mlx-stack 0.3.2__tar.gz → 0.3.3__tar.gz

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/.github/workflows/ci.yml

@@ -32,16 +32,16 @@ jobs:
         run: uv python install ${{ matrix.python-version }}
 
       - name: Install dependencies
-        run: uv sync --dev
+        run: make install
 
       - name: Lint
-        run: uv run ruff check src/ tests/
+        run: make lint
 
       - name: Typecheck
-        run: uv run python -m pyright
+        run: make typecheck
 
       - name: Test with coverage
-        run: uv run pytest --cov=src/mlx_stack --cov-report=xml -x -q --tb=short
+        run: make test
 
       - name: Upload coverage report
         uses: actions/upload-artifact@v4

mlx_stack-0.3.3/.github/workflows/release-please.yml

@@ -0,0 +1,100 @@
+name: Release Please
+
+on:
+  push:
+    branches: [main]
+
+permissions:
+  contents: write
+  pull-requests: write
+
+jobs:
+  release-please:
+    runs-on: ubuntu-latest
+    outputs:
+      release_created: ${{ steps.release.outputs.release_created }}
+      tag_name: ${{ steps.release.outputs.tag_name }}
+    steps:
+      - uses: googleapis/release-please-action@v4
+        id: release
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          config-file: release-please-config.json
+          manifest-file: .release-please-manifest.json
+
+      - name: Regenerate release notes
+        if: ${{ steps.release.outputs.release_created == 'true' }}
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          gh api repos/${{ github.repository }}/releases/generate-notes \
+            -f tag_name=${{ steps.release.outputs.tag_name }} \
+            --jq '.body' \
+          | gh release edit ${{ steps.release.outputs.tag_name }} \
+              --repo ${{ github.repository }} \
+              --notes-file -
+
+  ci:
+    name: CI
+    needs: release-please
+    if: ${{ needs.release-please.outputs.release_created == 'true' }}
+    runs-on: macos-latest
+    strategy:
+      matrix:
+        python-version: ["3.13"]
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+
+      - name: Install Python ${{ matrix.python-version }}
+        run: uv python install ${{ matrix.python-version }}
+
+      - name: Install dependencies
+        run: make install
+
+      - name: Lint
+        run: make lint
+
+      - name: Typecheck
+        run: make typecheck
+
+      - name: Test
+        run: make test
+
+  publish:
+    name: Build & Publish
+    needs: [release-please, ci]
+    if: ${{ needs.release-please.outputs.release_created == 'true' }}
+    runs-on: ubuntu-latest
+
+    permissions:
+      id-token: write
+
+    environment:
+      name: pypi
+      url: https://pypi.org/p/mlx-stack
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v6
+        with:
+          ref: ${{ needs.release-please.outputs.tag_name }}
+          fetch-depth: 0
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+
+      - name: Build package
+        run: uv build
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

mlx_stack-0.3.3/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.3.3"
+}

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/CHANGELOG.md

@@ -4,6 +4,13 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.3.3](https://github.com/weklund/mlx-stack/compare/v0.3.2...v0.3.3) (2026-04-02)
+
+
+### Bug Fixes
+
+* replace raw status strings with ServiceHealth enum and assert behavior in tests ([#13](https://github.com/weklund/mlx-stack/issues/13)) ([ef0161c](https://github.com/weklund/mlx-stack/commit/ef0161c9f0b64096917fd06045786aad49fe8c93))
+
 ## [0.3.2](https://github.com/weklund/mlx-stack/compare/v0.3.1...v0.3.2) (2026-04-02)
 
 

mlx_stack-0.3.3/Makefile

@@ -0,0 +1,20 @@
+.PHONY: install lint typecheck test check
+
+## Install dev dependencies
+install:
+	uv sync --dev
+
+## Lint source and tests
+lint:
+	uv run ruff check src/ tests/
+
+## Run type checker across the full project
+typecheck:
+	uv run python -m pyright
+
+## Run tests with coverage
+test:
+	uv run pytest --cov=src/mlx_stack -x -q --tb=short
+
+## Run all checks (same as CI)
+check: lint typecheck test

