rumdl_lib/

rule.rs

//!
//! This module defines the Rule trait and related types for implementing linting rules in rumdl.

use dyn_clone::DynClone;
use serde::{Deserialize, Serialize};
use std::ops::Range;
use thiserror::Error;

use crate::lint_context::LintContext;

// Macro to implement box_clone for Rule implementors
#[macro_export]
macro_rules! impl_rule_clone {
    ($ty:ty) => {
        impl $ty {
            fn box_clone(&self) -> Box<dyn Rule> {
                Box::new(self.clone())
            }
        }
    };
}

#[derive(Debug, Error)]
pub enum LintError {
    #[error("Invalid input: {0}")]
    InvalidInput(String),
    #[error("Fix failed: {0}")]
    FixFailed(String),
    #[error("IO error: {0}")]
    IoError(#[from] std::io::Error),
    #[error("Parsing error: {0}")]
    ParsingError(String),
}

pub type LintResult = Result<Vec<LintWarning>, LintError>;

#[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
pub struct LintWarning {
    pub message: String,
    pub line: usize, // 1-indexed start line
    /// 1-indexed start column, measured in **characters** (not bytes).
    /// When deriving a column from a byte offset (regex match, `str::find`,
    /// parser byte offset), convert with `range_utils::byte_to_char_count` or a
    /// character-based range helper. A raw byte offset mis-positions the
    /// highlight on lines containing multi-byte UTF-8.
    pub column: usize,
    pub end_line: usize, // 1-indexed end line
    /// 1-indexed end column, measured in **characters** (see `column`). Use
    /// `str::chars().count()`, not `str::len()`, when computing a span width.
    pub end_column: usize,
    pub severity: Severity,
    pub fix: Option<Fix>,
    pub rule_name: Option<String>,
}

impl LintWarning {
    /// The 1-indexed lines whose content this warning covers, used to decide
    /// whether an inline disable comment suppresses it.
    ///
    /// `end_column` is exclusive, so a range ending at column 1 of a later line
    /// stops at that line's start and covers none of it: a warning spanning
    /// whole lines with their terminators ends there. That line is excluded,
    /// so a `disable-line` comment on the content after a flagged run does not
    /// suppress the run. An `end_line` before `line` falls back to `line` alone.
    pub fn covered_lines(&self) -> std::ops::RangeInclusive<usize> {
        let end = if self.end_line > self.line && self.end_column <= 1 {
            self.end_line - 1
        } else {
            self.end_line.max(self.line)
        };
        self.line..=end
    }
}

/// One atomic fix attached to a `LintWarning`.
///
/// `range`/`replacement` describe the primary edit. `additional_edits`
/// carries any *paired* edits that must apply together with the primary one
/// for the result to be a valid document — for example, MD054's conversion
/// of an inline link to a reference style produces the in-place link rewrite
/// **and** a reference-definition append at end-of-file; applying only one
/// half would leave a dangling reference.
///
/// All fix consumers (the LSP code-action layer, CLI counters, the
/// `apply_warning_fixes` helper) treat the primary edit and the
/// `additional_edits` as a single unit. The field is empty by default so
/// rules that only need a single-location fix can keep using
/// `Fix::new(range, replacement)`; only rules that need multi-location
/// atomicity populate it via `Fix::with_additional_edits(...)`.
///
/// `additional_edits` is intentionally a flat `Vec<Fix>` — nesting beyond
/// one level isn't needed today and would complicate the apply contract.
/// Apply order is "primary first, then additional in their declared order"
/// when offsets are non-overlapping; consumers that batch multiple fixes
/// across warnings still sort by `range.start` descending so earlier offsets
/// remain valid as later edits mutate the buffer.
#[derive(Debug, PartialEq, Clone, Default, Serialize, Deserialize)]
pub struct Fix {
    /// Byte range in the text supplied to the producing API. Public document
    /// linting includes any leading BOM; raw `Rule::check` ranges address the
    /// rule's `LintContext::content`.
    pub range: Range<usize>,
    pub replacement: String,
    /// Edits applied atomically with the primary `range`/`replacement` pair.
    /// Empty for the common single-edit case. See struct docs for semantics.
    #[serde(default, skip_serializing_if = "Vec::is_empty")]
    pub additional_edits: Vec<Fix>,
}

