mng/better-ratchet-tests

libs/imbue_common/imbue/imbue_common/ratchet_testing/core.py

@@ -105,36 +105,41 @@ def _get_all_files_with_extension(
     folder_path: Path,
     extension: FileExtension | None,
 ) -> tuple[Path, ...]:
-    """Get all git-tracked files in a folder (cached).
+    """Get all non-gitignored files that exist on disk in a folder (cached).
 
     If extension is provided, only files matching that extension are returned.
-    If extension is None, all tracked files are returned.
+    If extension is None, all non-ignored files are returned.
+
+    Uses git ls-files with --cached and --others to include both tracked
+    and untracked files while respecting .gitignore rules. Filters the
+    result to only files that actually exist on disk, so deleted files
+    that are still in the git index are excluded.
     """
     glob_pattern = f"*{extension}" if extension is not None else "*"
     try:
         result = subprocess.run(
-            ["git", "ls-files", glob_pattern],
+            ["git", "ls-files", "--cached", "--others", "--exclude-standard", glob_pattern],
             cwd=folder_path,
             capture_output=True,
             text=True,
             check=True,
         )
     except subprocess.CalledProcessError as e:
-        raise GitCommandError(f"Failed to list git-tracked files in {folder_path}") from e
+        raise GitCommandError(f"Failed to list files in {folder_path}") from e
 
     file_paths = [folder_path / line.strip() for line in result.stdout.splitlines() if line.strip()]
-    return tuple(file_paths)
+    return tuple(f for f in file_paths if f.exists())
 
 
 def _get_non_ignored_files_with_extension(
     folder_path: Path,
     extension: FileExtension | None,
     excluded_path_patterns: tuple[str, ...] = (),
 ) -> tuple[Path, ...]:
-    """Get git-tracked files in a folder, with optional path exclusions.
+    """Get non-gitignored files on disk in a folder, with optional path exclusions.
 
     If extension is provided, only files matching that extension are returned.
-    If extension is None, all tracked files are returned.
+    If extension is None, all non-ignored files are returned.
 
     Each pattern in excluded_path_patterns is matched against file paths using Path.match(),
     which matches from the right for relative patterns (e.g., "test_*.py" matches any file
@@ -259,10 +264,10 @@ def get_ratchet_failures(
     pattern: RegexPattern,
     excluded_path_patterns: tuple[str, ...] = (),
 ) -> tuple[RatchetMatchChunk, ...]:
-    """Find all regex matches in git-tracked files and return them sorted by file path and line number.
+    """Find all regex matches in non-gitignored files on disk and return them sorted by file path and line number.
 
     If extension is provided, only files matching that extension are searched.
-    If extension is None, all tracked files are searched.
+    If extension is None, all non-ignored files are searched.
 
     Blame dates are not computed here; they are resolved on demand via _resolve_blame_dates()
     when a failure message needs to be formatted.

libs/imbue_common/imbue/imbue_common/ratchet_testing/core_test.py

@@ -10,6 +10,7 @@
 from imbue.imbue_common.ratchet_testing.core import LineNumber
 from imbue.imbue_common.ratchet_testing.core import RatchetMatchChunk
 from imbue.imbue_common.ratchet_testing.core import RegexPattern
+from imbue.imbue_common.ratchet_testing.core import _get_all_files_with_extension
 from imbue.imbue_common.ratchet_testing.core import _get_non_ignored_files_with_extension
 from imbue.imbue_common.ratchet_testing.core import _read_file_contents
 from imbue.imbue_common.ratchet_testing.core import format_ratchet_failure_message
@@ -156,6 +157,50 @@ def test_get_non_ignored_files_excludes_by_glob_pattern(git_repo: Path) -> None:
     assert result[0].name == "main.py"
 
 
+def test_get_all_files_excludes_deleted_but_tracked_files(git_repo: Path) -> None:
+    """Deleted files still in the git index should not be returned."""
+    tracked_file = git_repo / "tracked.py"
+    tracked_file.write_text("print('hello')")
+
+    subprocess.run(["git", "add", "."], cwd=git_repo, check=True, capture_output=True)
+    subprocess.run(
+        ["git", "commit", "-m", "Add tracked file"],
+        cwd=git_repo,
+        check=True,
+        capture_output=True,
+    )
+
+    # Delete the file from disk but don't stage the deletion
+    tracked_file.unlink()
+
+    result = _get_all_files_with_extension(git_repo, FileExtension(".py"))
+    result_names = [p.name for p in result]
+    assert "tracked.py" not in result_names
+
+
+def test_get_all_files_includes_untracked_non_ignored_files(git_repo: Path) -> None:
+    """Untracked files that are not gitignored should be returned."""
+    # Create a committed file
+    committed_file = git_repo / "committed.py"
+    committed_file.write_text("print('committed')")
+    subprocess.run(["git", "add", "."], cwd=git_repo, check=True, capture_output=True)
+    subprocess.run(
+        ["git", "commit", "-m", "Initial commit"],
+        cwd=git_repo,
+        check=True,
+        capture_output=True,
+    )
+
+    # Create an untracked file (not added to git, not ignored)
+    untracked_file = git_repo / "untracked.py"
+    untracked_file.write_text("print('untracked')")
+
+    result = _get_all_files_with_extension(git_repo, FileExtension(".py"))
+    result_names = [p.name for p in result]
+    assert "committed.py" in result_names
+    assert "untracked.py" in result_names
+
+
 def test_read_file_contents_caches_results(tmp_path: Path) -> None:
     test_file = tmp_path / "test.txt"
     test_file.write_text("original content")