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mlx-stack
-Version: 0.3.2
+Version: 0.3.3
 Summary: CLI control plane for local LLM infrastructure on Apple Silicon
 Project-URL: Homepage, https://github.com/weklund/mlx-stack
 Project-URL: Repository, https://github.com/weklund/mlx-stack

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/src/mlx_stack/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '0.3.2'
-__version_tuple__ = version_tuple = (0, 3, 2)
+__version__ = version = '0.3.3'
+__version_tuple__ = version_tuple = (0, 3, 3)
 
 __commit_id__ = commit_id = None

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/src/mlx_stack/cli/status.py

@@ -14,17 +14,22 @@ from rich.console import Console
 from rich.table import Table
 from rich.text import Text
 
-from mlx_stack.core.stack_status import StatusResult, run_status, status_to_dict
+from mlx_stack.core.stack_status import (
+    ServiceHealth,
+    StatusResult,
+    run_status,
+    status_to_dict,
+)
 
 console = Console(stderr=True)
 
 # Status display styling — maps state to Rich markup
-_STATUS_STYLES: dict[str, str] = {
-    "healthy": "[bold green]healthy[/bold green]",
-    "degraded": "[bold yellow]degraded[/bold yellow]",
-    "down": "[bold red]down[/bold red]",
-    "crashed": "[bold red]crashed[/bold red]",
-    "stopped": "[dim]stopped[/dim]",
+_STATUS_STYLES: dict[ServiceHealth, str] = {
+    ServiceHealth.HEALTHY: "[bold green]healthy[/bold green]",
+    ServiceHealth.DEGRADED: "[bold yellow]degraded[/bold yellow]",
+    ServiceHealth.DOWN: "[bold red]down[/bold red]",
+    ServiceHealth.CRASHED: "[bold red]crashed[/bold red]",
+    ServiceHealth.STOPPED: "[dim]stopped[/dim]",
 }
 
 

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/src/mlx_stack/core/stack_status.py

@@ -16,6 +16,7 @@ Implements 5-state reporting per service:
 from __future__ import annotations
 
 from dataclasses import dataclass, field
+from enum import StrEnum
 from typing import Any
 
 from mlx_stack.core.paths import get_stacks_dir
@@ -33,6 +34,16 @@ from mlx_stack.core.stack_up import LITELLM_HEALTH_PATH, LITELLM_SERVICE_NAME
 VLLM_HEALTH_PATH = "/v1/models"
 
 
+class ServiceHealth(StrEnum):
+    """Health state of a managed service."""
+
+    HEALTHY = "healthy"
+    DEGRADED = "degraded"
+    DOWN = "down"
+    CRASHED = "crashed"
+    STOPPED = "stopped"
+
+
 # --------------------------------------------------------------------------- #
 # Data classes
 # --------------------------------------------------------------------------- #
@@ -45,7 +56,7 @@ class ServiceStatus:
     tier: str
     model: str
     port: int
-    status: str  # "healthy", "degraded", "down", "crashed", "stopped"
+    status: ServiceHealth
     uptime: float | None  # seconds, None for stopped/crashed
     uptime_display: str  # human-readable string or "-"
     response_time: float | None  # seconds, None if no HTTP response
@@ -171,7 +182,7 @@ def run_status(stack_name: str = "default") -> StatusResult:
             tier=tier_name,
             model=model,
             port=port,
-            status=svc_status["status"],
+            status=ServiceHealth(svc_status["status"]),
             uptime=svc_status["uptime"],
             uptime_display=format_uptime(svc_status["uptime"]),
             response_time=svc_status["response_time"],
@@ -190,7 +201,7 @@ def run_status(stack_name: str = "default") -> StatusResult:
         tier="litellm",
         model="proxy",
         port=litellm_port,
-        status=litellm_status["status"],
+        status=ServiceHealth(litellm_status["status"]),
         uptime=litellm_status["uptime"],
         uptime_display=format_uptime(litellm_status["uptime"]),
         response_time=litellm_status["response_time"],

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/src/mlx_stack/core/watchdog.py