impl Fix {
    /// Construct a single-edit fix. Use this for the overwhelming common case
    /// where a fix is one in-place replacement.
    pub fn new(range: Range<usize>, replacement: String) -> Self {
        Self {
            range,
            replacement,
            additional_edits: Vec::new(),
        }
    }

    /// Construct a multi-edit fix bundle. The primary edit is applied first,
    /// followed by every entry in `additional_edits` as part of the same
    /// atomic operation.
    pub fn with_additional_edits(range: Range<usize>, replacement: String, additional_edits: Vec<Fix>) -> Self {
        Self {
            range,
            replacement,
            additional_edits,
        }
    }
}

#[derive(Debug, PartialEq, Eq, Hash, Clone, Copy, Serialize, schemars::JsonSchema)]
#[serde(rename_all = "lowercase")]
pub enum Severity {
    Error,
    Warning,
    Info,
}

impl<'de> serde::Deserialize<'de> for Severity {
    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
    where
        D: serde::Deserializer<'de>,
    {
        let s = String::deserialize(deserializer)?;
        match s.to_lowercase().as_str() {
            "error" => Ok(Severity::Error),
            "warning" => Ok(Severity::Warning),
            "info" => Ok(Severity::Info),
            _ => Err(serde::de::Error::custom(format!(
                "Invalid severity: '{s}'. Valid values: error, warning, info"
            ))),
        }
    }
}

/// Type of rule for selective processing
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum RuleCategory {
    Heading,
    List,
    CodeBlock,
    Link,
    Image,
    Html,
    Emphasis,
    Whitespace,
    Blockquote,
    Table,
    FrontMatter,
    Other,
}

/// Capability of a rule to fix issues
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum FixCapability {
    /// Rule can automatically fix all violations it detects
    FullyFixable,
    /// Rule can fix some violations based on context
    ConditionallyFixable,
    /// Rule cannot fix violations (by design)
    Unfixable,
}

/// Declares what cross-file data a rule needs
///
/// Most rules only need single-file context and should use `None` (the default).
/// Rules that need to validate references across files (like MD051) should use `Workspace`.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
pub enum CrossFileScope {
    /// Single-file only - no cross-file analysis needed (default for 99% of rules)
    #[default]
    None,
    /// Needs workspace-wide index for cross-file validation
    Workspace,
}

/// A warning an inline disable comment kept out of a document's results.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct SuppressedWarning {
    /// Canonical id of the rule that raised the warning, e.g. `MD013`
    pub rule_name: String,
    /// 1-indexed line of the warning's range the rule was found disabled on
    pub line: usize,
    /// The kind of directive that disabled the rule there
    pub layer: crate::inline_config::DisableLayer,
}

/// What a run's inline disable comments actually suppressed.
///
/// Assembled once per document, after every single-file rule has run, so a rule
/// reading it sees the complete picture.
#[derive(Debug, Clone, Default)]
pub struct SuppressionReport {
    /// Every warning an inline disable comment removed, in the order raised
    pub suppressed: Vec<SuppressedWarning>,
    /// Canonical ids of the rules whose findings this report accounts for.
    ///
    /// A rule outside this set produced nothing the report can see, so nothing
    /// can be concluded about a comment naming it.
    pub judged_rules: std::collections::HashSet<String>,
}

pub trait Rule: DynClone + Send + Sync {
    fn name(&self) -> &'static str;
    fn description(&self) -> &'static str;
    /// Return raw diagnostics, including findings covered by inline disables.
    /// The lint driver filters them and records which comments suppressed them
    /// for observers such as MD087. Filtering them here loses that accounting.
    /// Disabled regions may still be excluded from inferred settings, and `fix`
    /// must honor them when rewriting content.
    fn check(&self, ctx: &LintContext) -> LintResult;
    fn fix(&self, ctx: &LintContext) -> Result<String, LintError>;

