Stop shelling generated test commands and fail closed on git and coverage.

A package path is one argument. Git status uses NUL records and returns git's status. A failed coverage command leaves reports unread, coverage keeps the longest path match, and site walks no longer recurse.

src/mutator/cli.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import os
 import subprocess
 import sys
 from dataclasses import dataclass, field
@@ -13,7 +14,44 @@ from mutator.engine import mutate_file, restore_backups, scan_file
 from mutator.report import format_results, format_site_log
 from mutator.runner import CommandRunner
 
-HELP = """\
+
+def _skipped_directory_text() -> str:
+    """The directories the walker skips, named the same way in --help."""
+
+    names = {
+        ".clj-kondo",
+        ".git",
+        ".hg",
+        ".idea",
+        ".metrics",
+        ".svn",
+        ".venv",
+        "__pycache__",
+        "__tests__",
+        "build",
+        "coverage",
+        "dist",
+        "node_modules",
+        "out",
+        "spec",
+        "specs",
+        "target",
+        "test",
+        "testdata",
+        "tests",
+        "venv",
+        "vendor",
+    }
+    try:
+        crapper = ensure_crapper()
+        names = set(crapper.discover.SKIP_DIRS) | set(crapper.discover.TEST_DIRS)
+    except ImportError:
+        pass
+    ordered = sorted(names)
+    return ", ".join(ordered[:-1]) + ", and " + ordered[-1]
+
+
+HELP = f"""\
 Usage: mutator [options] [path-or-filter ...]
 
 Discover mutation sites in Clojure, Java, Go, TypeScript, Rust, and Python,
@@ -67,7 +105,7 @@ Arguments:
                     path contains this text are mutated.
 
 With no paths, source files under the project root are mutated. Directories
-named test, tests, spec, specs, vendor, node_modules, and target are skipped.
+named {_skipped_directory_text()} are skipped.
 
 The default, once a snapshot exists, reruns survivors and sites in functions
 whose text changed. Killed mutants in unchanged functions are kept.
@@ -254,26 +292,55 @@ def parse_args(argv: list[str] | None = None) -> Options:
     return options
 
 
-def _changed_files(root: Path) -> list[Path]:
+class GitStatusError(Exception):
+    """git status could not be read. `code` is git's own status."""
+
+    def __init__(self, code: int, message: str):
+        super().__init__(message)
+        self.code = code
+        self.message = message
+
+
+def _git(root: Path, *args: str) -> bytes:
     result = subprocess.run(
-        ["git", "status", "--porcelain"],
+        ["git", "-C", os.fspath(root), *args],
-        cwd=root,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
         check=False,
-        capture_output=True,
-        text=True,
     )
     if result.returncode != 0:
-        print(result.stderr.strip() or "git status failed", file=sys.stderr)
+        message = os.fsdecode(result.stderr).strip() or "git status failed"
-        return []
+        raise GitStatusError(result.returncode, message)
+    return result.stdout
+
+
+def _changed_files(root: Path) -> list[Path]:
+    """Added and modified files under `root`, as paths from the repo root.
+
+    `-z` keeps spaces and non-ASCII names intact. A missing path, including a
+    deletion, is left out. A rename is the new file only.
+    """
+
+    toplevel = Path(os.fsdecode(_git(root, "rev-parse", "--show-toplevel")).strip())
+    status = _git(
+        root,
+        "status",
+        "--porcelain",
+        "-z",
+        "--no-renames",
+        "--untracked-files=all",
+        "--",
+        ".",
+    )
+    root_resolved = root.resolve()
     found: list[Path] = []
-    for line in result.stdout.splitlines():
+    for entry in status.split(b"\0"):
-        if len(line) < 4:
+        if len(entry) <= 3:
             continue
-        path_text = line[3:].strip()
+        path = (toplevel / os.fsdecode(entry[3:])).resolve()
-        if " -> " in path_text:
+        if not path.is_file() or not path.is_relative_to(root_resolved):
-            path_text = path_text.split(" -> ", 1)[1]
+            continue
-        path_text = path_text.strip('"')
+        found.append(path)
-        found.append((root / path_text).resolve())
     return found
 
 
@@ -292,8 +359,13 @@ def _positionals(root: Path, args: list[str]) -> tuple[list[Path], list[str]]:
 
 
 def _changed_sources(root: Path, crapper) -> list[Path]:
+    skip = set(crapper.discover.SKIP_DIRS) | set(crapper.discover.TEST_DIRS)
+    root_resolved = root.resolve()
     files = []
     for path in _changed_files(root):
+        relative = path.relative_to(root_resolved)
+        if not skip.isdisjoint(relative.parts):
+            continue
         if crapper.discover.language_of(path) is None:
             continue
         if crapper.discover.is_test_file(path):
@@ -358,13 +430,32 @@ def select_files(options: Options) -> list[Path]:
     return sorted(set(_inside_root(_matching(files, filters), root)), key=lambda item: item.as_posix())
 
