repotoire 0.8.2

//! AST-driven extraction of [`super::predict::Evidence`] for Python
//! command-injection call sites.
//!
//! # Why a separate module
//!
//! The scorer in [`super::predict`] takes plain data
//! ([`super::predict::Evidence`]) so it can be unit-tested without an
//! AST. This module's job is to populate that struct from a
//! `tree_sitter::Node` for a Python `call` expression.
//!
//! Splitting the two halves matches Phase 2a/2b's
//! `{insecure_crypto,path_traversal}::evidence` split.
//!
//! # What this module knows about
//!
//! - Walking up from the call node to the enclosing
//!   `function_definition` (for name + parameter list) and
//!   `class_definition` (informational).
//! - Detecting the `shell=` kwarg (using the existing
//!   `python_kwarg_truthy` helper with `unknown_default=true`, the
//!   same conservative default the existing `mod.rs` API matcher uses).
//! - Classifying the first positional argument's origin into one of
//!   `Literal` / `ConfigSource` / `RequestSource` / `Parameter` /
//!   `Unknown` (only meaningful for **string-form** calls like
//!   `os.system(cmd)`).
//! - Classifying `argv[0]` and detecting "all elements are literals"
//!   for **list-form** calls (`subprocess.run([...])`).
//! - Reading the source line for `# repotoire: command-static[<reason>]`
//!   or `# repotoire: command-user-controlled[<source>]` annotations.
//!
//! # What this module deliberately does NOT do
//!
//! - Does not look for evidence in non-Python languages. Phase 2d
//!   scope is Python-only per decisions doc D4.
//! - Does not follow data flow across statements. The first-arg-origin
//!   classification looks at the syntactic form of the argument: bare
//!   identifier matched against the enclosing function's parameters,
//!   attribute chain matched against config/request lexicons, string
//!   literal, or unknown. Documented v0 limitation in the decisions doc.
//! - Does not consult the graph for enclosing scope. AST walking is
//!   sufficient and avoids a detector → graph dependency.
//!
//! # Status (commit-by-commit context)
//!
//! Lands ahead of the integration commit in `mod.rs`. The
//! `#![allow(dead_code)]` below silences clippy until that wire-up
//! commit invokes `extract_python_evidence()` from
//! `build_dual_branch_python_finding`. The 23 unit tests at the
//! bottom of this module exercise the public surface end-to-end, so
//! the "never used" diagnostic is structurally misleading — every
//! function and helper is hit by at least one `#[test]`. The allow is
//! removed in the next commit.
#![allow(dead_code)]

use super::predict::{
    extract_command_static_reason, extract_command_user_controlled_source, matches_config_object,
    matches_request_object, Argv0Origin, Evidence, FirstArgOrigin,
};
use crate::detectors::security::ast_helpers::{
    collect_named_args, enclosing_python_function, node_text, python_function_param_names,
    python_kwarg_truthy,
};
use tree_sitter::Node;

/// Extract typed evidence from a Python command-injection call node.
///
/// `call_node` must be a `call` AST node whose function names a
/// command-execution API (e.g. `os.system`, `subprocess.run`).
/// `source` is the file's raw bytes. `lines` is the pre-split
/// source-line slice the scanner already builds; used only for
/// annotation lookup.
///
/// Never panics; missing fields produce `None`/`false`/`Unknown`/
/// defaults in the corresponding Evidence field.
pub(super) fn extract_python_evidence<'a>(
    call_node: Node<'a>,
    source: &'a [u8],
    lines: &[&str],
) -> Evidence {
    let mut ev = Evidence::default();

    // ── Enclosing function (for name + parameter list) and class. ──
    let enclosing_fn = enclosing_python_function(call_node);
    if let Some(fn_node) = enclosing_fn {
        if let Some(name_node) = fn_node.child_by_field_name("name") {
            if let Some(name) = node_text(name_node, source) {
                ev.enclosing_function = Some(name.to_string());
            }
        }
    }
    ev.enclosing_class = enclosing_python_class_name(call_node, source);

    let param_names: Vec<String> = enclosing_fn
        .map(|fn_node| python_function_param_names(fn_node, source))
        .unwrap_or_default();

