ripvec-core 4.0.7

//! Per-language entry-point detection for the `find_dead_code` MCP tool
//! (4.1.0).
//!
//! The trait [`EntryPointDetector`] and its per-language implementors
//! ([`RustEntryDetector`], [`PythonEntryDetector`], [`GoEntryDetector`])
//! identify the syntactic shapes that act as roots of the call graph: the
//! BFS reachability walk for dead-code detection seeds from the union of
//! all [`EntryPoint`]s emitted across the indexed corpus.
//!
//! This module is X1 of the 4.1.0 series; the actual reachability walk and
//! cluster discovery (`RepoGraph::compute_dead_code`) lands in X2. The MCP
//! tool wrapper lands in X3. The remaining language detectors land in X4.
//! See `docs/FIND_DEAD_CODE_DESIGN.md` Section 2 for the per-language
//! entry-point survey and Section 3 for the algorithm that consumes this
//! output.
//!
//! ## Type B (Wired-Stub) self-audit note
//!
//! Until X2 lands, every public item in this module is consumed only from
//! the integration tests under `crates/ripvec-core/tests/entry_points.rs`.
//! `scripts/check_wiring_gaps.sh` will report these as Type B findings.
//! The findings are **explicitly deferred** to X2 — see the Section 9
//! PLAN.md entry — not silently dangling. Do not annotate with
//! `#[doc(hidden)]`: the doc-visibility surface is part of the X2 contract
//! and is the intended public API of the dead-code module.

use std::path::{Path, PathBuf};

use streaming_iterator::StreamingIterator;
use tree_sitter::{Node, Parser, Query, QueryCursor};

/// Classification of why a [`Definition`](crate::repo_map::Definition)-shaped
/// item is treated as an entry point for the dead-code reachability walk.
///
/// Categories follow Section 2 of `docs/FIND_DEAD_CODE_DESIGN.md`. The
/// classification is per-detection, not per-definition: the same
/// `pub fn` can appear as both [`EntryPointKind::Main`] (for binaries) and
/// [`EntryPointKind::LibraryExport`] (for libraries) depending on how the
/// containing crate is structured. Downstream consumers (X2) treat each
/// detection independently.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum EntryPointKind {
    /// A binary-crate `main`-shaped entry: `fn main()` in Rust, `func main()`
    /// in Go, the `if __name__ == "__main__"` block in Python.
    Main,

    /// A public-API surface item: `pub` re-exports in Rust libraries,
    /// `__all__` exports in Python, capitalised names in Go libraries.
    LibraryExport,

    /// A test entry: `#[test]` / `#[bench]` in Rust, `def test_*` /
    /// `*_test.py` in Python, `func TestX` / `BenchmarkX` / `ExampleX` /
    /// `FuzzX` in Go.
    Test,

    /// A foreign-function-interface entry: `#[no_mangle]` /
    /// `extern "C"` in Rust, cgo `//export` in Go.
    Ffi,

    /// A procedural-macro entry: `#[proc_macro]`, `#[proc_macro_derive]`,
    /// `#[proc_macro_attribute]` in Rust.
    ProcMacro,

    /// A package-initialisation entry: `func init()` in Go.
    Init,

    /// A build-script entry: Cargo's `build.rs`.
    BuildScript,
}

/// A single entry-point detection in one source file.
///
/// Per-detection, not per-definition — the same `pub fn` can produce
/// multiple `EntryPoint` instances (one for each matching predicate).
/// Downstream consumers should treat each detection as an independent
/// reachability seed.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct EntryPoint {
    /// The symbol name of the entry point. For Rust this is the function
    /// item identifier; for Python it is the function or module-level
    /// expression name; for Go it is the function declaration identifier.
    pub name: String,

    /// Why this item was treated as an entry point.
    pub kind: EntryPointKind,

    /// The source file the entry point was detected in.
    pub file_path: PathBuf,

    /// 1-based line number of the entry point declaration. Matches the
    /// `start_line` field of [`crate::repo_map::Definition`].
    pub line: u32,
}