 
-def _prepare_coverage(options: Options, root: Path, files: list[Path]) -> None:
+def _prepare_coverage(options: Options, root: Path, files: list[Path]) -> int:
-    if options.no_coverage or options.use_existing_coverage:
+    """Run coverage generation. A failed command's code is returned as-is.
-        return
+
-    if options.scan:
+    Reports already on disk are left unread when the command fails, so a stale
-        return
+    100% report cannot score the run.
+    """
+
+    if options.no_coverage or options.use_existing_coverage or options.scan:
+        return 0
+    if options.coverage_command:
+        result = CommandRunner(verbose=options.verbose).run(options.coverage_command, root, None)
+        if result.code != 0:
+            print(
+                f"Coverage command exited {result.code}. Reports already on disk will not be read.",
+                file=sys.stderr,
+            )
+            tail = "\n".join(result.output.splitlines()[-20:])
+            if tail:
+                print(tail, file=sys.stderr)
+            return result.code
+        return 0
     crapper = ensure_crapper()
-    crapper.runners.run_coverage(root, files, options.coverage_command)
+    code = crapper.runners.run_coverage(root, files, None)
+    if isinstance(code, int) and code != 0:
+        return code
+    return 0
 
 
 def _coverage_for(options: Options, root: Path, path: Path) -> set[int] | None:
@@ -437,7 +528,7 @@ def _finish(baseline_failed: bool, survived: bool) -> int:
 
 def _mutate_files(options: Options, root: Path, files: list[Path]) -> int:
     runner = CommandRunner(verbose=options.verbose)
-    baselines: dict[tuple[str, str], tuple[bool, float, str]] = {}
+    baselines: dict[tuple[tuple[str, ...], str], tuple[bool, float, str]] = {}
     forms = []
     written: list[str] = []
     baseline_failed = False
@@ -478,13 +569,19 @@ def run(argv: list[str] | None = None) -> int:
         print(missing, file=sys.stderr)
         return 2
     root = options.project_root.resolve()
+    try:
         files = select_files(options)
+    except GitStatusError as exc:
+        print(exc.message, file=sys.stderr)
+        return exc.code
     if not files:
         print("No source files to mutate.")
         return 0
     if not options.scan:
         _restore(root)
-    _prepare_coverage(options, root, files)
+    coverage = _prepare_coverage(options, root, files)
+    if coverage != 0:
+        return coverage
     if options.scan:
         return _scan(options, root, files)
     return _mutate_files(options, root, files)

src/mutator/coverage.py

@@ -29,18 +29,32 @@ def _exact(normalized: dict, candidates: list[str]):
     return None
 
 
-def _suffix_hit(key: str, candidates: list[str], suffix_match) -> bool:
+def _parts(path: str) -> tuple[str, ...]:
-    for candidate in candidates:
+    return tuple(part for part in path.split("/") if part)
-        if suffix_match(key, candidate) or suffix_match(candidate, key):
-            return True
-    return False
 
 
-def _by_suffix(normalized: dict, candidates: list[str], suffix_match):
+def _rank(key_parts: tuple[str, ...], sources: list[tuple[str, ...]]) -> tuple[int, int] | None:
-    for key, value in normalized.items():
+    """How well one report key names one of the source paths.
-        if _suffix_hit(key, candidates, suffix_match):
+
-            return value
+    Rank `(0, extra)` is a key that ends with the source and may add a prefix.
-    return None
+    Rank `(1, missing)` is a shorter key that the source path ends with. The
+    smaller rank is the longer shared suffix. Two different keys at the same
+    rank are ambiguous.
+    """
+
+    best: tuple[int, int] | None = None
+    for source in sources:
+        if not source or not key_parts:
+            continue
+        if len(key_parts) >= len(source) and key_parts[-len(source) :] == source:
+            rank = (0, len(key_parts) - len(source))
+        elif len(source) > len(key_parts) and source[-len(key_parts) :] == key_parts:
+            rank = (1, len(source) - len(key_parts))
+        else:
+            continue
+        if best is None or rank < best:
+            best = rank
+    return best
 
 
 def _lookup(index: dict, source_path: str):
@@ -53,7 +67,26 @@ def _lookup(index: dict, source_path: str):
     found = _exact(normalized, candidates)
     if found is not None:
         return found
-    return _by_suffix(normalized, candidates, crapper.coverage.suffix_match)
+    sources: list[tuple[str, ...]] = []
+    for candidate in candidates:
+        parts = _parts(candidate)
+        if parts and parts not in sources:
+            sources.append(parts)
+    best_rank: tuple[int, int] | None = None
+    best: list[tuple[tuple[str, ...], object]] = []
+    for key, value in normalized.items():
+        key_parts = _parts(key)
+        rank = _rank(key_parts, sources)
+        if rank is None:
+            continue
+        if best_rank is None or rank < best_rank:
+            best_rank = rank
+            best = [(key_parts, value)]
+        elif rank == best_rank:
+            best.append((key_parts, value))
+    if len({parts for parts, _value in best}) != 1:
+        return None
+    return best[0][1]
 
 
 def _covered_from_pairs(lines: dict[int, tuple[int, int]] | None) -> set[int] | None:
@@ -131,8 +164,7 @@ def _jacoco_index(root: Path) -> dict[str, set[int]]:
 def _go_lines(profile, source_path: str) -> set[int] | None:
     if not profile:
         return None
-    crapper = ensure_crapper()
+    segments = _lookup(profile, source_path)
-    segments = crapper.coverage._profile_segments(profile, source_path)
     if segments is None:
         return None
     covered: set[int] = set()

src/mutator/engine.py

@@ -10,7 +10,7 @@ from mutator.functions import form_digests, form_spans, sites_in_file
 from mutator.metrics import load_history, source_key, write_results
 from mutator.model import FormResult, RunResult, Site
 from mutator.report import format_scan
-from mutator.runner import CommandRunner, test_plan
+from mutator.runner import Command, CommandRunner, display_command, test_plan
 from mutator.workers import run_mutants
 