@@ -34,6 +34,7 @@ from mlx_stack.core.process import (
     write_pid_file,
 )
 from mlx_stack.core.stack_status import (
+    ServiceHealth,
     ServiceStatus,
     run_status,
 )
@@ -576,10 +577,10 @@ def poll_cycle(
         # Try to reset flap state if service has been stable
         reset_flap_state(tracker)
 
-        if svc.status != "crashed":
+        if svc.status != ServiceHealth.CRASHED:
             # Service is not crashed — reset consecutive failure counter
             # if it was previously restarted and is now healthy
-            if svc.status == "healthy" and tracker.consecutive_failures > 0:
+            if svc.status == ServiceHealth.HEALTHY and tracker.consecutive_failures > 0:
                 tracker.consecutive_failures = 0
             continue
 

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/tests/unit/test_cli_pull.py

@@ -20,6 +20,7 @@ import json
 from pathlib import Path
 from unittest.mock import MagicMock, patch
 
+import pytest
 from click.testing import CliRunner
 
 from mlx_stack.cli.main import cli
@@ -619,8 +620,9 @@ class TestDownloadModel:
         self,
         mock_run: MagicMock,
         tmp_path: Path,
+        capsys: pytest.CaptureFixture[str],
     ) -> None:
-        """Successful download on first attempt."""
+        """Successful download on first attempt shows completion message."""
         from rich.console import Console
 
         from mlx_stack.core.pull import download_model
@@ -628,7 +630,15 @@ class TestDownloadModel:
         local_dir = tmp_path / "model"
         console = Console()
         download_model("mlx-community/test-4bit", local_dir, console)
-        mock_run.assert_called_once()
+
+        # Verify correct repo and path were passed to the download function
+        args = mock_run.call_args
+        assert args[0][0] == "mlx-community/test-4bit"
+        assert args[0][1] == local_dir
+
+        # Verify user-facing success message was printed
+        output = capsys.readouterr().out
+        assert "Download complete" in output
 
     @patch("mlx_stack.core.pull._run_download")
     def test_retry_on_first_failure(
@@ -1022,7 +1032,7 @@ class TestPullCLI:
         mock_download: MagicMock,
         mlx_stack_home: Path,
     ) -> None:
-        """VAL-PULL-006: --force re-downloads."""
+        """VAL-PULL-006: --force re-downloads even when model exists."""
         mock_catalog.return_value = [_make_entry()]
 
         # Create existing model
@@ -1034,7 +1044,9 @@ class TestPullCLI:
         runner = CliRunner()
         result = runner.invoke(cli, ["pull", "qwen3.5-8b", "--force"])
         assert result.exit_code == 0
-        mock_download.assert_called_once()
+        # Must show success message (not the "already exists" skip)
+        assert "is ready" in result.output
+        assert "already exists" not in result.output
 
     @patch(
         "mlx_stack.core.pull.download_model",
@@ -1135,14 +1147,23 @@ class TestPullCLI:
         mock_download: MagicMock,
         mlx_stack_home: Path,
     ) -> None:
-        """VAL-CROSS-014: --bench flag accepted and runs after download."""
+        """VAL-CROSS-014: --bench flag runs benchmark and shows output."""
         mock_catalog.return_value = [_make_entry()]
 
-        with patch("mlx_stack.cli.pull._run_post_download_bench") as mock_bench:
+        with patch("mlx_stack.core.benchmark.run_benchmark") as mock_bench:
+            mock_bench.return_value = MagicMock(
+                prompt_tps_mean=150.0,
+                prompt_tps_std=5.0,
+                gen_tps_mean=80.0,
+                gen_tps_std=2.5,
+            )
             runner = CliRunner()
             result = runner.invoke(cli, ["pull", "qwen3.5-8b", "--bench"])
             assert result.exit_code == 0
-            mock_bench.assert_called_once()
+            # Must show benchmark output to the user
+            assert "Running post-download benchmark" in result.output
+            assert "Prompt TPS" in result.output
+            assert "Gen TPS" in result.output
 