/// Per-language entry-point detector.
///
/// Designed for consumption by `RepoGraph::compute_dead_code` in
/// 4.1.0-X2. Until X2 lands, the only consumers are the integration tests
/// under `crates/ripvec-core/tests/entry_points.rs` — see the
/// module-level docstring for the Type B (Wired-Stub) self-audit note.
///
/// Implementations parse the source once per call. The parsing cost is
/// trivial (tree-sitter is O(n) and the source is already in memory at
/// detection time), and stateless parsers compose more cleanly than a
/// shared parser cache across the three (and eventually eleven) language
/// detectors. X2's `RepoGraph::compute_dead_code` already iterates
/// per-file, so the per-file parse adds no additional walk cost.
pub trait EntryPointDetector {
    /// Return every entry point declared in this source file.
    ///
    /// `source` is the full UTF-8 contents of `file_path`. The path is
    /// passed alongside `source` so detectors that consider filename
    /// patterns (e.g. Python's `test_*.py` and `*_test.py`,
    /// Rust's `build.rs`) can use both signals.
    ///
    /// If parsing fails, returns an empty vector — entry-point detection
    /// is best-effort and should never abort the dead-code walk.
    fn detect(&self, source: &str, file_path: &Path) -> Vec<EntryPoint>;
}

// ---------------------------------------------------------------------------
// Rust detector
// ---------------------------------------------------------------------------

/// Rust entry-point detector.
///
/// Detects (per `docs/FIND_DEAD_CODE_DESIGN.md` Section 2):
/// - `pub fn main()` and bare `fn main()` (Main)
/// - `pub fn` items in `lib.rs` / `mod.rs` (LibraryExport)
/// - Items annotated with `#[test]` or `#[bench]` (Test)
/// - Items annotated with `#[no_mangle]` or marked `extern "C"` (Ffi)
/// - Items annotated with `#[proc_macro]`, `#[proc_macro_derive]`, or
///   `#[proc_macro_attribute]` (ProcMacro)
/// - The entire `build.rs` file is treated as a single BuildScript entry
///   point (the build script's `main` is the cargo-known entry).
#[derive(Debug, Default, Clone, Copy)]
pub struct RustEntryDetector;

impl EntryPointDetector for RustEntryDetector {
    fn detect(&self, source: &str, file_path: &Path) -> Vec<EntryPoint> {
        let mut entries = Vec::new();
        let Some(tree) = parse_with(source, &tree_sitter_rust::LANGUAGE.into()) else {
            return entries;
        };
        let root = tree.root_node();
        let bytes = source.as_bytes();

        // Treat the entire build.rs file as a single BuildScript entry.
        // The crate's main may be named anything inside build.rs (cargo
        // calls the file's main), so we emit one entry at line 1.
        if file_path.file_name().and_then(|s| s.to_str()) == Some("build.rs") {
            entries.push(EntryPoint {
                name: "build.rs".to_string(),
                kind: EntryPointKind::BuildScript,
                file_path: file_path.to_path_buf(),
                line: 1,
            });
        }

        let is_lib_or_mod_rs = matches!(
            file_path.file_name().and_then(|s| s.to_str()),
            Some("lib.rs" | "mod.rs")
        );

        // Walk every function_item declaration recursively. For each item:
        //   - inspect its preceding attribute_item siblings for #[test],
        //     #[bench], #[no_mangle], #[proc_macro*]
        //   - inspect the function_item's own modifiers for `extern "C"`
        //   - inspect the name for `main`
        //   - if file is lib.rs/mod.rs and the item is `pub`, emit
        //     LibraryExport
        visit_rust_node(&root, bytes, file_path, is_lib_or_mod_rs, &mut entries);
        entries
    }
}

fn visit_rust_node(
    node: &Node<'_>,
    bytes: &[u8],
    file_path: &Path,
    is_lib_or_mod_rs: bool,
    out: &mut Vec<EntryPoint>,
) {
    if node.kind() == "function_item" {
        rust_classify_function(node, bytes, file_path, is_lib_or_mod_rs, out);
    }
    let mut cursor = node.walk();
    for child in node.children(&mut cursor) {
        visit_rust_node(&child, bytes, file_path, is_lib_or_mod_rs, out);
    }
}

fn rust_classify_function(
    node: &Node<'_>,
    bytes: &[u8],
    file_path: &Path,
    is_lib_or_mod_rs: bool,
    out: &mut Vec<EntryPoint>,
) {
    // Find the function name. function_item has a `name` field whose
    // value is an identifier child.
    let name_node = node.child_by_field_name("name");
    let Some(name_node) = name_node else { return };
    let Ok(name) = std::str::from_utf8(&bytes[name_node.start_byte()..name_node.end_byte()]) else {
        return;
    };
    let line = u32::try_from(node.start_position().row + 1).unwrap_or(u32::MAX);

    // Gather attributes that immediately precede this function. In
    // tree-sitter-rust the attributes are SIBLING attribute_item nodes,
    // not children of the function_item, so we walk previous siblings.
    let attrs = collect_preceding_rust_attrs(node, bytes);

    // A single function item may match multiple predicates (e.g. a
    // `#[no_mangle] pub extern "C" fn main` in `lib.rs` is both Ffi
    // and Main and LibraryExport). Emit one EntryPoint per matching
    // predicate; the BFS in X2 treats each detection as a distinct
    // reachability seed.

