !severity_allowed {
            return Ok(false);
        }

        // Check ignore rules — if any rule matches, the issue is dropped.
        // Title and description are fetched lazily: only when a rule's path
        // matches and the rule also specifies a title/description pattern.
        if !self.ignore_rules.is_empty() {
            let file_path_str = file_path.to_string();
            let mut title_str: Option<String> = None;
            let mut description_text: Option<Option<String>> = None;

            for rule in &self.ignore_rules {
                if !rule.path.matches(&file_path_str) {
                    continue;
                }
                if let Some(ref title_pat) = rule.title {
                    if title_str.is_none() {
                        title_str = Some(issue.title().await?.to_unstyled_string());
                    }
                    if !title_pat.matches(title_str.as_deref().unwrap()) {
                        continue;
                    }
                }
                if let Some(ref desc_pat) = rule.description {
                    if description_text.is_none() {
                        description_text =
                            Some(issue.description().await?.map(|s| s.to_unstyled_string()));
                    }
                    match description_text.as_ref().unwrap().as_deref() {
                        Some(desc) if desc_pat.matches(desc) => {}
                        _ => continue,
                    }
                }
                // All specified fields matched — ignore this issue.
                return Ok(false);
            }
        }

        Ok(true)
    }
}

/// A list of issues captured with [`CollectibleIssuesExt::peek_issues`].
#[turbo_tasks::value(shared)]
#[derive(Debug)]
pub struct CapturedIssues {
    issues: AutoSet<ResolvedVc<Box<dyn Issue>>>,
    tracer: ResolvedVc<DelegatingImportTracer>,
}

impl CapturedIssues {
    /// Returns an iterator over the issues.
    pub fn iter(&self) -> impl Iterator<Item = ResolvedVc<Box<dyn Issue>>> + '_ {
        self.issues.iter().copied()
    }

    // Returns all the issues as formatted `PlainIssues`.
    pub async fn get_plain_issues(&self, filter: &IssueFilter) -> Result<Vec<ReadRef<PlainIssue>>> {
        let mut list = self
            .issues
            .iter()
            .map(async |issue| {
                if filter.matches(*issue).await? {
                    Ok(Some(
                        PlainIssue::from_issue(**issue, Some(*self.tracer)).await?,
                    ))
                } else {
                    Ok(None)
                }
            })
            .try_flat_join()
            .await?;
        list.sort();
        Ok(list)
    }
}

#[turbo_tasks::task_input]
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, TraceRawVcs, Encode, Decode)]
pub struct IssueSource {
    source: ResolvedVc<Box<dyn Source>>,
    range: Option<SourceRange>,
}

/// The end position is the first character after the range
#[turbo_tasks::task_input]
#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash, TraceRawVcs, Encode, Decode)]
enum SourceRange {
    LineColumn(SourcePos, SourcePos),
    ByteOffset(u32, u32),
}

impl IssueSource {
    // Sometimes we only have the source file that causes an issue, not the
    // exact location, such as as in some generated code.
    pub fn from_source_only(source: ResolvedVc<Box<dyn Source>>) -> Self {
        IssueSource {
            source,
            range: None,
        }
    }

    /// Drops the precise range while preserving the source file.
    pub fn without_range(self) -> Self {
        IssueSource {
            range: None,
            ..self
        }
    }

    pub fn from_line_col(
        source: ResolvedVc<Box<dyn Source>>,
        start: SourcePos,
        end: SourcePos,
    ) -> Self {
        IssueSource {
            source,
            range: Some(SourceRange::LineColumn(start, end)),
        }
    }

    pub fn from_single_line_col(source: ResolvedVc<Box<dyn Source>>, pos: SourcePos) -> Self {
        IssueSource {
            source,
            range: Some(SourceRange::LineColumn(
                pos,
                SourcePos {
                    line: pos.line,
                    // The end position is the first character after the range
                    column: pos.column + 1,
                },
            )),
        }
    }

    async fn into_plain(self) -> Result<PlainIssueSource> {
        let Self { mut source, range } = self;

        let range = if let Some(range) = range {
            let mut range = match range {
                SourceRange::LineColumn(start, end) => Some((start, end)),
                SourceRange::ByteOffset(start, end) => {
                    // Defensively read the content, an error there should not prevent all issue
                    // formatting.  Best practice is for `content` to return `NotFound` instead of
                    // an error.
                    if let Ok(content) = self.source.content().lines().await
                        && let FileLinesContent::Lines(lines) = &*content
                    {
                        let start = find_line_and_column(lines.as_ref(), start);
                        let end = find_line_and_column(lines.as_ref(), end);
                        Some((start, end))
                    } else {
                        None
                    }
                }
            };