    // ── Argument collection. ──
    let mut all_args: Vec<Node<'_>> = Vec::new();
    let mut positional_args: Vec<Node<'_>> = Vec::new();
    if let Some(args_node) = call_node.child_by_field_name("arguments") {
        all_args = collect_named_args(args_node);
        positional_args = all_args
            .iter()
            .filter(|a| a.kind() != "keyword_argument" && a.kind() != "comment")
            .copied()
            .collect();
    }

    // ── shell=True kwarg detection (matches mod.rs:1204 semantics). ──
    //
    // `unknown_default=true` means: if `shell=` is present but its value
    // is a non-literal expression (variable, function call, ...), treat
    // it as truthy. This is the conservative default the existing API
    // matcher uses for `shell=` and matches the predictor's intent: a
    // non-literal `shell=` value should not silently be treated as
    // False, since at runtime it might be True.
    ev.kw_shell_true = python_kwarg_truthy(&all_args, "shell", source, true);

    // ── First positional arg classification. ──
    //
    // The classification has two modes depending on whether the first
    // positional arg is a list/tuple (list-form call like
    // `subprocess.run([...])`) or anything else (string-form call like
    // `os.system(cmd)`).
    if let Some(first) = positional_args.first() {
        match first.kind() {
            "list" | "tuple" => {
                // List-form: populate argv0_origin and
                // argv_list_all_literals; leave first_arg_origin = None
                // (or Unknown) because the predictor handles the
                // list-form path through argv0_origin instead.
                let (argv0, all_literals) = classify_list_form_args(*first, source, &param_names);
                ev.argv0_origin = Some(argv0);
                ev.argv_list_all_literals = all_literals;
            }
            _ => {
                // String-form: classify the first arg's origin.
                ev.first_arg_origin = Some(classify_first_arg_origin(*first, source, &param_names));
            }
        }
    }

    // ── Source-line annotations. ──
    let line_idx = call_node.start_position().row;
    if let Some(line) = lines.get(line_idx) {
        ev.command_static_annotation = extract_command_static_reason(line);
        ev.command_user_controlled_annotation = extract_command_user_controlled_source(line);
    }

    ev
}

// ──────────────────────────────────────────────────────────���──────────────────
// First-arg classification (string-form calls)
// ─────────────────────────────────────────────────────────────────────────────

/// Classify a Python expression node into a [`FirstArgOrigin`].
///
/// Decisions, in priority order (mirrors `path_traversal/evidence.rs`):
///
/// 1. String literal (no f-string interpolation) → `Literal`.
/// 2. Bare identifier matching an enclosing-function parameter →
///    `Parameter { name }`.
/// 3. Attribute / call / subscript whose textual chain matches the
///    request lexicon → `RequestSource`.
/// 4. ... matching the config lexicon → `ConfigSource`.
/// 5. Anything else → `Unknown`.
fn classify_first_arg_origin(
    node: Node<'_>,
    source: &[u8],
    param_names: &[String],
) -> FirstArgOrigin {
    match node.kind() {
        "string" => {
            // Pure string literal unless it contains an interpolation.
            let mut cursor = node.walk();
            for child in node.children(&mut cursor) {
                if child.kind() == "interpolation" {
                    return FirstArgOrigin::Unknown;
                }
            }
            FirstArgOrigin::Literal
        }
        "concatenated_string" => {
            // `"foo" "bar"` adjacent literal concat: Literal if all are
            // literals, Unknown otherwise.
            let mut cursor = node.walk();
            for child in node.children(&mut cursor) {
                if classify_first_arg_origin(child, source, param_names) != FirstArgOrigin::Literal
                {
                    return FirstArgOrigin::Unknown;
                }
            }
            FirstArgOrigin::Literal
        }
        "identifier" => {
            let Some(name) = node_text(node, source) else {
                return FirstArgOrigin::Unknown;
            };
            if param_names.iter().any(|p| p == name) {
                FirstArgOrigin::Parameter {
                    name: name.to_string(),
                }
            } else {
                FirstArgOrigin::Unknown
            }
        }
        "attribute" | "call" | "subscript" => {
            let Some(text) = node_text(node, source) else {
                return FirstArgOrigin::Unknown;
            };
            if matches_request_object(text) {
                FirstArgOrigin::RequestSource
            } else if matches_config_object(text) {
                FirstArgOrigin::ConfigSource
            } else {
                FirstArgOrigin::Unknown
            }
        }
        "parenthesized_expression" => {
            for i in 0..node.named_child_count() {
                if let Some(c) = node.named_child(i) {
                    return classify_first_arg_origin(c, source, param_names);
                }
            }
            FirstArgOrigin::Unknown
        }
        _ => FirstArgOrigin::Unknown,
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// List-form classification
// ─────────────────────────────────────────────────────────────────────────────

