tdd-cli 0.3.0__tar.gz → 0.4.1__tar.gz

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/CHANGELOG.md

@@ -4,6 +4,51 @@ All notable changes to this project are documented here.
 The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and the project adheres to [Semantic Versioning](https://semver.org/).
 
+## [0.4.1] - 2026-08-16
+
+### Changed
+
+- `collect()` runs one invocation per declared suite instead of one per test
+  file, falling back to the per-file loop for anything a batch did not account
+  for. Per-file collection was **77% of a real `run start`** — 313 subprocesses
+  costing 402s, against 117s to actually run every test — because cost scaled
+  with file count at a ~1.08s floor per invocation (the environment manager
+  resolving plus the runner booting). Measured 38.8x faster on a 60-file
+  project, with an identical collected set. R10.3's guarantee is unchanged: a
+  file that fails to collect is still attributed to itself and cannot destroy
+  the set, and a file the batch never reports is still collected individually,
+  so the set can only match or improve on the old one (#27).
+
+## [0.4.0] - 2026-08-16
+
+### Added
+
+- `baseline_captured` reports `run_s` and `collect_s` alongside `elapsed_s`. The
+  suite run and the per-file collection have unrelated cost models — one scales
+  with tests, the other with files — so a single total could not say which was
+  slow, and answering that meant measuring projects by hand outside the tool.
+- `TDD_TIMING=1` emits a `command_timing` line per subprocess on stderr
+  (`label`, `command`, `cwd`, `duration_ms`, `exit_code`), covering every
+  subprocess the tool spawns: suite runs, per-file collection, lint/typecheck
+  gates, doctor probes and artifact hooks. Off by default — the per-file loop
+  would otherwise emit one line per test file on every invocation. `label` is
+  one of `suite`, `collect`, `gate`, `doctor`; an unlabelled row comes from a
+  third-party adapter, since every built-in call site names itself (R8.4).
+
+### Fixed
+
+- `tdd doctor` no longer emits a blocker it cannot explain. Every failing check
+  now carries a `detail` naming what to fix, enforced by the checklist recorder
+  so a check added later inherits the guarantee. Previously `worktree clean`
+  failed with `detail: ""`, leaving an agent with `resolve_blocker` and nothing
+  to resolve — it re-ran doctor and read the identical output.
+- `worktree clean` is scoped to dirt a run would actually read: a declared
+  project root, a declared artifact path, or `tdd.toml`. Build residue is
+  excluded via `config.is_ignored`, so doctor's own `uv run` / `vitest list`
+  probes (`.venv`, `node_modules`, caches) can no longer be what makes doctor
+  fail. Unrelated dirt is reported in the passing check's `detail` rather than
+  blocking the run.
+
 ## [0.3.0] - 2026-08-10
 
 ### Added

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: tdd-cli
-Version: 0.3.0
+Version: 0.4.1
 Summary: Ledger-backed TDD process controller for autonomous coding agents
 Project-URL: Homepage, https://github.com/geuben/tdd-cli
 Project-URL: Repository, https://github.com/geuben/tdd-cli

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/pyproject.toml

@@ -47,7 +47,7 @@ packages = ["src/tddcli"]
 include = ["src", "tests", "examples", "README.md", "LICENSE", "CHANGELOG.md", "SECURITY.md"]
 
 [dependency-groups]
-dev = ["pytest>=8.0", "pytest-json-report>=1.5", "ruff>=0.5"]
+dev = ["pytest>=8.0", "pytest-json-report>=1.5", "ruff>=0.5", "zizmor>=1.29"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests"]

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/src/tddcli/__init__.py

@@ -3,4 +3,4 @@
 State is derived from observed test execution, never asserted by the caller.
 """
 
-__version__ = "0.3.0"
+__version__ = "0.4.1"

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/src/tddcli/adapters/base.py

@@ -4,11 +4,13 @@ from __future__ import annotations
 
 import os
 import subprocess
+import time
 from collections import Counter
 from dataclasses import dataclass, field
 from pathlib import Path
 
 from .. import leases
+from ..envelope import heartbeat
 
 NOT_FOUND = "not_found"
 NOT_COLLECTED = "not_collected"
@@ -78,10 +80,26 @@ def clip_failure(text: str, limit: int = 1500) -> str:
     return f"{text[:head]}\n… [clipped] …\n{text[-tail:]}"
 