    /// Return the native fix when computing findings already produces it.
    /// The result belongs only to this configured instance and exact context;
    /// callers discard it with the findings after any source rewrite.
    #[doc(hidden)]
    fn check_with_prepared_fix(&self, ctx: &LintContext) -> Result<(Vec<LintWarning>, Option<String>), LintError> {
        self.check(ctx).map(|warnings| (warnings, None))
    }

    /// Apply this rule's native fixer using its already-computed findings.
    /// Callers must supply raw diagnostics from this configured instance and
    /// this exact immutable context. No findings survive a source rewrite.
    /// Custom rules keep their normal fixer unless they explicitly opt in.
    #[doc(hidden)]
    fn fix_with_diagnostics(&self, ctx: &LintContext, _warnings: &[LintWarning]) -> Result<String, LintError> {
        self.fix(ctx)
    }

    /// Whether CLI fixing is exactly the application of this rule's diagnostic
    /// edits, filtered by inline disables. The engine can then consume the
    /// already-computed edits without checking the same source a second time.
    ///
    /// Opt in only when `fix()` performs no additional transformations. The
    /// default preserves custom fix implementations and third-party rules.
    /// When opting in, `should_skip(ctx)` must imply that applying the raw
    /// check diagnostics would leave the source unchanged as well.
    #[doc(hidden)]
    fn fixes_from_diagnostics(&self) -> bool {
        false
    }

    /// Built-in-only certificate for proving that a source revision needs no fix.
    /// External rules retain their existing check and fix behavior by default.
    #[doc(hidden)]
    fn native_format_policy(&self) -> Option<crate::format_proof::NativeFormatPolicy> {
        None
    }

    /// Check if this rule should quickly skip processing based on content.
    fn should_skip(&self, _ctx: &LintContext) -> bool {
        false
    }

    /// Get the category of this rule for selective processing
    fn category(&self) -> RuleCategory {
        RuleCategory::Other // Default implementation returns Other
    }

    /// Whether the content-category prefilter may skip this rule.
    ///
    /// The prefilter reads the document's shape alone: a `Link` rule is skipped
    /// for a document holding no links. A rule whose configuration widens what
    /// it reads answers `false` for that configuration, so that `should_skip`
    /// and `check` decide instead. MD051 and MD057 read frontmatter values on
    /// request, and a document can carry those with no link syntax at all.
    fn skippable_by_category(&self) -> bool {
        true
    }

    /// Return this concrete rule instance for type-based queries.
    fn as_any(&self) -> &dyn std::any::Any;

    // DocumentStructure has been merged into LintContext - this method is no longer used
    // fn as_maybe_document_structure(&self) -> Option<&dyn MaybeDocumentStructure> {
    //     None
    // }

    /// Returns the rule name and default config table if the rule has config.
    /// If a rule implements this, it MUST be defined on the `impl Rule for ...` block,
    /// not just the inherent impl.
    ///
    /// This is user-facing: it backs `rumdl config`, `rumdl config --defaults` and
    /// `rumdl explain`, so every value here must be a real default the user could write
    /// back into a config file. A key whose default cannot be written down (an unset
    /// `Option`) is therefore absent. Config *validation* reads [`Rule::config_schema`]
    /// instead, which keeps such keys.
    fn default_config_section(&self) -> Option<(String, toml::Value)> {
        None
    }

    /// Returns the rule name and every config key the rule accepts, for validation.
    ///
    /// A key with no representable default (an unset `Option`, or a deserializer that
    /// accepts several TOML types) carries a sentinel value: the key name is recognized
    /// while its type check is skipped. Sentinels contain a NUL byte and must never
    /// reach user-facing output, which is why this is separate from
    /// [`Rule::default_config_section`].
    ///
    /// Defaults to the user-facing table, which is correct for a rule whose every key
    /// has a representable default.
    fn config_schema(&self) -> Option<(String, toml::Value)> {
        self.default_config_section()
    }