    // #[proc_macro], #[proc_macro_derive], #[proc_macro_attribute].
    if attrs.iter().any(|a| {
        a.starts_with("proc_macro_derive")
            || a.starts_with("proc_macro_attribute")
            || a == "proc_macro"
            || a.starts_with("proc_macro(")
    }) {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::ProcMacro,
            file_path: file_path.to_path_buf(),
            line,
        });
    }

    // #[test] / #[bench].
    if attrs.iter().any(|a| a == "test" || a == "bench") {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Test,
            file_path: file_path.to_path_buf(),
            line,
        });
    }

    // FFI: #[no_mangle] OR `extern "C"` in the function declaration.
    let function_text =
        std::str::from_utf8(&bytes[node.start_byte()..node.end_byte()]).unwrap_or("");
    let has_extern_c =
        rust_function_has_extern_c(node, bytes) || function_text.contains("extern \"C\"");
    if attrs.iter().any(|a| a == "no_mangle") || has_extern_c {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Ffi,
            file_path: file_path.to_path_buf(),
            line,
        });
    }

    // Main: `fn main` (with or without `pub`).
    if name == "main" {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Main,
            file_path: file_path.to_path_buf(),
            line,
        });
    }

    // LibraryExport: `pub fn` in lib.rs / mod.rs.
    if is_lib_or_mod_rs && rust_function_is_pub(node, bytes) {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::LibraryExport,
            file_path: file_path.to_path_buf(),
            line,
        });
    }
}

/// Collect the text of every `#[...]` attribute node that immediately
/// precedes this function_item in source order. The returned strings are
/// the attribute path/identifier (e.g. `"test"`, `"no_mangle"`,
/// `"proc_macro_derive(Foo)"`), with the leading `#[` and trailing `]`
/// stripped, and any leading `outer_attribute_item` `#[` punctuation
/// removed.
fn collect_preceding_rust_attrs(node: &Node<'_>, bytes: &[u8]) -> Vec<String> {
    let mut attrs = Vec::new();
    let mut prev = node.prev_sibling();
    while let Some(p) = prev {
        if p.kind() == "attribute_item" || p.kind() == "inner_attribute_item" {
            // The attribute_item child structure is `# [ attribute ]`;
            // pull the `attribute` child and use its text.
            let mut cursor = p.walk();
            let mut attr_text: Option<String> = None;
            for child in p.children(&mut cursor) {
                if child.kind() == "attribute"
                    && let Ok(text) =
                        std::str::from_utf8(&bytes[child.start_byte()..child.end_byte()])
                {
                    attr_text = Some(text.to_string());
                }
            }
            if let Some(t) = attr_text {
                attrs.push(t);
            }
            prev = p.prev_sibling();
        } else if p.kind().starts_with("line_comment") || p.kind().starts_with("block_comment") {
            prev = p.prev_sibling();
        } else {
            break;
        }
    }
    attrs
}

/// Return true if the function_item node has a `pub` visibility modifier.
fn rust_function_is_pub(node: &Node<'_>, bytes: &[u8]) -> bool {
    let mut cursor = node.walk();
    for child in node.children(&mut cursor) {
        if child.kind() == "visibility_modifier"
            && let Ok(text) = std::str::from_utf8(&bytes[child.start_byte()..child.end_byte()])
        {
            return text.starts_with("pub");
        }
    }
    false
}

/// Return true if the function_item has an `extern "C"` ABI declaration
/// as a function-modifier child (e.g. `pub extern "C" fn bar()`).
fn rust_function_has_extern_c(node: &Node<'_>, bytes: &[u8]) -> bool {
    let mut cursor = node.walk();
    for child in node.children(&mut cursor) {
        // tree-sitter-rust uses `function_modifiers` containing
        // `extern_modifier`; the latter's child is a `string_literal`
        // with the ABI name.
        if child.kind() != "function_modifiers" {
            continue;
        }
        let mut inner = child.walk();
        for grandchild in child.children(&mut inner) {
            if grandchild.kind() == "extern_modifier"
                && let Ok(text) =
                    std::str::from_utf8(&bytes[grandchild.start_byte()..grandchild.end_byte()])
                && text.contains("\"C\"")
            {
                return true;
            }
        }
    }
    false
}