            // If we have a source map, map the line/column to the original source.
            if let Some((start, end)) = range {
                let mapped = source_pos(source, start, end).await?;

                if let Some((mapped_source, start, end)) = mapped {
                    range = Some((start, end));
                    source = mapped_source;
                }
            }
            range
        } else {
            None
        };
        Ok(PlainIssueSource {
            asset: PlainSource::from_source(*source).await?,
            range,
        })
    }

    /// Create an [`IssueSource`] from an [`UnparsableJson`] error, using its
    /// start/end location if available.
    pub fn from_unparsable_json(
        source: ResolvedVc<Box<dyn Source>>,
        error: &UnparsableJson,
    ) -> Self {
        match (error.start_location, error.end_location) {
            (None, None) => Self::from_source_only(source),
            (Some((line, column)), None) | (None, Some((line, column))) => Self::from_line_col(
                source,
                SourcePos { line, column },
                SourcePos { line, column },
            ),
            (Some((start_line, start_column)), Some((end_line, end_column))) => {
                Self::from_line_col(
                    source,
                    SourcePos {
                        line: start_line,
                        column: start_column,
                    },
                    SourcePos {
                        line: end_line,
                        column: end_column,
                    },
                )
            }
        }
    }

    /// Create a [`IssueSource`] from byte offsets given by an swc ast node
    /// span.
    ///
    /// Arguments:
    ///
    /// * `source`: The source code in which to look up the byte offsets.
    /// * `start`: The start index of the span. Must use **1-based** indexing.
    /// * `end`: The end index of the span. Must use **1-based** indexing.
    pub fn from_swc_offsets(source: ResolvedVc<Box<dyn Source>>, start: u32, end: u32) -> Self {
        IssueSource {
            source,
            range: match (start == 0, end == 0) {
                (true, true) => None,
                (false, false) => Some(SourceRange::ByteOffset(start - 1, end - 1)),
                (false, true) => Some(SourceRange::ByteOffset(start - 1, start - 1)),
                (true, false) => Some(SourceRange::ByteOffset(end - 1, end - 1)),
            },
        }
    }

    /// Returns an `IssueSource` representing a span of code in the `source`.
    /// Positions are derived from byte offsets and stored as lines and columns.
    /// Requires a binary search of the source text to perform this.
    ///
    /// Arguments:
    ///
    /// * `source`: The source code in which to look up the byte offsets.
    /// * `start`: Byte offset into the source that the text begins. 0-based index and inclusive.
    /// * `end`: Byte offset into the source that the text ends. 0-based index and exclusive.
    pub async fn from_byte_offset(
        source: ResolvedVc<Box<dyn Source>>,
        start: u32,
        end: u32,
    ) -> Result<Self> {
        Ok(IssueSource {
            source,
            range: if let FileLinesContent::Lines(lines) = &*source.content().lines().await? {
                let start = find_line_and_column(lines.as_ref(), start);
                let end = find_line_and_column(lines.as_ref(), end);
                Some(SourceRange::LineColumn(start, end))
            } else {
                None
            },
        })
    }

    /// Returns the file path for the source file.
    pub async fn file_path(&self) -> Result<FileSystemPath> {
        Ok(self.source.ident().await?.path.clone())
    }