libs/mng/imbue/mng/cli/test_provision.py

@@ -1,5 +1,4 @@
 import json
-import time
 from pathlib import Path
 
 import pluggy
@@ -9,6 +8,7 @@
 from imbue.mng.cli.create import create
 from imbue.mng.cli.provision import provision
 from imbue.mng.cli.stop import stop
+from imbue.mng.utils.polling import wait_for
 from imbue.mng.utils.testing import ModalSubprocessTestEnv
 from imbue.mng.utils.testing import get_short_random_string
 from imbue.mng.utils.testing import run_mng_subprocess
@@ -23,7 +23,7 @@ def test_provision_existing_agent(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that provisioning an existing agent succeeds."""
-    agent_name = f"test-provision-{int(time.time())}"
+    agent_name = f"test-provision-{get_short_random_string()}"
     create_test_agent(agent_name)
 
     result = cli_runner.invoke(
@@ -44,7 +44,7 @@ def test_provision_with_user_command(
     tmp_path: Path,
 ) -> None:
     """Test that provisioning with --user-command executes the command."""
-    agent_name = f"test-prov-cmd-{int(time.time())}"
+    agent_name = f"test-prov-cmd-{get_short_random_string()}"
     marker_file = tmp_path / "provision_marker.txt"
 
     create_test_agent(agent_name)
@@ -73,7 +73,7 @@ def test_provision_with_env_var(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that provisioning with --env sets environment variables."""
-    agent_name = f"test-prov-env-{int(time.time())}"
+    agent_name = f"test-prov-env-{get_short_random_string()}"
 
     create_test_agent(agent_name)
 
@@ -111,7 +111,7 @@ def test_provision_preserves_existing_env_vars(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that provisioning preserves existing environment variables."""
-    agent_name = f"test-prov-env-preserve-{int(time.time())}"
+    agent_name = f"test-prov-env-preserve-{get_short_random_string()}"
     session_name = f"{mng_test_prefix}{agent_name}"
 
     with tmux_session_cleanup(session_name):
@@ -166,7 +166,7 @@ def test_provision_with_upload_file(
     tmp_path: Path,
 ) -> None:
     """Test that provisioning with --upload-file transfers the file."""
-    agent_name = f"test-prov-upload-{int(time.time())}"
+    agent_name = f"test-prov-upload-{get_short_random_string()}"
 
     # Create a local file to upload
     local_file = tmp_path / "upload_source.txt"
@@ -212,7 +212,7 @@ def test_provision_with_agent_option(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --agent option works as an alternative to positional argument."""
-    agent_name = f"test-prov-opt-{int(time.time())}"
+    agent_name = f"test-prov-opt-{get_short_random_string()}"
 
     create_test_agent(agent_name)
 
@@ -255,7 +255,7 @@ def test_provision_json_output(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --format json produces JSON output."""
-    agent_name = f"test-prov-json-{int(time.time())}"
+    agent_name = f"test-prov-json-{get_short_random_string()}"
 
     create_test_agent(agent_name)
 
@@ -286,7 +286,7 @@ def test_provision_stopped_agent(
     need the agent process running. Previously, provisioning a stopped agent
     failed because the agent lookup required the agent to be running.
     """
-    agent_name = f"test-prov-stopped-{int(time.time())}"
+    agent_name = f"test-prov-stopped-{get_short_random_string()}"
 
     create_test_agent(agent_name)
 
@@ -323,7 +323,7 @@ def test_provision_stopped_agent_with_user_command(
     process is stopped, since provisioning operates on the host, not
     the agent process.
     """
-    agent_name = f"test-prov-stopped-cmd-{int(time.time())}"
+    agent_name = f"test-prov-stopped-cmd-{get_short_random_string()}"
     marker_file = tmp_path / "stopped_provision_marker.txt"
 
     create_test_agent(agent_name)
@@ -365,7 +365,7 @@ def test_provision_running_agent_restarts_by_default(
     The agent should be stopped before provisioning and restarted after,
     and should be running after provisioning completes.
     """
-    agent_name = f"test-prov-restart-{int(time.time())}"
+    agent_name = f"test-prov-restart-{get_short_random_string()}"
     session_name = create_test_agent(agent_name)
 
     # Verify agent is running before provisioning
@@ -381,8 +381,13 @@ def test_provision_running_agent_restarts_by_default(
 
     assert result.exit_code == 0, f"Provision failed with: {result.output}"
 