 
+#: Opt-in per-command timing. Off by default: the per-file collect loop would emit
+#: one line per test file on every invocation, drowning the heartbeats that exist
+#: to make a slow baseline legible.
+TIMING_ENV = "TDD_TIMING"
+
+
 def run_command(
     command: str, cwd: Path, timeout: int = 1800,
     extra_env: dict[str, str] | None = None,
+    label: str | None = None,
 ) -> tuple[int, str, str]:
+    """Every subprocess the tool spawns passes through here, which makes it the one
+    place worth timing: suite runs, per-file collection, lint and typecheck gates,
+    doctor probes, artifact hooks.
+
+    `label` is what makes the rows groupable. This function sees a command string
+    and a cwd — not which project or phase asked for it — so an unlabelled timing
+    is readable by a human and useless to a query.
+    """
+    started = time.monotonic()
     proc = subprocess.run(
         command,
         shell=True,
@@ -91,6 +109,15 @@ def run_command(
         timeout=timeout,
         env=None if extra_env is None else {**os.environ, **extra_env},
     )
+    if os.environ.get(TIMING_ENV):
+        heartbeat(
+            event="command_timing",
+            label=label,
+            command=command,
+            cwd=str(cwd),
+            duration_ms=int((time.monotonic() - started) * 1000),
+            exit_code=proc.returncode,
+        )
     return proc.returncode, proc.stdout, proc.stderr
 
 
@@ -130,6 +157,7 @@ class Adapter:
                 command.replace("{workers}", str(workers)),
                 self.root,
                 extra_env={"TDD_WORKERS": str(workers), **(extra_env or {})},
+                label="suite",
             )
 
     def _test_cmd(self) -> str:
@@ -161,6 +189,48 @@ class Adapter:
         return "a body that fails loudly, never working logic"
 
     def collect(self) -> Collection:
+        """Enumerate the project's tests: one invocation per declared suite, with
+        the per-file loop kept for what that cannot account for (issue #27).
+
+        Per file, collection cost scaled with *file count* rather than test count —
+        measured at 77% of a whole `run start`, against a floor of ~1.08s per
+        invocation that is the environment manager resolving plus the runner
+        booting, not collection work.
+
+        The batch and the loop do not discover the same way: a batch uses the
+        runner's own config, the loop walks `test_paths`. So the loop still runs for
+        every file the batch did not report — whether the batch failed, returned
+        nothing, or simply never mentioned that file. R10.3's guarantee is
+        unchanged: one uncollectable file is attributed to itself, and cannot
+        destroy the set. What changes is that the healthy case no longer pays for
+        the broken one.
+        """
+        result = Collection()
+        unaccounted = {str(p.relative_to(self.root)) for p in self._test_files()}
+        for command, env in self._collect_invocations():
+            batch = self._collect_batch(command, env)
+            if batch is None:
+                continue
+            tests, files = batch
+            result.tests |= tests
+            unaccounted -= files
+        return self._collect_per_file(unaccounted, result)
+
+    def _collect_invocations(self) -> list[tuple[str, dict[str, str] | None]]:
+        """One collection command per declared suite: the default plus each
+        override's (R7.13), so an override's files are enumerated with its own
+        command and env."""
+        raise NotImplementedError
+
+    def _collect_batch(
+        self, command: str, env: dict[str, str] | None
+    ) -> tuple[set[str], set[str]] | None:
+        """`(test ids, files seen)` for one whole-suite invocation, or None when it
+        produced nothing usable — the signal to leave its files to the loop."""
+        raise NotImplementedError
+
+    def _collect_per_file(self, rels: set[str], result: Collection) -> Collection:
+        """R10.3's loop, now reached only for files no batch accounted for."""
         raise NotImplementedError
 
     def collectable(self) -> GateResult:
@@ -184,7 +254,7 @@ class Adapter:
     def _gate(self, commands: list[str]) -> GateResult:
         chunks = []
         for cmd in commands:
-            code, out, err = run_command(cmd, self.root)
+            code, out, err = run_command(cmd, self.root, label="gate")
             if code != 0:
                 chunks.append(f"$ {cmd}\n{out}\n{err}".strip())
         return GateResult(ok=not chunks, output="\n\n".join(chunks)[:4000])

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/src/tddcli/adapters/pytest_adapter.py