    /// If this source implements `GenerateSourceMap`, returns an
    /// `AdditionalIssueSource` that wraps the source in a `GeneratedCodeSource`
    /// (stripping source-map support) so the generated code is shown alongside
    /// the original in error messages. Returns `None` otherwise.
    pub async fn to_generated_code_source(&self) -> Result<Option<AdditionalIssueSource>> {
        if ResolvedVc::try_sidecast::<Box<dyn GenerateSourceMap>>(self.source).is_some() {
            let description = self.source.description().await?;
            let generated = Vc::upcast::<Box<dyn Source>>(GeneratedCodeSource::new(*self.source))
                .to_resolved()
                .await?;
            return Ok(Some(AdditionalIssueSource {
                description: format!("Generated code of {}", description).into(),
                source: IssueSource {
                    source: generated,
                    // The range is intentionally copied verbatim: the offsets
                    // are already in generated-source coordinates (they came
                    // from parsing the loader output), so no remapping is
                    // needed here.
                    range: self.range,
                },
            }));
        }
        Ok(None)
    }
}

impl IssueSource {
    /// Returns bytes offsets corresponding the source range in the format used by swc's Spans.
    pub async fn to_swc_offsets(&self) -> Result<Option<(u32, u32)>> {
        Ok(match &self.range {
            Some(range) => match range {
                SourceRange::ByteOffset(start, end) => Some((*start + 1, *end + 1)),
                SourceRange::LineColumn(start, end) => {
                    if let FileLinesContent::Lines(lines) = &*self.source.content().lines().await? {
                        let start = find_offset(lines.as_ref(), *start) + 1;
                        let end = find_offset(lines.as_ref(), *end) + 1;
                        Some((start, end))
                    } else {
                        None
                    }
                }
            },
            _ => None,
        })
    }
}

async fn source_pos(
    source: ResolvedVc<Box<dyn Source>>,
    start: SourcePos,
    end: SourcePos,
) -> Result<Option<(ResolvedVc<Box<dyn Source>>, SourcePos, SourcePos)>> {
    let Some(generator) = ResolvedVc::try_sidecast::<Box<dyn GenerateSourceMap>>(source) else {
        return Ok(None);
    };

    let srcmap = generator.generate_source_map();
    let Some(srcmap) = &*SourceMap::new_from_rope_cached(srcmap).await? else {
        return Ok(None);
    };

    let find = async |line: u32, col: u32| {
        let TokenWithSource {
            token,
            source_content,
        } = &srcmap.lookup_token_and_source(line, col).await?;

        match token {
            crate::source_map::Token::Synthetic(t) => anyhow::Ok((
                SourcePos {
                    line: t.generated_line as _,
                    column: t.generated_column as _,
                },
                *source_content,
            )),
            crate::source_map::Token::Original(t) => anyhow::Ok((
                SourcePos {
                    line: t.original_line as _,
                    column: t.original_column as _,
                },
                *source_content,
            )),
        }
    };

    let (start, content_1) = find(start.line, start.column).await?;
    let (end, content_2) = find(end.line, end.column).await?;

    let Some((content_1, content_2)) = content_1.zip(content_2) else {
        return Ok(None);
    };

    if content_1 != content_2 {
        return Ok(None);
    }

    Ok(Some((content_1, start, end)))
}

/// A labeled issue source used to provide additional context in error messages.
/// For example, when a webpack loader produces broken code, the primary source
/// shows the original file, while an additional source shows the generated code.
#[turbo_tasks::value(shared)]
pub struct AdditionalIssueSource {
    pub description: RcStr,
    pub source: IssueSource,
}

#[turbo_tasks::value(shared, transparent)]
pub struct AdditionalIssueSources(Vec<AdditionalIssueSource>);

#[turbo_tasks::value_impl]
impl AdditionalIssueSources {
    #[turbo_tasks::function]
    pub fn empty() -> Vc<Self> {
        Vc::cell(Vec::new())
    }
}

// A structured reference to a file with module level details for displaying in an import trace
#[derive(
    Serialize,
    PartialEq,
    Eq,
    PartialOrd,
    Ord,
    Clone,
    Debug,
    TraceRawVcs,
    NonLocalValue,
    DeterministicHash,
)]
#[serde(rename_all = "camelCase")]
pub struct PlainTraceItem {
    // The name of the filesystem
    pub fs_name: RcStr,
    // The root path of the filesystem, for constructing links
    pub root_path: RcStr,
    // The path of the file, relative to the filesystem root
    pub path: RcStr,
    // An optional label attached to the module that clarifies where in the module graph it is.
    pub layer: Option<RcStr>,
}