    /// Returns config key aliases for this rule
    /// This allows rules to accept alternative config key names for backwards compatibility
    fn config_aliases(&self) -> Option<std::collections::HashMap<String, String>> {
        None
    }

    /// Returns the list of config keys whose deserializer accepts more than one TOML
    /// type (e.g. either a scalar or a list). The schema is built from a serialized
    /// default that can only encode one variant, so the validator would reject the
    /// alternative form. The registry replaces the schema entry for each listed key
    /// with a polymorphic sentinel so type checking is skipped while the key name
    /// is still validated. The registry rewrites [`Rule::config_schema`], so
    /// `default_config_section()` keeps returning the clean user-facing default.
    fn polymorphic_config_keys(&self) -> &'static [&'static str] {
        &[]
    }

    /// Declares the fix capability of this rule
    fn fix_capability(&self) -> FixCapability {
        FixCapability::FullyFixable // Safe default for backward compatibility
    }

    /// Declares cross-file analysis requirements for this rule
    ///
    /// Returns `CrossFileScope::None` by default, meaning the rule only needs
    /// single-file context. Rules that need workspace-wide data should override
    /// this to return `CrossFileScope::Workspace`.
    fn cross_file_scope(&self) -> CrossFileScope {
        CrossFileScope::None
    }

    /// Contribute data to the workspace index during linting
    ///
    /// Called during the single-file linting phase for rules that return
    /// `CrossFileScope::Workspace`. Rules should extract headings, links,
    /// and other data needed for cross-file validation.
    ///
    /// This is called as a side effect of linting, so LintContext is already
    /// created - no duplicate parsing required.
    fn contribute_to_index(&self, _ctx: &LintContext, _file_index: &mut crate::workspace_index::FileIndex) {
        // Default: no contribution
    }

    /// Perform cross-file validation after all files have been linted
    ///
    /// Called once per file after the entire workspace has been indexed.
    /// Rules receive the file_index (from contribute_to_index) and the full
    /// workspace_index for cross-file lookups.
    ///
    /// Note: This receives the FileIndex instead of LintContext to avoid re-parsing
    /// each file. The FileIndex was already populated during contribute_to_index.
    ///
    /// Rules can use workspace_index methods for cross-file validation:
    /// - `get_file(path)` - to look up headings in target files (for MD051)
    /// - `files()` - to iterate all indexed files
    ///
    /// Returns additional warnings for cross-file issues. These are appended
    /// to the single-file warnings.
    fn cross_file_check(
        &self,
        _file_path: &std::path::Path,
        _file_index: &crate::workspace_index::FileIndex,
        _workspace_index: &crate::workspace_index::WorkspaceIndex,
    ) -> LintResult {
        Ok(Vec::new()) // Default: no cross-file warnings
    }

    /// Whether this rule reports on the inline comments that suppressed warnings
    ///
    /// Recording every suppression costs work on each file, so the linting driver
    /// does it only when a rule asks for it. A rule answering `true` receives the
    /// result through `check_suppressions`.
    fn observes_suppressions(&self) -> bool {
        false
    }

    /// Report on a document's inline disable comments
    ///
    /// Called once per document after every single-file rule has run, for rules
    /// that return `true` from `observes_suppressions`. The report says which
    /// warnings the comments removed and which rules the run can account for.
    ///
    /// Returns warnings that join the single-file warnings.
    fn check_suppressions(&self, _ctx: &LintContext, _report: &SuppressionReport) -> LintResult {
        Ok(Vec::new()) // Default: nothing to report
    }

    /// Factory: create a rule from config (if present), or use defaults.
    fn from_config(_config: &crate::config::Config) -> Box<dyn Rule>
    where
        Self: Sized,
    {
        panic!(
            "from_config not implemented for rule: {}",
            std::any::type_name::<Self>()
        );
    }
}