// ---------------------------------------------------------------------------
// Python detector
// ---------------------------------------------------------------------------

/// Python entry-point detector.
///
/// Detects (per `docs/FIND_DEAD_CODE_DESIGN.md` Section 2):
/// - `if __name__ == "__main__":` blocks at module top level (Main)
/// - Top-level functions named in `__all__` (LibraryExport)
/// - Functions starting with `test_` in files matching `test_*.py` /
///   `*_test.py` or under a `tests/` directory (Test)
///
/// Framework decorators (`@click.command`, `@app.route`,
/// `@pytest.fixture`) are not yet captured — see X4 for the framework
/// pass.
#[derive(Debug, Default, Clone, Copy)]
pub struct PythonEntryDetector;

impl EntryPointDetector for PythonEntryDetector {
    fn detect(&self, source: &str, file_path: &Path) -> Vec<EntryPoint> {
        let mut entries = Vec::new();
        let Some(tree) = parse_with(source, &tree_sitter_python::LANGUAGE.into()) else {
            return entries;
        };
        let root = tree.root_node();
        let bytes = source.as_bytes();

        let is_test_file = python_is_test_file(file_path);

        // Module top-level statements.
        let mut cursor = root.walk();
        for child in root.children(&mut cursor) {
            match child.kind() {
                "if_statement" if python_is_dunder_main_block(&child, bytes) => {
                    let line = u32::try_from(child.start_position().row + 1).unwrap_or(u32::MAX);
                    entries.push(EntryPoint {
                        name: "__main__".to_string(),
                        kind: EntryPointKind::Main,
                        file_path: file_path.to_path_buf(),
                        line,
                    });
                }
                "expression_statement" => {
                    // `__all__ = [...]` is an expression_statement
                    // containing an assignment.
                    if let Some(names) = python_extract_dunder_all(&child, bytes) {
                        let line =
                            u32::try_from(child.start_position().row + 1).unwrap_or(u32::MAX);
                        for n in names {
                            entries.push(EntryPoint {
                                name: n,
                                kind: EntryPointKind::LibraryExport,
                                file_path: file_path.to_path_buf(),
                                line,
                            });
                        }
                    }
                }
                "function_definition" | "decorated_definition" => {
                    let fn_node = if child.kind() == "decorated_definition" {
                        child.child_by_field_name("definition")
                    } else {
                        Some(child)
                    };
                    if let Some(fn_node) = fn_node
                        && fn_node.kind() == "function_definition"
                        && let Some(name_node) = fn_node.child_by_field_name("name")
                        && let Ok(name) = std::str::from_utf8(
                            &bytes[name_node.start_byte()..name_node.end_byte()],
                        )
                        && is_test_file
                        && name.starts_with("test_")
                    {
                        let line =
                            u32::try_from(fn_node.start_position().row + 1).unwrap_or(u32::MAX);
                        entries.push(EntryPoint {
                            name: name.to_string(),
                            kind: EntryPointKind::Test,
                            file_path: file_path.to_path_buf(),
                            line,
                        });
                    }
                }
                _ => {}
            }
        }

        entries
    }
}

fn python_is_test_file(file_path: &Path) -> bool {
    let Some(file_name) = file_path.file_name().and_then(|s| s.to_str()) else {
        return false;
    };
    let is_py = Path::new(file_name)
        .extension()
        .is_some_and(|ext| ext.eq_ignore_ascii_case("py"));
    if !is_py {
        return false;
    }
    let stem = Path::new(file_name)
        .file_stem()
        .and_then(|s| s.to_str())
        .unwrap_or("");
    if stem.starts_with("test_") || stem.ends_with("_test") {
        return true;
    }
    // Any component named `tests` in the parent directory chain.
    file_path
        .components()
        .any(|c| c.as_os_str() == std::ffi::OsStr::new("tests"))
}

fn python_is_dunder_main_block(node: &Node<'_>, bytes: &[u8]) -> bool {
    // if condition: comparison `__name__ == "__main__"`.
    let cond = node.child_by_field_name("condition");
    let Some(cond) = cond else { return false };
    let Ok(text) = std::str::from_utf8(&bytes[cond.start_byte()..cond.end_byte()]) else {
        return false;
    };
    // Tolerate single or double quotes around `__main__`.
    let normalized = text.replace(' ', "");
    normalized.contains("__name__==\"__main__\"")
        || normalized.contains("__name__=='__main__'")
        || normalized.contains("\"__main__\"==__name__")
        || normalized.contains("'__main__'==__name__")
}