-    # Agent should still be running after provisioning (restarted)
-    assert tmux_session_exists(session_name), "Agent should be running after provision with restart"
+    # Agent should still be running after provisioning (restarted).
+    # Use wait_for to tolerate brief delays when the agent is restarted under heavy xdist load.
+    wait_for(
+        lambda: tmux_session_exists(session_name),
+        timeout=10.0,
+        error_message="Agent should be running after provision with restart",
+    )
 
 
 @pytest.mark.tmux
@@ -392,7 +397,7 @@ def test_provision_running_agent_no_restart_keeps_running(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --no-restart does not stop/restart a running agent."""
-    agent_name = f"test-prov-norestart-{int(time.time())}"
+    agent_name = f"test-prov-norestart-{get_short_random_string()}"
     session_name = create_test_agent(agent_name)
 
     # Verify agent is running before provisioning
@@ -423,7 +428,7 @@ def test_provision_stopped_agent_stays_stopped_with_restart(
     When is_restart=True, only agents that were running before provisioning should
     be restarted. A stopped agent should remain stopped.
     """
-    agent_name = f"test-prov-stopped-norestart-{int(time.time())}"
+    agent_name = f"test-prov-stopped-norestart-{get_short_random_string()}"
     session_name = create_test_agent(agent_name)
 
     # Stop the agent

libs/mng/imbue/mng/cli/test_rename.py

@@ -116,8 +116,13 @@ def test_rename_running_agent_renames_tmux_session(
         assert rename_result.exit_code == 0, f"Rename failed: {rename_result.output}"
         assert "Renamed agent:" in rename_result.output
 
-        # The old session should be gone, the new one should exist
-        assert tmux_session_exists(new_session_name), "New tmux session should exist"
+        # The old session should be gone, the new one should exist.
+        # Use wait_for to tolerate brief propagation delays under heavy xdist load.
+        wait_for(
+            lambda: tmux_session_exists(new_session_name),
+            timeout=10.0,
+            error_message=f"New tmux session {new_session_name} should exist after rename",
+        )
         assert not tmux_session_exists(old_session_name), "Old tmux session should not exist"
 
 

libs/mng/imbue/mng/cli/test_snapshot.py

@@ -2,7 +2,6 @@
 
 import json
 import subprocess
-import time
 from pathlib import Path
 from typing import Any
 
@@ -27,7 +26,7 @@ def test_snapshot_create_local_agent_rejects_unsupported_provider(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that snapshot create fails for a local agent (unsupported provider)."""
-    agent_name = f"test-snap-create-{int(time.time())}"
+    agent_name = f"test-snap-create-{get_short_random_string()}"
     create_test_agent(agent_name)
 
     result = cli_runner.invoke(
@@ -48,7 +47,7 @@ def test_snapshot_create_dry_run_jsonl_resolves_local_agent(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --dry-run with --format jsonl outputs structured data on stdout."""
-    agent_name = f"test-snap-dryrun-jsonl-{int(time.time())}"
+    agent_name = f"test-snap-dryrun-jsonl-{get_short_random_string()}"
     create_test_agent(agent_name)
 
     result = cli_runner.invoke(
@@ -117,7 +116,7 @@ def test_snapshot_create_on_error_continue_reports_failure(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --on-error continue reports the error and exits 1 (doesn't crash)."""
-    agent_name = f"test-snap-onerror-cont-{int(time.time())}"
+    agent_name = f"test-snap-onerror-cont-{get_short_random_string()}"
     create_test_agent(agent_name)
 
     result = cli_runner.invoke(
@@ -138,7 +137,7 @@ def test_snapshot_create_on_error_abort_reports_failure(
     plugin_manager: pluggy.PluginManager,
 ) -> None:
     """Test that --on-error abort also fails (with abort message)."""
-    agent_name = f"test-snap-onerror-abort-{int(time.time())}"
+    agent_name = f"test-snap-onerror-abort-{get_short_random_string()}"
     create_test_agent(agent_name)
 
     result = cli_runner.invoke(

libs/mng_recursive/imbue/mng_recursive/provisioning.py

@@ -59,7 +59,7 @@ def _upload_deploy_files(
     # do this in parallel, since there can sometimes be a bunch of things to transfer
     # first, figure out all directories and do a single mkdir -p that captures all of them:
     remote_paths: list[str] = []
-    for dest_path, source in deploy_files.items():
+    for dest_path in deploy_files:
         resolved_path = _resolve_remote_path(dest_path, remote_home)
         remote_paths.append(shlex.quote(str(resolved_path.parent)))
     mkdir_result = host.execute_command(f"mkdir -p {' '.join(remote_paths)}")
@@ -71,7 +71,7 @@ def _upload_deploy_files(
     count = 0
     with ConcurrencyGroupExecutor(
         parent_cg=mng_ctx.concurrency_group, name="upload_deploy_files", max_workers=32
-    ) as executor:
+    ) as _executor:
         for dest_path, source in deploy_files.items():
             resolved_path = _resolve_remote_path(dest_path, remote_home)