@@ -199,7 +199,7 @@ class PytestAdapter(Adapter):
         ]
         for cmd, env in probes:
             code, out, err = run_command(
-                f"{cmd} --collect-only -q", self.root, extra_env=env
+                f"{cmd} --collect-only -q", self.root, extra_env=env, label="doctor"
             )
             if code != 0:
                 chunks.append(out.strip())
@@ -215,7 +215,9 @@ class PytestAdapter(Adapter):
         if not self.project.overrides:
             return GateResult(ok=True)
         probe = f"{self._test_cmd().replace('{workers}', '0')} --collect-only -q"
-        code, out, err = run_command(probe, self.root, extra_env=self._suite_env(None))
+        code, out, err = run_command(
+            probe, self.root, extra_env=self._suite_env(None), label="doctor"
+        )
         reached = sorted({
             f for f in (
                 line.split("::", 1)[0]
@@ -233,22 +235,49 @@ class PytestAdapter(Adapter):
             " cannot reach them (e.g. `pytest tests/`)."
         ))
 
-    def collect(self) -> Collection:
+    def _collect_invocations(self) -> list[tuple[str, dict[str, str] | None]]:
+        """An override without a `collect_command` collects with its `test_command`
+        — `--collect-only` composes with any run command."""
+        return [(self._collect_cmd(), self._suite_env(None))] + [
+            (ov.collect_command or ov.test_command, self._suite_env(ov))
+            for ov in self.project.overrides
+        ]
+
+    def _nodeids(self, out: str) -> tuple[set[str], set[str]]:
+        tests: set[str] = set()
+        files: set[str] = set()
+        for line in out.splitlines():
+            line = line.strip()
+            if "::" in line and not line.startswith(("=", "-", "no tests")):
+                tests.add(self.qualify(line))
+                files.add(line.split("::", 1)[0])
+        return tests, files
+
+    def _collect_batch(
+        self, command: str, env: dict[str, str] | None
+    ) -> tuple[set[str], set[str]] | None:
+        code, out, err = run_command(
+            f"{command} --collect-only -q", self.root, extra_env=env, label="collect"
+        )
+        if code != 0:
+            return None
+        # An empty result needs no special case: it accounts for no files, so every
+        # file falls to the loop exactly as a failure would.
+        return self._nodeids(out)
+
+    def _collect_per_file(self, rels: set[str], result: Collection) -> Collection:
         """Per file (R10.3) — one uncollectable module must not destroy the whole set."""
-        result = Collection()
-        for path in self._test_files():
-            rel = path.relative_to(self.root)
-            base, env = self._collect_cmd_for(str(rel))
+        for rel in sorted(rels):
+            base, env = self._collect_cmd_for(rel)
             code, out, err = run_command(
-                f"{base} --collect-only -q {shlex.quote(str(rel))}",
+                f"{base} --collect-only -q {shlex.quote(rel)}",
                 self.root,
                 extra_env=env,
+                label="collect",
             )
             if code != 0:
-                result.failed_files[str(rel)] = (err or out).strip()[:800]
+                result.failed_files[rel] = (err or out).strip()[:800]
                 continue
-            for line in out.splitlines():
-                line = line.strip()
-                if "::" in line and not line.startswith(("=", "-", "no tests")):
-                    result.tests.add(self.qualify(line))
+            tests, _ = self._nodeids(out)
+            result.tests |= tests
         return result

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/src/tddcli/adapters/vitest_adapter.py

@@ -177,7 +177,7 @@ class VitestAdapter(Adapter):
         """
         chunks = []
         code, out, err = run_command(
-            self._collect_cmd(), self.root, extra_env=self._suite_env(None)
+            self._collect_cmd(), self.root, extra_env=self._suite_env(None), label="doctor"
         )
         if code != 0:
             chunks.append((err or out).strip())
@@ -190,7 +190,7 @@ class VitestAdapter(Adapter):
                 )
                 continue
             code, out, err = run_command(
-                ov.collect_command, self.root, extra_env=self._suite_env(ov)
+                ov.collect_command, self.root, extra_env=self._suite_env(ov), label="doctor"
             )
             if code != 0:
                 chunks.append((err or out).strip())
@@ -204,7 +204,7 @@ class VitestAdapter(Adapter):
         if not self.project.overrides:
             return GateResult(ok=True)
         code, out, err = run_command(
-            self._collect_cmd(), self.root, extra_env=self._suite_env(None)
+            self._collect_cmd(), self.root, extra_env=self._suite_env(None), label="doctor"
         )
         reached = sorted({
             f for f in (
@@ -223,10 +223,43 @@ class VitestAdapter(Adapter):
             " config (test.exclude) or scope its include globs."
         ))
 