/// Classify a Python `list` or `tuple` node passed as the first
/// positional argument to a command-exec API.
///
/// Returns `(argv0_origin, all_literals)`:
///
/// - `argv0_origin`: origin of the first **element** of the list. Note
///   that the existing detector at `command_injection/mod.rs` already
///   reclassifies the canonical shell-c shape (argv[0] of `sh`/`bash`/
///   ... + argv[1] of `-c`) to `PySubprocessShellC` before this point,
///   so we don't see `["sh", "-c", x]` here — we see the post-classified
///   API. The first element here is just argv[0] of an
///   `execve`-semantics list.
/// - `all_literals`: true if every element is a string literal (with
///   no f-string interpolation). The strongest Benign signal.
fn classify_list_form_args(
    list_node: Node<'_>,
    source: &[u8],
    param_names: &[String],
) -> (Argv0Origin, bool) {
    let mut elements: Vec<Node<'_>> = Vec::new();
    let mut cursor = list_node.walk();
    for child in list_node.named_children(&mut cursor) {
        // Skip `comment` and similar non-element named children if any.
        if child.kind() == "comment" {
            continue;
        }
        elements.push(child);
    }

    let all_literals = !elements.is_empty()
        && elements
            .iter()
            .all(|el| is_static_string_literal(*el, source));

    let argv0 = elements
        .first()
        .map(|el| classify_argv0(*el, source, param_names))
        .unwrap_or(Argv0Origin::Other);

    (argv0, all_literals)
}

/// True if `node` is a string literal with no interpolation. Mirrors
/// the conservative posture in `classify_first_arg_origin`.
///
/// `_source` is accepted for signature uniformity with the broader
/// "give me a node, give me the bytes" extractor convention used by
/// the other helpers in this module; the recursion itself only needs
/// node kinds, not bytes. clippy's `only_used_in_recursion` flagged
/// this; underscore-prefix preserves the calling convention without
/// suppressing the lint globally.
fn is_static_string_literal(node: Node<'_>, _source: &[u8]) -> bool {
    match node.kind() {
        "string" => {
            let mut cursor = node.walk();
            for child in node.children(&mut cursor) {
                if child.kind() == "interpolation" {
                    return false;
                }
            }
            true
        }
        "concatenated_string" => {
            let mut cursor = node.walk();
            for child in node.children(&mut cursor) {
                if !is_static_string_literal(child, _source) {
                    return false;
                }
            }
            true
        }
        _ => false,
    }
}

fn classify_argv0(node: Node<'_>, source: &[u8], param_names: &[String]) -> Argv0Origin {
    match node.kind() {
        "string" | "concatenated_string" => {
            if is_static_string_literal(node, source) {
                Argv0Origin::Literal
            } else {
                Argv0Origin::Other
            }
        }
        "identifier" => {
            let Some(name) = node_text(node, source) else {
                return Argv0Origin::Other;
            };
            if param_names.iter().any(|p| p == name) {
                Argv0Origin::Parameter {
                    name: name.to_string(),
                }
            } else {
                Argv0Origin::Other
            }
        }
        _ => Argv0Origin::Other,
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Enclosing-scope helper
// ─────────────────────────────────────────────────────────────────────────────