 
@@ -180,7 +180,15 @@ def _warn_many(file_key: str, selected, mutation_warning: int) -> None:
         )
 
 
-def _remember_baseline(baselines, cache_key, runner, command: str, cwd: Path) -> None:
+def _command_key(command: Command) -> tuple[str, ...]:
+    """A hashable identity. A list cannot be a dict key."""
+
+    if isinstance(command, str):
+        return ("sh", command)
+    return ("argv", *command)
+
+
+def _remember_baseline(baselines, cache_key, runner, command: Command, cwd: Path) -> None:
     if cache_key in baselines:
         return
     baseline = runner.run(command, cwd, None)
@@ -191,8 +199,8 @@ def _remember_baseline(baselines, cache_key, runner, command: str, cwd: Path) ->
     baselines[cache_key] = (True, baseline.seconds, "")
 
 
-def _baseline_failure(file_key: str, command: str, tail: str) -> RunResult:
+def _baseline_failure(file_key: str, command: Command, tail: str) -> RunResult:
-    message = f"Baseline failed for {file_key}: {command}"
+    message = f"Baseline failed for {file_key}: {display_command(command)}"
     if tail:
         message = f"{message}\n{tail}"
     return RunResult(path=file_key, forms=[], written=[], baseline_failed=True, baseline_message=message)
@@ -215,7 +223,7 @@ def _apply_selected(
     if not selected:
         return None
     command, cwd = test_plan(root, path, language, test_command)
-    cache_key = (command, str(cwd))
+    cache_key = (_command_key(command), str(cwd))
     _remember_baseline(baselines, cache_key, runner, command, cwd)
     ok, seconds, tail = baselines[cache_key]
     if not ok:
@@ -264,7 +272,7 @@ def mutate_file(
     test_command: str | None,
     timeout_factor: float,
     mutation_warning: int,
-    baselines: dict[tuple[str, str], tuple[bool, float, str]],
+    baselines: dict[tuple[tuple[str, ...], str], tuple[bool, float, str]],
     max_workers: int | None = None,
 ) -> RunResult:
     """Mutate one file and write its namespaces into `.metrics/mutate`."""

src/mutator/runner.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import os
+import shlex
 import signal
 import subprocess
 import sys
@@ -10,6 +11,8 @@ import time
 from dataclasses import dataclass
 from pathlib import Path
 
+Command = str | list[str]
+
 
 @dataclass(frozen=True)
 class CommandResult:
@@ -59,27 +62,46 @@ def _prefer_worker_sources(cwd: Path, environment: dict) -> None:
     environment["PYTHONPATH"] = os.pathsep.join(entries)
 
 
+def display_command(command: Command) -> str:
+    """A shell-looking rendering. A list is quoted so it can be pasted."""
+
+    if isinstance(command, str):
+        return command
+    return " ".join(shlex.quote(part) for part in command)
+
+
 class CommandRunner:
     def __init__(self, verbose: bool = False):
         self.verbose = verbose
 
-    def run(self, command: str, cwd: Path, timeout: float | None) -> CommandResult:
+    def run(self, command: Command, cwd: Path, timeout: float | None) -> CommandResult:
+        """Run `command` in `cwd`.
+
+        A list is an argument vector and does not go through a shell, so a
+        path with spaces or metacharacters stays one argument. A string is a
+        command the user typed in `--test-command` or `--coverage-command`.
+        """
+
         if self.verbose:
-            print(f"+ ({cwd}) {command}", file=sys.stderr)
+            print(f"+ ({cwd}) {display_command(command)}", file=sys.stderr)
         started = time.monotonic()
         environment = os.environ.copy()
         environment["PYTHONDONTWRITEBYTECODE"] = "1"
         _prefer_worker_sources(cwd, environment)
+        try:
             process = subprocess.Popen(
                 command,
                 cwd=cwd,
-            shell=True,
+                shell=isinstance(command, str),
                 stdout=subprocess.PIPE,
                 stderr=subprocess.STDOUT,
                 start_new_session=True,
                 text=True,
                 env=environment,
             )
+        except OSError as exc:
+            seconds = time.monotonic() - started
+            return CommandResult(code=127, timed_out=False, seconds=seconds, output=str(exc))
         try:
             output, _err = process.communicate(timeout=timeout)
             code = process.returncode if process.returncode is not None else 1
@@ -112,20 +134,20 @@ def _read(path: Path) -> str:
         return ""
 
 
-def _clojure_command(directory: Path) -> str:
+def _clojure_command(directory: Path) -> list[str]:
     deps = directory / "deps.edn"
     bb = directory / "bb.edn"
     if bb.is_file() and not deps.is_file():
         text = _read(bb)
         if "spec" in text:
-            return "bb spec --tag ~no-mutate"
+            return ["bb", "spec", "--tag", "~no-mutate"]
-        return "bb test"
+        return ["bb", "test"]
     text = _read(deps)
     if ":spec" in text and "speclj" in text:
-        return "clj -M:spec --tag ~no-mutate"
+        return ["clj", "-M:spec", "--tag", "~no-mutate"]
     if ":spec" in text:
-        return "clj -M:spec"
+        return ["clj", "-M:spec"]
-    return "clj -M:test"
+    return ["clj", "-M:test"]
 
 
 def _project_interpreter(directory: Path) -> str:
@@ -142,7 +164,7 @@ def _project_interpreter(directory: Path) -> str:
     return sys.executable
 