/// Extract the string literals from a top-level `__all__ = [...]`
/// assignment. Returns `None` if the statement is not such an assignment.
fn python_extract_dunder_all(node: &Node<'_>, bytes: &[u8]) -> Option<Vec<String>> {
    // expression_statement -> assignment (left, right)
    let mut cursor = node.walk();
    for child in node.children(&mut cursor) {
        if child.kind() == "assignment" {
            let left = child.child_by_field_name("left")?;
            let right = child.child_by_field_name("right")?;
            let left_text = std::str::from_utf8(&bytes[left.start_byte()..left.end_byte()]).ok()?;
            if left_text.trim() != "__all__" {
                return None;
            }
            // right is typically a `list` or `tuple` node containing
            // `string` children.
            let mut names = Vec::new();
            let mut inner = right.walk();
            for grandchild in right.children(&mut inner) {
                if grandchild.kind() != "string" {
                    continue;
                }
                // Walk the string node to find string_content child.
                let mut sc = grandchild.walk();
                let mut content_text: Option<String> = None;
                for sg in grandchild.children(&mut sc) {
                    if sg.kind() == "string_content"
                        && let Ok(t) = std::str::from_utf8(&bytes[sg.start_byte()..sg.end_byte()])
                    {
                        content_text = Some(t.to_string());
                    }
                }
                if let Some(t) = content_text {
                    names.push(t);
                } else if let Ok(raw) =
                    std::str::from_utf8(&bytes[grandchild.start_byte()..grandchild.end_byte()])
                {
                    // Fallback: strip the outer quotes from the raw
                    // string text.
                    let trimmed = raw.trim_matches(|c| c == '"' || c == '\'');
                    names.push(trimmed.to_string());
                }
            }
            return Some(names);
        }
    }
    None
}

// ---------------------------------------------------------------------------
// Go detector
// ---------------------------------------------------------------------------

/// Go entry-point detector.
///
/// Detects (per `docs/FIND_DEAD_CODE_DESIGN.md` Section 2):
/// - `func main()` in `package main` (Main)
/// - `func init()` (Init) — runs automatically at package load
/// - Functions starting with `Test`, `Benchmark`, `Example`, `Fuzz` (Test)
/// - Exported names (starting with uppercase) in library packages
///   (LibraryExport)
#[derive(Debug, Default, Clone, Copy)]
pub struct GoEntryDetector;

impl EntryPointDetector for GoEntryDetector {
    fn detect(&self, source: &str, file_path: &Path) -> Vec<EntryPoint> {
        let mut entries = Vec::new();
        let Some(tree) = parse_with(source, &tree_sitter_go::LANGUAGE.into()) else {
            return entries;
        };
        let root = tree.root_node();
        let bytes = source.as_bytes();

        // Determine the package name. `package main` enables `main` as
        // the binary entry; non-main packages are libraries whose
        // exported names are library entries.
        let package_name = go_package_name(&root, bytes).unwrap_or_default();
        let is_main_package = package_name == "main";

        // Walk top-level function_declaration / method_declaration nodes.
        let mut cursor = root.walk();
        for child in root.children(&mut cursor) {
            match child.kind() {
                "function_declaration" => {
                    if let Some(name_node) = child.child_by_field_name("name")
                        && let Ok(name) = std::str::from_utf8(
                            &bytes[name_node.start_byte()..name_node.end_byte()],
                        )
                    {
                        let line =
                            u32::try_from(child.start_position().row + 1).unwrap_or(u32::MAX);
                        go_classify(name, line, is_main_package, file_path, &mut entries);
                    }
                }
                "method_declaration" => {
                    // Methods participate in LibraryExport only — main / init
                    // / Test* are exclusively free functions.
                    if let Some(name_node) = child.child_by_field_name("name")
                        && let Ok(name) = std::str::from_utf8(
                            &bytes[name_node.start_byte()..name_node.end_byte()],
                        )
                        && !is_main_package
                        && go_is_exported(name)
                    {
                        let line =
                            u32::try_from(child.start_position().row + 1).unwrap_or(u32::MAX);
                        entries.push(EntryPoint {
                            name: name.to_string(),
                            kind: EntryPointKind::LibraryExport,
                            file_path: file_path.to_path_buf(),
                            line,
                        });
                    }
                }
                _ => {}
            }
        }