impl PlainTraceItem {
    async fn from_asset_ident(asset: ReadRef<AssetIdent>) -> Result<Self> {
        // TODO(lukesandberg): How should we display paths? it would be good to display all paths
        // relative to the cwd or the project root.
        let fs_path = asset.path.clone();
        let fs_name = fs_path.fs.to_string().owned().await?;
        let root_path = fs_path.fs.root().await?.path.clone();
        let path = fs_path.path.clone();
        let layer = asset.layer.as_ref().map(Layer::user_friendly_name).cloned();
        Ok(Self {
            fs_name,
            root_path,
            path,
            layer,
        })
    }
}

pub type PlainTrace = Vec<PlainTraceItem>;

// Flatten and simplify this set of import traces into a simpler format for formatting.
async fn into_plain_trace(traces: Vec<Vec<ReadRef<AssetIdent>>>) -> Result<Vec<PlainTrace>> {
    let mut plain_traces = traces
        .into_iter()
        .map(async |trace| {
            let mut plain_trace = trace
                .into_iter()
                .filter(|asset| {
                    // If there are nested assets, this is a synthetic module which is likely to be
                    // confusing/distracting.  Just skip it.
                    asset.assets.is_empty()
                })
                .map(PlainTraceItem::from_asset_ident)
                .try_join()
                .await?;

            // After simplifying the trace, we may end up with apparent duplicates.
            // Consider this example:
            // Import trace:
            // ./[project]/app/global.scss.css [app-client] (css) [app-client]
            // ./[project]/app/layout.js [app-client] (ecmascript) [app-client]
            // ./[project]/app/layout.js [app-rsc] (client reference proxy) [app-rsc]
            // ./[project]/app/layout.js [app-rsc] (ecmascript) [app-rsc]
            // ./[project]/app/layout.js [app-rsc] (ecmascript, Next.js Server Component) [app-rsc]
            //
            // In that case, there are an number of 'shim modules' that are inserted by next with
            // different `modifiers` that are used to model the server->client hand off.  The
            // simplification performed by `PlainTraceItem::from_asset_ident` drops these
            // 'modifiers' and so we would end up with 'app/layout.js' appearing to be duplicated
            // several times.  These modules are implementation details of the application so we
            // just deduplicate them here.

            plain_trace.dedup();

            Ok(plain_trace)
        })
        .try_join()
        .await?;

    // Trim any empty traces and traces that only contain 1 item.  Showing a trace that points to
    // the file with the issue is not useful.
    plain_traces.retain(|t| t.len() > 1);
    // Sort so the shortest traces come first, and break ties by the trace itself to ensure
    // stability
    plain_traces.sort_by(|a, b| {
        // Sort by length first, so that shorter traces come first.
        a.len().cmp(&b.len()).then_with(|| a.cmp(b))
    });

    // Now see if there are any overlaps
    // If two of the traces overlap that means one is a suffix of another one.  Because we are
    // computing shortest paths in the same graph and the shortest path algorithm we use is
    // deterministic.
    // Technically this is a quadratic algorithm since we need to compare each trace with all
    // subsequent traces, however there are rarely more than 3 traces and certainly never more
    // than 10.
    if plain_traces.len() > 1 {
        let mut i = 0;
        while i < plain_traces.len() - 1 {
            let mut j = plain_traces.len() - 1;
            while j > i {
                if plain_traces[j].ends_with(&plain_traces[i]) {
                    // Remove the longer trace.
                    // This typically happens due to things like server->client transitions where
                    // the same file appears multiple times under different modules identifiers.
                    // On the one hand the shorter trace is simpler, on the other hand the longer
                    // trace might be more 'interesting' and even relevant.
                    plain_traces.remove(j);
                }
                j -= 1;
            }
            i += 1;
        }
    }