 
-def _python_command(directory: Path) -> str:
+def _python_command(directory: Path) -> list[str]:
     interpreter = _project_interpreter(directory)
     pyproject = _read(directory / "pyproject.toml")
     if (
@@ -151,12 +173,16 @@ def _python_command(directory: Path) -> str:
         or "pytest" in pyproject
         or (directory / "setup.cfg").is_file() and "pytest" in _read(directory / "setup.cfg")
     ):
-        return f"{interpreter} -m pytest"
+        return [interpreter, "-m", "pytest"]
-    return f"{interpreter} -m unittest discover"
+    return [interpreter, "-m", "unittest", "discover"]
 
 
-def test_plan(root: Path, source: Path, language: str, override: str | None) -> tuple[str, Path]:
+def test_plan(root: Path, source: Path, language: str, override: str | None) -> tuple[Command, Path]:
-    """The test command and the directory it runs in."""
+    """The test command and the directory it runs in.
+
+    A user override stays a shell string. Every generated command is an
+    argument vector, so a package path is one argument.
+    """
 
     root = root.resolve()
     if override:
@@ -166,18 +192,18 @@ def test_plan(root: Path, source: Path, language: str, override: str | None) ->
         return _clojure_command(directory), directory
     if language == "java":
         directory = nearest(source, "pom.xml", root) or root
-        return "mvn -q test -DexcludeTags=no-mutate", directory
+        return ["mvn", "-q", "test", "-DexcludeTags=no-mutate"], directory
     if language == "go":
         directory = nearest(source, "go.mod", root) or root
         relative = source.resolve().parent.relative_to(directory.resolve())
         package = "." if relative == Path(".") else "./" + relative.as_posix()
-        return f"go test -count=1 {package}", directory
+        return ["go", "test", "-count=1", package], directory
     if language == "typescript":
         directory = nearest(source, "package.json", root) or root
-        return "npm test", directory
+        return ["npm", "test"], directory
     if language == "rust":
         directory = nearest(source, "Cargo.toml", root) or root
-        return "cargo test", directory
+        return ["cargo", "test"], directory
     if language == "python":
         directory = (
             nearest(source, "pyproject.toml", root)
@@ -186,4 +212,4 @@ def test_plan(root: Path, source: Path, language: str, override: str | None) ->
             or root
         )
         return _python_command(directory), directory
-    return "false", root
+    return ["false"], root

src/mutator/sites.py

@@ -227,12 +227,15 @@ def _clojure_node(node, data: bytes, found: list[RawSite]) -> None:
 
 
 def _visit_clojure(node, data: bytes, found: list[RawSite]) -> None:
-    _clojure_node(node, data, found)
+    # An explicit stack: a 1,200-deep form must not raise RecursionError.
-    for child in node.children:
+    stack = [node]
-        _visit_clojure(child, data, found)
+    while stack:
+        current = stack.pop()
+        _clojure_node(current, data, found)
+        stack.extend(reversed(current.children))
 
 
-def _visit(node, data: bytes, language: str, found: list[RawSite]) -> None:
+def _visit_node(node, data: bytes, language: str, found: list[RawSite]) -> None:
     parent = node.parent.type if node.parent is not None else ""
     if node.child_count == 0:
         token = _text(data, node)
@@ -257,8 +260,15 @@ def _visit(node, data: bytes, language: str, found: list[RawSite]) -> None:
     elif node.type == "bool_lit" and _text(data, node) in _BOOLEANS:
         text = _text(data, node)
         _add(found, node, text, _BOOLEANS[text], "boolean")
-    for child in node.children:
+
-        _visit(child, data, language, found)
+
+def _visit(node, data: bytes, language: str, found: list[RawSite]) -> None:
+    # An explicit stack: a 1,200-term expression must not raise RecursionError.
+    stack = [node]
+    while stack:
+        current = stack.pop()
+        _visit_node(current, data, language, found)
+        stack.extend(reversed(current.children))
 
 
 def discover_raw(source: str, language: str, path: str) -> list[RawSite]:

src/mutator/workers.py

@@ -18,6 +18,7 @@ from queue import Empty, Queue
 from threading import Lock
 
 from mutator.model import Site
+from mutator.runner import Command
 from mutator.sites import apply_site
 
 # Build output and the worker tree itself must not be shared. A link to
@@ -299,7 +300,7 @@ def _execute(
     relative: str,
     original: bytes,
     runner,
-    command: str,
+    command: Command,
     cwd: Path,
     timeout: float,
     file_key: str,
@@ -332,7 +333,7 @@ def _drain(
     relative: str,
     original: bytes,
     runner,
-    command: str,
+    command: Command,
     cwd: Path,
     timeout: float,
     file_key: str,
@@ -373,7 +374,7 @@ def _run_all(
     relative: str,
     original: bytes,
     runner,
-    command: str,
+    command: Command,
     cwd: Path,
     timeout: float,
     file_key: str,
@@ -412,7 +413,7 @@ def run_mutants(
     sites: list[Site],
     max_workers: int | None,
     runner,
-    command: str,
+    command: Command,
     cwd: Path,
     timeout: float,
     file_key: str,

tests/test_boundaries.py

@@ -6,6 +6,7 @@ alone when no input can tell the operators apart.
 """
 