// Implement the cloning logic for the Rule trait object
dyn_clone::clone_trait_object!(Rule);

/// Extension trait to add downcasting capabilities to Rule
pub trait RuleExt {
    fn downcast_ref<T: 'static>(&self) -> Option<&T>;
}

impl<R: Rule + 'static> RuleExt for Box<R> {
    fn downcast_ref<T: 'static>(&self) -> Option<&T> {
        if std::any::TypeId::of::<R>() == std::any::TypeId::of::<T>() {
            unsafe { Some(&*std::ptr::from_ref(self.as_ref()).cast::<T>()) }
        } else {
            None
        }
    }
}

// Inline config parsing functions are in inline_config.rs.
// Use InlineConfig::from_content() for the full inline configuration system,
// or inline_config::parse_disable_comment/parse_enable_comment for low-level parsing.

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

    #[test]
    fn test_severity_serialization() {
        let warning = LintWarning {
            message: "Test warning".to_string(),
            line: 1,
            column: 1,
            end_line: 1,
            end_column: 10,
            severity: Severity::Warning,
            fix: None,
            rule_name: Some("MD001".to_string()),
        };

        let serialized = serde_json::to_string(&warning).unwrap();
        assert!(serialized.contains("\"severity\":\"warning\""));

        let error = LintWarning {
            severity: Severity::Error,
            ..warning
        };

        let serialized = serde_json::to_string(&error).unwrap();
        assert!(serialized.contains("\"severity\":\"error\""));
    }

    #[test]
    fn test_fix_serialization() {
        let fix = Fix::new(0..10, "fixed text".to_string());

        let warning = LintWarning {
            message: "Test warning".to_string(),
            line: 1,
            column: 1,
            end_line: 1,
            end_column: 10,
            severity: Severity::Warning,
            fix: Some(fix),
            rule_name: Some("MD001".to_string()),
        };

        let serialized = serde_json::to_string(&warning).unwrap();
        assert!(serialized.contains("\"fix\""));
        assert!(serialized.contains("\"replacement\":\"fixed text\""));
    }

    #[test]
    fn test_rule_category_equality() {
        assert_eq!(RuleCategory::Heading, RuleCategory::Heading);
        assert_ne!(RuleCategory::Heading, RuleCategory::List);

        // Test all categories are distinct
        let categories = [
            RuleCategory::Heading,
            RuleCategory::List,
            RuleCategory::CodeBlock,
            RuleCategory::Link,
            RuleCategory::Image,
            RuleCategory::Html,
            RuleCategory::Emphasis,
            RuleCategory::Whitespace,
            RuleCategory::Blockquote,
            RuleCategory::Table,
            RuleCategory::FrontMatter,
            RuleCategory::Other,
        ];

        for (i, cat1) in categories.iter().enumerate() {
            for (j, cat2) in categories.iter().enumerate() {
                if i == j {
                    assert_eq!(cat1, cat2);
                } else {
                    assert_ne!(cat1, cat2);
                }
            }
        }
    }

    #[test]
    fn test_lint_error_conversions() {
        use std::io;

        // Test From<io::Error>
        let io_error = io::Error::new(io::ErrorKind::NotFound, "file not found");
        let lint_error: LintError = io_error.into();
        match lint_error {
            LintError::IoError(_) => {}
            _ => panic!("Expected IoError variant"),
        }

        // Test Display trait
        let invalid_input = LintError::InvalidInput("bad input".to_string());
        assert_eq!(invalid_input.to_string(), "Invalid input: bad input");

        let fix_failed = LintError::FixFailed("couldn't fix".to_string());
        assert_eq!(fix_failed.to_string(), "Fix failed: couldn't fix");

        let parsing_error = LintError::ParsingError("parse error".to_string());
        assert_eq!(parsing_error.to_string(), "Parsing error: parse error");
    }
}