    Ok(plain_traces)
}

#[turbo_tasks::value(shared)]
#[derive(Clone, Debug, PartialOrd, Ord, DeterministicHash, Serialize)]
pub enum IssueStage {
    Config,
    AppStructure,
    ProcessModule,
    /// Read file.
    Load,
    SourceTransform,
    Parse,
    /// TODO: Add index of the transform
    Transform,
    Analysis,
    Resolve,
    Bindings,
    CodeGen,
    Emit,
    Unsupported,
    Misc,
    Other(RcStr),
}

impl Display for IssueStage {
    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
        match self {
            IssueStage::Config => write!(f, "config"),
            IssueStage::Resolve => write!(f, "resolve"),
            IssueStage::ProcessModule => write!(f, "process module"),
            IssueStage::Load => write!(f, "load"),
            IssueStage::SourceTransform => write!(f, "source transform"),
            IssueStage::Parse => write!(f, "parse"),
            IssueStage::Transform => write!(f, "transform"),
            IssueStage::Analysis => write!(f, "analysis"),
            IssueStage::Bindings => write!(f, "bindings"),
            IssueStage::CodeGen => write!(f, "code gen"),
            IssueStage::Emit => write!(f, "emit"),
            IssueStage::Unsupported => write!(f, "unsupported"),
            IssueStage::AppStructure => write!(f, "app structure"),
            IssueStage::Misc => write!(f, "misc"),
            IssueStage::Other(s) => write!(f, "{s}"),
        }
    }
}

#[turbo_tasks::value(serialization = "skip")]
#[derive(Clone, Debug, PartialOrd, Ord)]
pub struct PlainIssue {
    pub severity: IssueSeverity,
    pub stage: IssueStage,

    pub title: StyledString,
    pub file_path: RcStr,

    pub description: Option<StyledString>,
    pub detail: Option<StyledString>,
    pub documentation_link: RcStr,

    pub source: Option<PlainIssueSource>,
    pub additional_sources: Vec<PlainAdditionalIssueSource>,
    pub import_traces: Vec<PlainTrace>,
}

/// A collection of [`PlainIssue`]s collected from a single source.
///
/// Returned by [`collect_issues`] so that the (plain) issues can be read strongly
/// consistently from a top-level task and handed to a non-turbo-task [`IssueReporter`].
#[turbo_tasks::value(serialization = "skip")]
#[derive(Debug)]
pub struct PlainIssues(pub Vec<ReadRef<PlainIssue>>);

#[turbo_tasks::value(serialization = "skip")]
#[derive(Clone, Debug, PartialOrd, Ord)]
pub struct PlainAdditionalIssueSource {
    pub description: RcStr,
    pub source: PlainIssueSource,
}

fn hash_plain_issue(issue: &PlainIssue, hasher: &mut Xxh3Hash64Hasher, full: bool) {
    hasher.write_ref(&issue.severity);
    hasher.write_ref(&issue.file_path);
    hasher.write_ref(&issue.stage);
    hasher.write_ref(&issue.title);
    hasher.write_ref(&issue.description);
    hasher.write_ref(&issue.detail);
    hasher.write_ref(&issue.documentation_link);

    if let Some(source) = &issue.source {
        hasher.write_value(1_u8);
        // I'm assuming we don't need to hash the contents. Not 100% correct, but
        // probably 99%.
        hasher.write_ref(&source.range);
    } else {
        hasher.write_value(0_u8);
    }

    // `additional_sources` is intentionally not hashed: it carries supplementary
    // display info (e.g. generated code from a loader) that does not change the
    // identity of the underlying problem.  Two issues that differ only in their
    // generated-code snippet still represent the same root cause and should be
    // deduplicated.

    if full {
        hasher.write_ref(&issue.import_traces);
    }
}

impl PlainIssue {
    /// We need deduplicate issues that can come from unique paths, but represent the same
    /// underlying problem. E.g., a parse error for a file that is compiled in both client and
    /// server contexts.
    ///
    /// Passing `full` will also hash any sub-issues and processing paths. While useful for
    /// generating exact matching hashes, it's possible for the same issue to pass from multiple
    /// processing paths, making for overly verbose logging.
    pub fn internal_hash_ref(&self, full: bool) -> u64 {
        let mut hasher = Xxh3Hash64Hasher::new();
        hash_plain_issue(self, &mut hasher, full);
        hasher.finish()
    }
}