 import hashlib
+import subprocess
 import sys
 import threading
 from pathlib import Path
@@ -13,6 +14,7 @@ from pathlib import Path
 import pytest
 
 from mutator.cli import (
+    GitStatusError,
     _changed_files,
     _is_source,
     _prepare_coverage,
@@ -21,7 +23,7 @@ from mutator.cli import (
     run,
     select_files,
 )
-from mutator.coverage import _candidates, _jacoco_index, _suffix_hit, covered_lines
+from mutator.coverage import _candidates, _jacoco_index, covered_lines
 from mutator.crapper_link import ensure_crapper
 from mutator.edn import loads
 from mutator.engine import (
@@ -246,7 +248,6 @@ def test_grammar_guards_and_non_head_clojure_symbols(tmp_path):
 def test_coverage_indexes_use_package_names_form_counts_and_exact_paths(tmp_path):
     assert _candidates(lambda value: value, "") == []
     assert _candidates(lambda _value: "same", "src/app.py") == ["same"]
-    assert _suffix_hit("demo/Board.java", ["Other.java"], lambda left, right: left.endswith(right)) is False
 
     report = tmp_path / "target" / "site" / "jacoco" / "jacoco.xml"
     report.parent.mkdir(parents=True)
@@ -288,7 +289,7 @@ def test_commands_follow_the_nearest_project_and_a_killed_process(tmp_path):
     both.mkdir()
     (both / "bb.edn").write_text("{:tasks {test (clojure)}}\n", encoding="utf-8")
     (both / "deps.edn").write_text("{:deps {}}\n", encoding="utf-8")
-    assert _clojure_command(both) == "clj -M:test"
+    assert _clojure_command(both) == ["clj", "-M:test"]
 
     root = tmp_path / "proj"
     sub = root / "svc"
@@ -299,28 +300,28 @@ def test_commands_follow_the_nearest_project_and_a_killed_process(tmp_path):
     clojure.write_text("(defn place [] 1)\n", encoding="utf-8")
     command, directory = command_for(root, clojure, "clojure", None)
     assert directory == sub
-    assert command == "bb test"
+    assert command == ["bb", "test"]
 
     (sub / "pom.xml").write_text("<project/>\n", encoding="utf-8")
     java = sub / "src" / "A.java"
     java.write_text("class A { int place(){ return 1; } }\n", encoding="utf-8")
     command, directory = command_for(root, java, "java", None)
     assert directory == sub
-    assert command == "mvn -q test -DexcludeTags=no-mutate"
+    assert command == ["mvn", "-q", "test", "-DexcludeTags=no-mutate"]
 
     (sub / "package.json").write_text("{}\n", encoding="utf-8")
     typescript = sub / "src" / "a.ts"
     typescript.write_text("export function place(){ return 1; }\n", encoding="utf-8")
     command, directory = command_for(root, typescript, "typescript", None)
     assert directory == sub
-    assert command == "npm test"
+    assert command == ["npm", "test"]
 
     (sub / "Cargo.toml").write_text("[package]\nname='demo'\n", encoding="utf-8")
     rust = sub / "src" / "lib.rs"
     rust.write_text("fn place() -> i32 { 1 }\n", encoding="utf-8")
     command, directory = command_for(root, rust, "rust", None)
     assert directory == sub
-    assert command == "cargo test"
+    assert command == ["cargo", "test"]
 
     (sub / "go.mod").write_text("module example.com/demo\n", encoding="utf-8")
     go = sub / "pkg" / "widget.go"
@@ -328,14 +329,14 @@ def test_commands_follow_the_nearest_project_and_a_killed_process(tmp_path):
     go.write_text("package pkg\nfunc Run() int { return 1 }\n", encoding="utf-8")
     command, directory = command_for(root, go, "go", None)
     assert directory == sub
-    assert command == "go test -count=1 ./pkg"
+    assert command == ["go", "test", "-count=1", "./pkg"]
 
     (sub / "pyproject.toml").write_text("[project]\ndependencies=['pytest']\n", encoding="utf-8")
     python = sub / "app.py"
     python.write_text("def place():\n    return 1\n", encoding="utf-8")
     command, directory = command_for(root, python, "python", None)
     assert directory == sub
-    assert command.endswith("-m pytest")
+    assert command[-2:] == ["-m", "pytest"]
 
     result = CommandRunner().run(f"{sys.executable} -c 'import time; time.sleep(30)'", tmp_path, 0.4)
     assert result.timed_out is True
@@ -398,28 +399,93 @@ def test_help_and_conflicts_use_different_streams(capsys):
     assert captured.out == ""
 
 
-def test_git_status_keeps_short_names_renames_and_the_failure_text(monkeypatch, tmp_path, capsys):
+def _init_repo(path: Path) -> None:
-    class Result:
+    subprocess.run(["git", "init"], cwd=path, check=True, capture_output=True)
-        def __init__(self, code, out="", err=""):
+    subprocess.run(["git", "config", "user.email", "[email protected]"], cwd=path, check=True)
-            self.returncode = code
+    subprocess.run(["git", "config", "user.name", "Dev"], cwd=path, check=True)
-            self.stdout = out
-            self.stderr = err
 