/// Walk up from `node` to the nearest enclosing `class_definition` and
/// return its name. Returns `None` at module level.
///
/// Same shape as Phase 2a/2b's `enclosing_python_class_name`. Hoisting
/// to `ast_helpers.rs` is the rule-of-three trigger; deferred to a
/// follow-up cleanup since the three copies (2a/2b/2d) are 8 lines
/// each and consolidation would force a forward-coupling parameter
/// (`source: &[u8]`) into a helper that previously didn't need it.
fn enclosing_python_class_name<'a>(node: Node<'a>, source: &'a [u8]) -> Option<String> {
    let mut cur = node.parent()?;
    loop {
        if cur.kind() == "class_definition" {
            let name = cur.child_by_field_name("name")?;
            return node_text(name, source).map(str::to_string);
        }
        if cur.kind() == "module" {
            return None;
        }
        cur = cur.parent()?;
    }
}

// ─────────────────────────────────────────────────────────────────────────────
// Tests
// ─────────────────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;
    use crate::detectors::ast_fingerprint::parse_root_ext;
    use crate::parsers::lightweight::Language;

    /// Parse `source` as Python and find the first `call` node whose
    /// function chain ends in the named attribute or identifier.
    fn first_call_with_attr<'tree>(
        tree: &'tree tree_sitter::Tree,
        source: &[u8],
        attr_name: &str,
    ) -> tree_sitter::Node<'tree> {
        fn walk<'a>(
            node: tree_sitter::Node<'a>,
            source: &[u8],
            attr_name: &str,
        ) -> Option<tree_sitter::Node<'a>> {
            if node.kind() == "call" {
                if let Some(func) = node.child_by_field_name("function") {
                    if func.kind() == "attribute" {
                        if let Some(attr) = func.child_by_field_name("attribute") {
                            if node_text(attr, source) == Some(attr_name) {
                                return Some(node);
                            }
                        }
                    } else if func.kind() == "identifier"
                        && node_text(func, source) == Some(attr_name)
                    {
                        return Some(node);
                    }
                }
            }
            let mut cursor = node.walk();
            for child in node.children(&mut cursor) {
                if let Some(found) = walk(child, source, attr_name) {
                    return Some(found);
                }
            }
            None
        }
        walk(tree.root_node(), source, attr_name)
            .unwrap_or_else(|| panic!("no call ending in `.{attr_name}` in source"))
    }

    fn extract(source: &str, attr: &str) -> Evidence {
        let bytes = source.as_bytes();
        let tree = parse_root_ext(source, Language::Python, "py").expect("parse python");
        let lines: Vec<&str> = source.lines().collect();
        let call = first_call_with_attr(&tree, bytes, attr);
        extract_python_evidence(call, bytes, &lines)
    }

    // ── Enclosing scope ──

    #[test]
    fn extracts_enclosing_class_and_function() {
        let src = "
import subprocess
class CmdRunner:
    def run(self, name):
        return subprocess.run(['echo', name])
";
        let ev = extract(src, "run");
        assert_eq!(ev.enclosing_class.as_deref(), Some("CmdRunner"));
        assert_eq!(ev.enclosing_function.as_deref(), Some("run"));
    }

    #[test]
    fn module_level_call_has_no_enclosing_function() {
        let src = "import os\nos.system('date')\n";
        let ev = extract(src, "system");
        assert!(ev.enclosing_function.is_none());
        assert!(ev.enclosing_class.is_none());
    }