#[turbo_tasks::value_impl]
impl PlainIssue {
    /// Translate an [Issue] into a [PlainIssue]. A more regular structure suitable for printing and
    /// serialization.
    #[turbo_tasks::function]
    pub async fn from_issue(
        issue: ResolvedVc<Box<dyn Issue>>,
        import_tracer: Option<ResolvedVc<DelegatingImportTracer>>,
    ) -> Result<Vc<Self>> {
        Ok(
            Self::from_issue_ref(&*issue.into_trait_ref().await?, import_tracer)
                .await?
                .cell(),
        )
    }
}

impl PlainIssue {
    pub async fn from_issue_ref(
        trait_ref: &dyn Issue,
        import_tracer: Option<ResolvedVc<DelegatingImportTracer>>,
    ) -> Result<Self> {
        let severity = trait_ref.severity();
        let file_path = trait_ref.file_path().await?;
        let file_path_str = file_path.to_string_ref().await?;

        Ok(Self {
            severity,
            file_path: file_path_str,
            stage: trait_ref.stage(),
            title: trait_ref.title().await?,
            description: trait_ref.description().await?,
            detail: trait_ref.detail().await?,
            documentation_link: trait_ref.documentation_link(),
            source: {
                if let Some(s) = trait_ref.source() {
                    Some(s.into_plain().await?)
                } else {
                    None
                }
            },
            additional_sources: {
                trait_ref
                    .additional_sources()
                    .await?
                    .into_iter()
                    .map(async |s| {
                        Ok(PlainAdditionalIssueSource {
                            source: s.source.into_plain().await?,
                            description: s.description,
                        })
                    })
                    .try_join()
                    .await?
            },
            import_traces: match import_tracer {
                Some(tracer) => {
                    into_plain_trace(tracer.await?.get_traces(file_path).await?).await?
                }
                None => vec![],
            },
        })
    }
}

#[turbo_tasks::value(serialization = "skip")]
#[derive(Clone, Debug, PartialOrd, Ord)]
pub struct PlainIssueSource {
    pub asset: ReadRef<PlainSource>,
    pub range: Option<(SourcePos, SourcePos)>,
}

#[turbo_tasks::value(serialization = "skip")]
#[derive(Clone, Debug, PartialOrd, Ord)]
pub struct PlainSource {
    pub ident: RcStr,
    pub file_path: RcStr,
    #[turbo_tasks(debug_ignore)]
    pub content: ReadRef<FileContent>,
}

#[turbo_tasks::value_impl]
impl PlainSource {
    #[turbo_tasks::function]
    pub async fn from_source(asset: ResolvedVc<Box<dyn Source>>) -> Result<Vc<PlainSource>> {
        // Defensively read the content, an error there should not prevent all issue
        // formatting.  Best practice is for `content` to return `NotFound` instead of
        // an error.
        let content = if let Ok(asset_content) = asset.content().await
            && let AssetContent::File(file_content) = &*asset_content
            && let Ok(file_content) = file_content.await
        {
            file_content
        } else {
            ReadRef::new_owned(FileContent::NotFound)
        };
        let ident = asset.ident();

        Ok(PlainSource {
            ident: ident.to_string().owned().await?,
            file_path: ident.await?.path.to_string_ref().await?,
            content,
        }
        .cell())
    }
}

#[async_trait]
#[turbo_tasks::value_trait]
pub trait IssueReporter {
    /// Reports already-collected issues to the user (e.g. to stdio). Returns whether fatal
    /// (program-ending) issues were present.
    ///
    /// This is intentionally *not* a `#[turbo_tasks::function]`: it performs no turbo-tasks
    /// reads of its own (the issues are collected ahead of time by [`collect_issues`]), so it
    /// is safe to call from a top-level task.
    ///
    /// # Arguments:
    ///
    /// * `issues` - The plain issues already collected from the source.
    /// * `source` - The root [`RawVc`] from which the issues were traced. Can be used by
    ///   implementers as a dedup key to determine which issues are new. This must be derived from
    ///   the `OperationVc` the issues were collected from.
    /// * `min_failing_severity` - The minimum issue severity level considered to fatally end the
    ///   program.
    async fn report_issues(
        &self,
        issues: ReadRef<PlainIssues>,
        source: RawVc,
        min_failing_severity: IssueSeverity,
    ) -> Result<bool>;
}

pub trait CollectibleIssuesExt
where
    Self: Sized,
{
    /// Returns all issues from `source`
    ///
    /// Must be called in a turbo-task as this constructs a `cell`
    fn peek_issues(self) -> CapturedIssues;