-    def fail(*args, **kwargs):
+
-        return Result(1, err="boom\n")
+def _commit(path: Path, message: str) -> None:
+    subprocess.run(["git", "add", "-A"], cwd=path, check=True, capture_output=True)
+    subprocess.run(["git", "commit", "-m", message], cwd=path, check=True, capture_output=True)
+
+
+def test_git_status_keeps_real_names_and_fails_closed(tmp_path, capsys):
+    outside = tmp_path / "outside"
+    outside.mkdir()
+    code = run(["--changed", "--root", str(outside)])
+    captured = capsys.readouterr()
+    assert code == 128
+    assert "not a git repository" in captured.err
+    assert "No source files" not in captured.out
+
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _init_repo(repo)
+    source = "def place(x):\n    return x > 0\n"
+    keep = repo / "src" / "café.py"
+    keep.parent.mkdir()
+    keep.write_text(source, encoding="utf-8")
+    nested = repo / "fresh" / "nested" / "new.py"
+    nested.parent.mkdir(parents=True)
+    nested.write_text(source, encoding="utf-8")
+    spaced = repo / "src" / "my file.py"
+    spaced.write_text(source, encoding="utf-8")
+    gone = repo / "gone.py"
+    gone.write_text(source, encoding="utf-8")
+    old = repo / "src" / "old.py"
+    old.write_text(source, encoding="utf-8")
+    _commit(repo, "base")
+    keep.write_text(source + "\n", encoding="utf-8")
+    nested.write_text(source + "\n", encoding="utf-8")
+    spaced.write_text(source + "\n", encoding="utf-8")
+    gone.unlink()
+    subprocess.run(
+        ["git", "mv", "src/old.py", "src/new.py"], cwd=repo, check=True, capture_output=True
+    )
+    skipped = repo / "target" / "skip.py"
+    skipped.parent.mkdir()
+    skipped.write_text(source, encoding="utf-8")
+
+    found = set(_changed_files(repo))
+    assert keep.resolve() in found
+    assert nested.resolve() in found
+    assert spaced.resolve() in found
+    assert (repo / "src" / "new.py").resolve() in found
+    assert gone.resolve() not in found
+    assert old.resolve() not in found
+
+    chosen = select_files(parse_args(["--root", str(repo), "--changed"]))
+    assert skipped.resolve() not in chosen
+    assert keep.resolve() in chosen
+    code = run(["--scan", "--no-coverage", "--root", str(repo), "--changed"])
+    assert code == 0
+    assert "café.py" in capsys.readouterr().out
+
+
+def test_changed_from_a_subdirectory_stays_inside_it(tmp_path):
+    repo = tmp_path / "repo"
+    sub = repo / "sub"
+    (sub / "src").mkdir(parents=True)
+    (repo / "src").mkdir()
+    _init_repo(repo)
+    below = sub / "src" / "below.py"
+    above = repo / "src" / "above.py"
+    body = "def place(x):\n    return x > 0\n"
+    below.write_text(body, encoding="utf-8")
+    above.write_text(body, encoding="utf-8")
+    chosen = select_files(parse_args(["--root", str(sub), "--changed"]))
+    assert chosen == [below.resolve()]
+
+
+def test_an_empty_git_error_uses_the_fallback_text(monkeypatch, tmp_path):
+    def fail(args, **kwargs):
+        return subprocess.CompletedProcess(args, 128, b"", b"   \n")
 
     monkeypatch.setattr("mutator.cli.subprocess.run", fail)
-    assert _changed_files(tmp_path) == []
+    with pytest.raises(GitStatusError, match="git status failed") as caught:
-    assert capsys.readouterr().err.strip() == "boom"
+        _changed_files(tmp_path)
-
+    assert caught.value.code == 128
-    def status(*args, **kwargs):
-        return Result(0, out="?? a\nR  src/old.py -> src/new.py\n")
-
-    monkeypatch.setattr("mutator.cli.subprocess.run", status)
-    found = _changed_files(tmp_path)
-    assert (tmp_path / "a").resolve() in found
-    assert (tmp_path / "src" / "new.py").resolve() in found
-    assert (tmp_path / "src" / "old.py").resolve() not in found
 
 
 def test_selection_joins_the_root_and_skips_unknown_files(tmp_path, capsys):

tests/test_cli.py

@@ -42,6 +42,8 @@ def test_help_and_conflicting_flags():
     help_options = parse_args(["--help"])
     assert help_options.exit_code == 0
     assert "uml-viewer" in help_options.message
+    assert "__pycache__" in help_options.message
+    assert "testdata" in help_options.message
     conflict = parse_args(["--mutate-all", "--since-last-run", "src/demo.py"])
     assert conflict.exit_code == 1
     assert run(["--help"]) == 0
@@ -92,3 +94,28 @@ def test_real_python_mutants_update_the_snapshot(tmp_path, capsys):
     assert ":namespace \"demo\"" in text
     assert ":id \"defn/add\"" in text
     assert path.read_text(encoding="utf-8").startswith("def add")
+
+
+def test_a_failed_coverage_command_does_not_read_the_report(tmp_path, capsys):
+    path = tmp_path / "src" / "app.py"
+    path.parent.mkdir()
+    path.write_text("def place(x):\n    return x > 0\n", encoding="utf-8")
+    report = tmp_path / "target" / "coverage" / "python" / "lcov.info"
+    report.parent.mkdir(parents=True)
+    report.write_text("SF:src/app.py\nDA:1,1\nDA:2,1\nend_of_record\n", encoding="utf-8")
+    code = run(
+        [
+            "--root",
+            str(tmp_path),
+            "--coverage-command",
+            "exit 3",
+            "--test-command",
+            "false",
+            "src/app.py",
+        ]
+    )
+    captured = capsys.readouterr()
+    assert code == 3
+    assert "will not be read" in captured.err
+    assert "Baseline failed" not in captured.err
+    assert path.read_text(encoding="utf-8").startswith("def place")