    // ── kw_shell_true ──

    #[test]
    fn shell_true_literal_detected() {
        let src = "
import subprocess
def run(cmd):
    return subprocess.run(cmd, shell=True)
";
        let ev = extract(src, "run");
        assert!(ev.kw_shell_true);
    }

    #[test]
    fn shell_false_literal_not_detected() {
        let src = "
import subprocess
def run(cmd):
    return subprocess.run(['echo', cmd], shell=False)
";
        let ev = extract(src, "run");
        assert!(!ev.kw_shell_true);
    }

    #[test]
    fn shell_absent_kwarg_not_detected() {
        let src = "
import subprocess
def run(cmd):
    return subprocess.run(['echo', cmd])
";
        let ev = extract(src, "run");
        assert!(!ev.kw_shell_true);
    }

    #[test]
    fn shell_non_literal_treated_as_truthy_conservative() {
        // `shell=enable_shell` where `enable_shell` is a variable —
        // conservative posture is to treat as truthy, mirroring
        // `mod.rs:1204` semantics (`unknown_default=true`).
        let src = "
import subprocess
def run(cmd, enable_shell):
    return subprocess.run(cmd, shell=enable_shell)
";
        let ev = extract(src, "run");
        assert!(
            ev.kw_shell_true,
            "non-literal shell= value should be treated as truthy"
        );
    }

    // ── First-arg origin (string-form) ──

    #[test]
    fn string_form_literal_first_arg() {
        let src = "
import os
def run():
    os.system('date')
";
        let ev = extract(src, "system");
        assert_eq!(ev.first_arg_origin, Some(FirstArgOrigin::Literal));
        assert!(ev.argv0_origin.is_none(), "string-form: argv0_origin None");
    }

    #[test]
    fn string_form_parameter_first_arg() {
        let src = "
import os
def run(cmd):
    os.system(cmd)
";
        let ev = extract(src, "system");
        match ev.first_arg_origin {
            Some(FirstArgOrigin::Parameter { ref name }) if name == "cmd" => {}
            other => panic!("expected Parameter {{ cmd }}, got {other:?}"),
        }
    }

    #[test]
    fn string_form_request_source_first_arg() {
        let src = "
import os
def view(request):
    os.system(request.GET['cmd'])
";
        let ev = extract(src, "system");
        assert_eq!(ev.first_arg_origin, Some(FirstArgOrigin::RequestSource));
    }

    #[test]
    fn string_form_config_source_first_arg() {
        let src = "
import os
def run():
    os.system(os.environ.get('CMD'))
";
        let ev = extract(src, "system");
        assert_eq!(ev.first_arg_origin, Some(FirstArgOrigin::ConfigSource));
    }

    #[test]
    fn string_form_fstring_interpolation_is_unknown() {
        let src = "
import os
def run(name):
    os.system(f'echo {name}')
";
        let ev = extract(src, "system");
        assert_eq!(ev.first_arg_origin, Some(FirstArgOrigin::Unknown));
    }

    // ── List-form: argv_list_all_literals ──

    #[test]
    fn list_form_all_static_literals() {
        let src = "
import subprocess
def run():
    subprocess.run(['git', 'status', '--porcelain'])
";
        let ev = extract(src, "run");
        assert!(ev.argv_list_all_literals);
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Literal));
        assert!(ev.first_arg_origin.is_none());
    }

    #[test]
    fn list_form_with_variable_element_not_all_literals() {
        let src = "
import subprocess
def run(branch):
    subprocess.run(['git', 'checkout', branch])
";
        let ev = extract(src, "run");
        assert!(!ev.argv_list_all_literals);
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Literal));
    }

    #[test]
    fn list_form_param_argv0() {
        let src = "
import subprocess
def run(binary, arg):
    subprocess.run([binary, arg])
";
        let ev = extract(src, "run");
        assert!(!ev.argv_list_all_literals);
        match ev.argv0_origin {
            Some(Argv0Origin::Parameter { ref name }) if name == "binary" => {}
            other => panic!("expected Parameter {{ binary }}, got {other:?}"),
        }
    }