    /// Drops all issues from `source`
    ///
    /// This unemits the issues. They will not propagate up.
    fn drop_issues(self);
}

impl<T> CollectibleIssuesExt for T
where
    T: CollectiblesSource + Copy + Send,
{
    fn peek_issues(self) -> CapturedIssues {
        CapturedIssues {
            issues: self.peek_collectibles(),

            tracer: DelegatingImportTracer {
                delegates: self.peek_collectibles(),
            }
            .resolved_cell(),
        }
    }

    fn drop_issues(self) {
        self.drop_collectibles::<Box<dyn Issue>>();
    }
}

/// Collects all issues emitted by `source` as resolved [`PlainIssue`]s.
///
/// This is an `operation` function so its (plain) result can be read *strongly consistently* (via
/// [`OperationVc::read_strongly_consistent`]) from a top-level task without tripping the
/// eventually-consistent-read assertion. The per-issue `PlainIssue::from_issue` reads happen
/// *inside* this task, where eventually-consistent reads are legal.
#[turbo_tasks::function(operation, root)]
async fn collect_issues(source: OperationVc<()>) -> Result<Vc<PlainIssues>> {
    let plain = source
        .peek_issues()
        .get_plain_issues(&IssueFilter::everything())
        .await?;
    Ok(PlainIssues(plain).cell())
}

/// A helper function to print out issues to the console.
///
/// Must be called in a turbo-task as this constructs a `cell`
pub async fn handle_issues<T: Send>(
    source_op: OperationVc<T>,
    issue_reporter: Vc<Box<dyn IssueReporter>>,
    min_failing_severity: IssueSeverity,
    path: Option<&str>,
    operation: Option<&str>,
) -> Result<()> {
    let source_vc = source_op.connect();
    let _ = source_op.resolve().strongly_consistent().await?;
    let source_raw = Vc::into_raw(source_vc);

    // Collect the issues in a dedicated `operation` task and read its *plain* result strongly
    // consistently. This is safe at the top level (unlike an eventually-consistent read), while
    // the per-issue reads happen inside `collect_issues`. The source is type-erased to
    // `OperationVc<()>` so a single non-generic task can collect issues for any source.
    let erased_source = OperationVc::<()>::try_from(source_raw)?;
    let issues = collect_issues(erased_source)
        .read_strongly_consistent()
        .await?;

    // `report_issues` is a plain async method; reach it via a `TraitRef`. Resolve the reporter
    // strongly consistently first so that `into_trait_ref` is a plain cell read (rather than an
    // eventually-consistent task-output read) at the top level.
    let reporter = issue_reporter
        .to_resolved()
        .strongly_consistent()
        .await?
        .into_trait_ref()
        .await?;
    let has_fatal = reporter
        .report_issues(issues, source_raw, min_failing_severity)
        .await?;

    if has_fatal {
        let mut message = "Fatal issue(s) occurred".to_owned();
        if let Some(path) = path.as_ref() {
            message += &format!(" in {path}");
        };
        if let Some(operation) = operation.as_ref() {
            message += &format!(" ({operation})");
        };

        bail!(message)
    } else {
        Ok(())
    }
}

fn find_line_and_column(lines: &[FileLine], offset: u32) -> SourcePos {
    match lines.binary_search_by(|line| line.bytes_offset.cmp(&offset)) {
        Ok(i) => SourcePos {
            line: i as u32,
            column: 0,
        },
        Err(i) => {
            if i == 0 {
                SourcePos {
                    line: 0,
                    column: offset,
                }
            } else {
                let line = &lines[i - 1];
                SourcePos {
                    line: (i - 1) as u32,
                    column: min(line.content.len() as u32, offset - line.bytes_offset),
                }
            }
        }
    }
}

fn find_offset(lines: &[FileLine], pos: SourcePos) -> u32 {
    let line = &lines[pos.line as usize];
    line.bytes_offset + pos.column
}