tests/test_coverage.py

@@ -55,3 +55,44 @@ def test_lcov_marks_executed_lines(tmp_path):
     source.write_text("def place():\n    return 1\n", encoding="utf-8")
     assert covered_lines(tmp_path, source, "python") == {1}
     assert covered_lines(tmp_path, source, "clojure") == {1}
+
+
+def test_lcov_does_not_let_a_shorter_path_steal_hits(tmp_path):
+    report = tmp_path / "target" / "coverage" / "python" / "lcov.info"
+    report.parent.mkdir(parents=True)
+    report.write_text(
+        "SF:b/app.py\nDA:2,1\nend_of_record\n"
+        "SF:a/b/app.py\nDA:1,1\nend_of_record\n",
+        encoding="utf-8",
+    )
+    short = tmp_path / "b" / "app.py"
+    long = tmp_path / "a" / "b" / "app.py"
+    short.parent.mkdir(parents=True)
+    long.parent.mkdir(parents=True)
+    short.write_text("def place():\n    return 1\n", encoding="utf-8")
+    long.write_text("def place():\n    return 1\n", encoding="utf-8")
+    assert covered_lines(tmp_path, long, "python") == {1}
+    assert covered_lines(tmp_path, short, "python") == {2}
+
+
+def test_go_profile_does_not_cross_files(tmp_path):
+    profile = tmp_path / "target" / "coverage" / "go" / "coverage.out"
+    profile.parent.mkdir(parents=True)
+    profile.write_text(
+        "mode: set\n"
+        "b/widget.go:2.1,2.2 1 1\n"
+        "a/b/widget.go:3.1,3.2 1 0\n",
+        encoding="utf-8",
+    )
+    short = tmp_path / "b" / "widget.go"
+    long = tmp_path / "a" / "b" / "widget.go"
+    short.parent.mkdir(parents=True)
+    long.parent.mkdir(parents=True)
+    short.write_text("package b\nfunc Run() {}\n", encoding="utf-8")
+    long.write_text("package b\nfunc Run() {}\n", encoding="utf-8")
+    assert covered_lines(tmp_path, short, "go") == {2}
+    assert covered_lines(tmp_path, long, "go") == set()
+
+    profile.write_text("mode: set\na/b/widget.go:2.1,2.2 1 1\n", encoding="utf-8")
+    assert covered_lines(tmp_path, short, "go") is None
+    assert covered_lines(tmp_path, long, "go") == {2}

tests/test_runner.py

@@ -1,6 +1,7 @@
 import sys
 
 from mutator.runner import CommandRunner, _clojure_command, _python_command, nearest
+from mutator.runner import test_plan as plan_command
 
 
 def test_nearest_walks_up_to_the_marker(tmp_path):
@@ -19,29 +20,29 @@ def test_clojure_commands(tmp_path):
     bb = tmp_path / "bb-spec"
     bb.mkdir()
     (bb / "bb.edn").write_text('{:tasks {spec "spec"}}', encoding="utf-8")
-    assert _clojure_command(bb) == "bb spec --tag ~no-mutate"
+    assert _clojure_command(bb) == ["bb", "spec", "--tag", "~no-mutate"]
 
     bb_test = tmp_path / "bb-test"
     bb_test.mkdir()
     (bb_test / "bb.edn").write_text("{:tasks {test (clojure \"-M:test\")}}", encoding="utf-8")
-    assert _clojure_command(bb_test) == "bb test"
+    assert _clojure_command(bb_test) == ["bb", "test"]
 
     spec = tmp_path / "spec"
     spec.mkdir()
     (spec / "deps.edn").write_text(
         "{:aliases {:spec {:extra-deps {speclj/speclj {}}}}}", encoding="utf-8"
     )
-    assert _clojure_command(spec) == "clj -M:spec --tag ~no-mutate"
+    assert _clojure_command(spec) == ["clj", "-M:spec", "--tag", "~no-mutate"]
 
     spec_only = tmp_path / "spec-only"
     spec_only.mkdir()
     (spec_only / "deps.edn").write_text("{:aliases {:spec {}}}", encoding="utf-8")
-    assert _clojure_command(spec_only) == "clj -M:spec"
+    assert _clojure_command(spec_only) == ["clj", "-M:spec"]
 
     plain = tmp_path / "plain"
     plain.mkdir()
     (plain / "deps.edn").write_text("{:deps {}}", encoding="utf-8")
-    assert _clojure_command(plain) == "clj -M:test"
+    assert _clojure_command(plain) == ["clj", "-M:test"]
 
 
 def _executable(path):
@@ -55,15 +56,20 @@ def test_python_command_uses_an_absolute_project_interpreter(tmp_path):
     dot = project / ".venv" / "bin" / "python"
     _executable(dot)
     (project / "pytest.ini").write_text("[pytest]\n", encoding="utf-8")
-    assert _python_command(project) == f"{dot.absolute()} -m pytest"
+    assert _python_command(project) == [str(dot.absolute()), "-m", "pytest"]
 
     other = tmp_path / "other"
     plain = other / "venv" / "bin" / "python"
     _executable(plain)
-    assert _python_command(other) == f"{plain.absolute()} -m unittest discover"
+    assert _python_command(other) == [str(plain.absolute()), "-m", "unittest", "discover"]
 