     @patch("mlx_stack.core.pull.download_model")
     @patch("mlx_stack.core.pull.check_disk_space", return_value=(True, 100.0))

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/tests/unit/test_cli_status.py

@@ -23,6 +23,7 @@ from click.testing import CliRunner
 
 from mlx_stack.cli.main import cli
 from mlx_stack.core.stack_status import (
+    ServiceHealth,
     ServiceStatus,
     StatusResult,
     _load_stack_for_status,
@@ -509,7 +510,7 @@ class TestStatusToDict:
                     tier="standard",
                     model="big-model",
                     port=8000,
-                    status="healthy",
+                    status=ServiceHealth.HEALTHY,
                     uptime=3600.0,
                     uptime_display="1h",
                     response_time=0.05,
@@ -519,7 +520,7 @@ class TestStatusToDict:
                     tier="fast",
                     model="fast-model",
                     port=8001,
-                    status="stopped",
+                    status=ServiceHealth.STOPPED,
                     uptime=None,
                     uptime_display="-",
                     response_time=None,
@@ -558,7 +559,7 @@ class TestStatusToDict:
                     tier="t1",
                     model="m1",
                     port=8000,
-                    status="healthy",
+                    status=ServiceHealth.HEALTHY,
                     uptime=60.0,
                     uptime_display="1m",
                     response_time=0.1,
@@ -723,27 +724,49 @@ class TestStatusCli:
         mock_status: MagicMock,
         mlx_stack_home: Path,
     ) -> None:
-        """VAL-STATUS-003: Table shows tier names, models, ports."""
+        """VAL-STATUS-003: Table shows per-tier names, models, ports, and distinct statuses."""
+
         _write_stack_yaml(mlx_stack_home)
 
-        mock_status.return_value = {
-            "status": "healthy",
-            "pid": 12345,
-            "uptime": 150.0,
-            "response_time": 0.05,
-        }
+        def side_effect(service_name: str, port: int, health_path: str = "") -> dict[str, Any]:
+            if service_name == "standard":
+                return {
+                    "status": ServiceHealth.HEALTHY,
+                    "pid": 12345,
+                    "uptime": 150.0,
+                    "response_time": 0.05,
+                }
+            if service_name == "fast":
+                return {
+                    "status": ServiceHealth.CRASHED,
+                    "pid": 99999,
+                    "uptime": None,
+                    "response_time": None,
+                }
+            return {
+                "status": ServiceHealth.STOPPED,
+                "pid": None,
+                "uptime": None,
+                "response_time": None,
+            }
+
+        mock_status.side_effect = side_effect
 
         runner = CliRunner()
         result = runner.invoke(cli, ["status"])
 
         assert result.exit_code == 0
+        # Tier names and models present
         assert "standard" in result.output
         assert "fast" in result.output
         assert "big-model" in result.output
         assert "fast-model" in result.output
+        # Ports present
         assert "8000" in result.output
         assert "8001" in result.output
-        assert "healthy" in result.output
+        # Each tier's distinct status must appear — verifies per-tier rendering
+        assert ServiceHealth.HEALTHY in result.output
+        assert ServiceHealth.CRASHED in result.output
 
     @patch("mlx_stack.core.stack_status.get_service_status")
     @patch("mlx_stack.core.stack_status._get_litellm_port", return_value=4000)

{mlx_stack-0.3.2 → mlx_stack-0.3.3}/tests/unit/test_launchd.py

@@ -638,7 +638,13 @@ class TestInstallAgent:
             path, was_reinstall = install_agent("/usr/bin/mlx-stack")
 
         assert was_reinstall is False
-        mock_write.assert_called_once()
+        assert path is mock_plist_path
+
+        # Verify plist data contains correct binary and label
+        plist_data = mock_write.call_args[0][0]
+        assert plist_data["Label"] == "com.mlx-stack.watchdog"
+        assert plist_data["ProgramArguments"][0] == "/usr/bin/mlx-stack"
+        assert "watch" in plist_data["ProgramArguments"]
 
     def test_reinstall(
         self, mlx_stack_home: Path, stack_definition: dict[str, Any]