-    def collect(self) -> Collection:
-        result = Collection()
-        for path in self._test_files():
-            rel = path.relative_to(self.root)
+    def _collect_invocations(self) -> list[tuple[str, dict[str, str] | None]]:
+        """An override with no `collect_command` gets no batch — `vitest list` knows
+        nothing of its config, and its files fall to the loop, which records the
+        missing `collect_command` against each of them as before."""
+        return [(self._collect_cmd(), self._suite_env(None))] + [
+            (ov.collect_command, self._suite_env(ov))
+            for ov in self.project.overrides if ov.collect_command
+        ]
+
+    def _collect_batch(
+        self, command: str, env: dict[str, str] | None
+    ) -> tuple[set[str], set[str]] | None:
+        """Unlike `_parse_list_output`, which pins every id to the one file it was
+        given, a whole-suite listing must read the file from each line — the same
+        `file > describe > name` shape, one file per line rather than one per run."""
+        code, out, err = run_command(command, self.root, extra_env=env, label="collect")
+        if code != 0:
+            return None
+        tests: set[str] = set()
+        files: set[str] = set()
+        for line in out.splitlines():
+            line = line.strip()
+            if " > " not in line:
+                continue
+            rel, _, remainder = line.partition(" > ")
+            full_name = " ".join(part.strip() for part in remainder.split(" > "))
+            if rel and full_name:
+                tests.add(self.qualify(f"{rel.strip()} > {full_name}"))
+                files.add(rel.strip())
+        # An empty result needs no special case: it accounts for no files, so every
+        # file falls to the loop exactly as a failure would.
+        return tests, files
+
+    def _collect_per_file(self, rels: set[str], result: Collection) -> Collection:
+        for rel_str in sorted(rels):
+            path = self.root / rel_str
+            rel = Path(rel_str)
             ov = self.project.override_for(str(rel))
             if ov is not None and not ov.collect_command:
                 result.failed_files[str(rel)] = (
@@ -237,7 +270,8 @@ class VitestAdapter(Adapter):
             base = ov.collect_command if ov else self._collect_cmd()
             env = self._suite_env(ov)
             code, out, err = run_command(
-                f"{base} {shlex.quote(str(rel))}", self.root, extra_env=env
+                f"{base} {shlex.quote(str(rel))}", self.root, extra_env=env,
+                label="collect",
             )
 
             payload = _extract_json(out)

{tdd_cli-0.3.0 → tdd_cli-0.4.1}/src/tddcli/cli.py

@@ -13,6 +13,7 @@ import socket
 import sqlite3
 import sys
 import time
+from collections.abc import Callable
 from datetime import datetime, timezone
 from pathlib import Path
 
@@ -217,16 +218,61 @@ def _legacy_artifacts(worktree: Path) -> list[Path]:
     return sorted(found)
 
 
-def cmd_doctor(args) -> Envelope:
-    worktree = _worktree()
+def _doctor_checklist() -> tuple[list[dict], Callable]:
+    """A checks list and its recorder, which refuses a blocker it cannot explain.
+
+    `resolve_blocker` with an empty `detail` is unfalsifiable: an agent is told to
+    fix something and given nothing to fix. It re-runs doctor, reads the identical
+    output, and loops. Enforcing the detail here means a check added later inherits
+    the guarantee instead of relying on its author to remember.
+    """
     checks: list[dict] = []
 
     def check(name, ok, detail="", project=None):
+        if not ok and not str(detail).strip():
+            raise AssertionError(
+                f"doctor check {name!r} would fail silently: a failing check must"
+                " name what to fix"
+            )
         entry = {"check": name, "ok": bool(ok), "detail": detail}
         if project is not None:
             entry["project"] = project
         checks.append(entry)
 