        entries
    }
}

fn go_package_name(root: &Node<'_>, bytes: &[u8]) -> Option<String> {
    let mut cursor = root.walk();
    for child in root.children(&mut cursor) {
        if child.kind() != "package_clause" {
            continue;
        }
        let mut inner = child.walk();
        for grandchild in child.children(&mut inner) {
            if grandchild.kind() == "package_identifier"
                && let Ok(text) =
                    std::str::from_utf8(&bytes[grandchild.start_byte()..grandchild.end_byte()])
            {
                return Some(text.to_string());
            }
        }
    }
    None
}

fn go_classify(
    name: &str,
    line: u32,
    is_main_package: bool,
    file_path: &Path,
    out: &mut Vec<EntryPoint>,
) {
    if name == "main" && is_main_package {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Main,
            file_path: file_path.to_path_buf(),
            line,
        });
        return;
    }
    if name == "init" {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Init,
            file_path: file_path.to_path_buf(),
            line,
        });
        return;
    }
    if name.starts_with("Test")
        || name.starts_with("Benchmark")
        || name.starts_with("Example")
        || name.starts_with("Fuzz")
    {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::Test,
            file_path: file_path.to_path_buf(),
            line,
        });
        return;
    }
    if !is_main_package && go_is_exported(name) {
        out.push(EntryPoint {
            name: name.to_string(),
            kind: EntryPointKind::LibraryExport,
            file_path: file_path.to_path_buf(),
            line,
        });
    }
}

/// Return true if `name` starts with an ASCII uppercase letter, which is
/// Go's syntactic rule for an exported (package-public) identifier.
fn go_is_exported(name: &str) -> bool {
    name.chars().next().is_some_and(|c| c.is_ascii_uppercase())
}

// ---------------------------------------------------------------------------
// Dispatch
// ---------------------------------------------------------------------------

/// Return the entry-point detector for a language identifier.
///
/// `language` is the lowercased language name as used in
/// `crate::languages` (`"rust"`, `"python"`, `"go"`). Returns `None` for
/// any language not yet covered by this wave; X4 will extend coverage to
/// JS/TS, Java, C/C++, Ruby, Scala, Kotlin, Swift, and Bash.
///
/// File-extension dispatch (`"rs"`, `"py"`, `"pyi"`, `"go"`) is also
/// accepted for caller convenience — the BFS walk in X2 carries
/// extensions, not language names, through its per-file loop.
#[must_use]
pub fn detector_for(language: &str) -> Option<Box<dyn EntryPointDetector>> {
    match language {
        "rust" | "rs" => Some(Box::new(RustEntryDetector)),
        "python" | "py" | "pyi" => Some(Box::new(PythonEntryDetector)),
        "go" => Some(Box::new(GoEntryDetector)),
        _ => None,
    }
}

// ---------------------------------------------------------------------------
// Internal helpers
// ---------------------------------------------------------------------------

/// Parse `source` with the given tree-sitter `Language`. Returns `None`
/// if the parser cannot be configured or the parse fails.
fn parse_with(source: &str, language: &tree_sitter::Language) -> Option<tree_sitter::Tree> {
    let mut parser = Parser::new();
    parser.set_language(language).ok()?;
    parser.parse(source, None)
}

// Unused-but-keep-for-X2 helpers. These ride alongside the detector
// implementations so X2 has a single import point for the BFS-time
// helpers.
//
// `query_match_lines` returns the 1-based line of every match of a
// compiled tree-sitter query against `source`. X2 will use this to
// post-process the raw RepoGraph definitions when an entry-point
// predicate fires on something that is not itself a Definition (e.g.
// the Python `if __name__ == "__main__"` block isn't a Definition —
// it's a top-level statement that anchors any function it calls).
//
// We expose it as `pub(crate)` so X2 can consume without it widening
// the public surface.

#[allow(dead_code)]
pub(crate) fn query_match_lines(
    source: &str,
    language: &tree_sitter::Language,
    query: &Query,
) -> Vec<u32> {
    let mut lines = Vec::new();
    let Some(tree) = parse_with(source, language) else {
        return lines;
    };
    let mut cursor = QueryCursor::new();
    let mut matches = cursor.matches(query, tree.root_node(), source.as_bytes());
    while let Some(m) = matches.next() {
        for cap in m.captures {
            let line = u32::try_from(cap.node.start_position().row + 1).unwrap_or(u32::MAX);
            lines.push(line);
        }
    }
    lines
}