    #[test]
    fn list_form_other_argv0() {
        let src = "
import subprocess
def run():
    subprocess.run([get_binary(), 'arg'])
";
        let ev = extract(src, "run");
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Other));
        assert!(!ev.argv_list_all_literals);
    }

    #[test]
    fn list_form_tuple_works_same_as_list() {
        let src = "
import subprocess
def run():
    subprocess.run(('git', 'status'))
";
        let ev = extract(src, "run");
        assert!(ev.argv_list_all_literals);
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Literal));
    }

    #[test]
    fn list_form_fstring_argv0_is_other() {
        let src = "
import subprocess
def run(x):
    subprocess.run([f'/bin/{x}', 'arg'])
";
        let ev = extract(src, "run");
        // f-string with interpolation is not a static literal.
        assert!(!ev.argv_list_all_literals);
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Other));
    }

    // ── Annotations (kind isolation) ──

    #[test]
    fn command_static_annotation_extracted() {
        let src = "
import subprocess
def run(branch):
    subprocess.run(['git', 'checkout', branch])  # repotoire: command-static[allowlisted]
";
        let ev = extract(src, "run");
        assert_eq!(ev.command_static_annotation.as_deref(), Some("allowlisted"));
        assert!(ev.command_user_controlled_annotation.is_none());
    }

    #[test]
    fn command_user_controlled_annotation_extracted() {
        let src = "
import subprocess
def run(branch):
    subprocess.run(['git', 'checkout', branch])  # repotoire: command-user-controlled[GET]
";
        let ev = extract(src, "run");
        assert_eq!(
            ev.command_user_controlled_annotation.as_deref(),
            Some("GET")
        );
        assert!(ev.command_static_annotation.is_none());
    }

    #[test]
    fn internal_path_annotation_ignored_by_command_extractor() {
        // The 2b annotation kind must NOT match the 2d extractor.
        let src = "
import subprocess
def run(p):
    subprocess.run(['cat', p])  # repotoire: internal-path[ok]
";
        let ev = extract(src, "run");
        assert!(ev.command_static_annotation.is_none());
        assert!(ev.command_user_controlled_annotation.is_none());
    }

    #[test]
    fn no_annotation_yields_none() {
        let src = "
import subprocess
def run(p):
    subprocess.run(['cat', p])
";
        let ev = extract(src, "run");
        assert!(ev.command_static_annotation.is_none());
        assert!(ev.command_user_controlled_annotation.is_none());
    }

    // ── Canonical worked example from decisions D1 ──

    #[test]
    fn worked_example_2_shell_true_param_interpolation() {
        // `subprocess.run(f"echo {name}", shell=True)` where `name` is
        // a function parameter. Evidence should fire `kw_shell_true`
        // (the f-string with interpolation classifies first_arg_origin
        // as Unknown, which is correct — the interpolated `name` is
        // captured by the existing arg_kind = Interpolated path).
        let src = "
import subprocess
def run(name):
    subprocess.run(f'echo {name}', shell=True)
";
        let ev = extract(src, "run");
        assert!(ev.kw_shell_true);
        assert_eq!(ev.first_arg_origin, Some(FirstArgOrigin::Unknown));
        assert!(ev.argv0_origin.is_none());
    }

    #[test]
    fn worked_example_4_mixed_list_literal_argv0() {
        // `subprocess.run(["xdg-open", url])` where `url` is a param.
        // Evidence should fire `argv0_origin = Literal` (not the
        // strongest signal but enough for Benign), NOT
        // `argv_list_all_literals` (because `url` is variable).
        let src = "
import subprocess
def open_url(url):
    subprocess.run(['xdg-open', url])
";
        let ev = extract(src, "run");
        assert_eq!(ev.argv0_origin, Some(Argv0Origin::Literal));
        assert!(!ev.argv_list_all_literals);
        assert!(!ev.kw_shell_true);
    }
}