+    return checks, check
+
+
+def _blocks_the_loop(rel_path: str, cfg) -> bool:
+    """Whether dirt at `rel_path` is somewhere a run would actually read it."""
+    if rel_path == config_mod.CONFIG_NAME:
+        return True
+    if cfg.owning_project(rel_path) is not None:
+        return True
+    return any(art.owns(rel_path) for art in cfg.artifacts.values())
+
+
+def _cleanliness_detail(blocking: list[str], unrelated: list[str]) -> str:
+    def listed(paths: list[str]) -> str:
+        head = ", ".join(paths[:5])
+        return head if len(paths) <= 5 else f"{head} (+{len(paths) - 5} more)"
+
+    if blocking:
+        return (
+            f"uncommitted changes a run would observe: {listed(blocking)}."
+            " Commit, stash, or gitignore them before `tdd run start`."
+        )
+    if unrelated:
+        return (
+            f"clean where a run reads; {len(unrelated)} unrelated path(s) left as-is:"
+            f" {listed(unrelated)}"
+        )
+    return ""
+
+
+def cmd_doctor(args) -> Envelope:
+    worktree = _worktree()
+    checks, check = _doctor_checklist()
+
     check("worktree resolvable", True, str(worktree))
     try:
         cfg = config_mod.load(worktree)
@@ -246,13 +292,20 @@ def cmd_doctor(args) -> Envelope:
         root = worktree / project.root
         check("root exists", root.is_dir(), str(root), project=name)
         check("adapter known", project.adapter in adapters.available(), project.adapter, project=name)
-        check("test_paths declared", bool(project.test_paths), project=name)
+        declared = bool(project.test_paths)
+        check(
+            "test_paths declared", declared,
+            "" if declared
+            else f"add `test_paths` to [project.{name}] in tdd.toml — without it no"
+                 " suite can be discovered for this project",
+            project=name,
+        )
         if project.adapter == "pytest":
             # The probe runs in the project's own environment (uv, poetry, pipenv,
             # pdm or the active venv) — hardcoding `uv run` here failed the check
             # on any non-uv project even with the plugin installed.
             probe = adapters.build(project, worktree).plugin_probe_cmd()
-            code, out, err = adapters.base.run_command(probe, root)
+            code, out, err = adapters.base.run_command(probe, root, label="doctor")
             check("pytest-json-report installed", code == 0, (err or "")[:200], project=name)
 
         # Run before `collectable()` so this actionable message wins over
@@ -289,11 +342,33 @@ def cmd_doctor(args) -> Envelope:
         projects[name] = {"ok": all(c["ok"] for c in checks[before:])}
 
     for art in cfg.artifacts.values():
-        check(f"artifact {art.name}: has check or regenerate", bool(art.check or art.regenerate))
+        # One evaluation feeds both `ok` and the detail: evaluating the condition
+        # twice lets them disagree, and a passing check that still says "add a hook"
+        # is the same misdirection as a failing check that says nothing.
+        has_hook = bool(art.check or art.regenerate)
+        check(
+            f"artifact {art.name}: has check or regenerate", has_hook,
+            "" if has_hook
+            else f"add `check` or `regenerate` to [artifact.{art.name}] in tdd.toml —"
+                 " freshness cannot be verified without one",
+        )
 
     stale = _legacy_artifacts(worktree)
-    check("no legacy state artifacts", not stale, ", ".join(str(s) for s in stale[:5]))
-    check("worktree clean", not gitutil.is_dirty(worktree))
+    check(
+        "no legacy state artifacts", not stale,
+        f"delete these pre-ledger state files: {', '.join(str(s) for s in stale[:5])}"
+        if stale else "",
+    )
+
+    # Only dirt a run would read can corrupt one. Blocking on everything else
+    # stopped agents on unrelated notes and editor settings, and — because the
+    # check named no path — gave them nothing to act on but a re-run. `is_ignored`
+    # also excludes doctor's own probe residue (`.venv`, `node_modules`, caches),
+    # so running doctor can no longer be what makes doctor fail.
+    dirt = sorted(p for p in gitutil.dirty_paths(worktree) if not cfg.is_ignored(p))
+    blocking = [p for p in dirt if _blocks_the_loop(p, cfg)]
+    unrelated = [p for p in dirt if p not in set(blocking)]
+    check("worktree clean", not blocking, _cleanliness_detail(blocking, unrelated))
 
     ok = all(c["ok"] for c in checks)
     return Envelope(
@@ -364,12 +439,18 @@ def _probe_projects(cfg, worktree, ledger, on_progress):
     for done, (name, project) in enumerate(cfg.projects.items(), start=1):
         adapter = adapters.build(project, worktree)
         started = time.monotonic()
-        verdict, collection = adapter.run(None), adapter.collect()
+        verdict = adapter.run(None)
+        ran = time.monotonic()
+        collection = adapter.collect()
         elapsed = time.monotonic() - started
         probes[name] = (verdict, collection)
+        # Split, not just totalled: `run` and `collect` have unrelated cost models
+        # — one scales with tests, the other with files — and a single number sends
+        # whoever asks "why was that slow?" out of the tool to measure by hand.
         heartbeat(
             event="baseline_captured", project=name,
             test_count=len(collection.tests), elapsed_s=round(elapsed, 2),
+            run_s=round(ran - started, 2), collect_s=round(elapsed - (ran - started), 2),
         )
         on_progress(done, name)
     return probes