     _executable(other / ".venv" / "bin" / "python")
-    assert _python_command(other) == f"{(other / '.venv' / 'bin' / 'python').absolute()} -m unittest discover"
+    assert _python_command(other) == [
+        str((other / ".venv" / "bin" / "python").absolute()),
+        "-m",
+        "unittest",
+        "discover",
+    ]
 
 
 def test_python_command_keeps_the_virtualenv_symlink(tmp_path):
@@ -74,7 +80,7 @@ def test_python_command_keeps_the_virtualenv_symlink(tmp_path):
     link.parent.mkdir(parents=True)
     link.symlink_to(target)
     (project / "pytest.ini").write_text("[pytest]\n", encoding="utf-8")
-    assert _python_command(project) == f"{link.absolute()} -m pytest"
+    assert _python_command(project) == [str(link.absolute()), "-m", "pytest"]
 
 
 def test_python_command_falls_back_when_the_virtualenv_cannot_run(tmp_path):
@@ -82,37 +88,61 @@ def test_python_command_falls_back_when_the_virtualenv_cannot_run(tmp_path):
     binary = project / ".venv" / "bin" / "python"
     binary.parent.mkdir(parents=True)
     binary.write_text("", encoding="utf-8")
-    assert _python_command(project) == f"{sys.executable} -m unittest discover"
+    assert _python_command(project) == [sys.executable, "-m", "unittest", "discover"]
 
     missing = tmp_path / "missing"
     missing.mkdir()
-    assert _python_command(missing) == f"{sys.executable} -m unittest discover"
+    assert _python_command(missing) == [sys.executable, "-m", "unittest", "discover"]
 
 
 def test_python_commands(tmp_path):
     pytest_dir = tmp_path / "py"
     pytest_dir.mkdir()
     (pytest_dir / "pytest.ini").write_text("[pytest]\n", encoding="utf-8")
-    assert _python_command(pytest_dir).endswith("-m pytest")
+    assert _python_command(pytest_dir)[-2:] == ["-m", "pytest"]
 
     unit = tmp_path / "unit"
     unit.mkdir()
-    assert _python_command(unit).endswith("-m unittest discover")
+    assert _python_command(unit)[-3:] == ["-m", "unittest", "discover"]
 
     cfg = tmp_path / "cfg"
     cfg.mkdir()
     (cfg / "setup.cfg").write_text("[tool:pytest]\n", encoding="utf-8")
-    assert _python_command(cfg).endswith("-m pytest")
+    assert _python_command(cfg)[-2:] == ["-m", "pytest"]
 
     project = tmp_path / "proj"
     project.mkdir()
     (project / "pyproject.toml").write_text("[project]\ndependencies=['pytest']\n", encoding="utf-8")
-    assert _python_command(project).endswith("-m pytest")
+    assert _python_command(project)[-2:] == ["-m", "pytest"]
 
     conf = tmp_path / "conf"
     conf.mkdir()
     (conf / "conftest.py").write_text("", encoding="utf-8")
-    assert _python_command(conf).endswith("-m pytest")
+    assert _python_command(conf)[-2:] == ["-m", "pytest"]
+
+
+def test_a_go_package_path_is_one_argument(tmp_path):
+    root = tmp_path / "proj"
+    package = root / "a;echo no"
+    package.mkdir(parents=True)
+    (root / "go.mod").write_text("module example.com/demo\n", encoding="utf-8")
+    source = package / "widget.go"
+    source.write_text("package main\nfunc Run() int { return 1 }\n", encoding="utf-8")
+    command, directory = plan_command(root, source, "go", None)
+    assert command == ["go", "test", "-count=1", "./a;echo no"]
+    assert directory == root
+    echoed = CommandRunner().run(["/bin/echo", command[-1]], directory, None)
+    assert echoed.code == 0
+    assert echoed.output.strip() == "./a;echo no"
+
+    spaced = root / "my dir" / "widget.go"
+    spaced.parent.mkdir()
+    spaced.write_text("package main\nfunc Run() int { return 1 }\n", encoding="utf-8")
+    command, _directory = plan_command(root, spaced, "go", None)
+    assert command == ["go", "test", "-count=1", "./my dir"]
+    override, override_dir = plan_command(root, source, "go", "go test ./...")
+    assert override == "go test ./..."
+    assert override_dir == root.resolve()
 
 
 def test_a_worker_overlay_is_imported_ahead_of_the_environment(tmp_path):

tests/test_sites.py

@@ -1,6 +1,7 @@
 from pathlib import Path
 
 from mutator.functions import form_id, is_private, project_functions, sites_in_file
+from mutator.sites import discover_raw
 
 
 def _sites(tmp_path: Path, relative: str, source: str):
@@ -194,3 +195,11 @@ export function mount(app) {
 def test_a_quoted_nullish_operator_is_not_a_site(tmp_path):
     source = 'export function place() {\n  return "a ?? b?.c";\n}\n'
     assert _sites(tmp_path, "src/demo/app.ts", source) == []
+
+
+def test_a_deep_expression_is_walked(tmp_path):
+    expr = " + ".join(["1"] * 1200)
+    python = f"def place():\n    return {expr}\n"
+    assert discover_raw(python, "python", "app.py")
+    go = f"package p\nfunc place() int {{ return {expr} }}\n"
+    assert discover_raw(go, "go", "app.go")