localcode 0.5.8__tar.gz → 0.5.9__tar.gz

{localcode-0.5.8/src/localcode.egg-info → localcode-0.5.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: localcode
-Version: 0.5.8
+Version: 0.5.9
 Summary: High-performance AI coding on consumer hardware.
 Author: LocalCode contributors
 License-Expression: Apache-2.0
@@ -152,6 +152,8 @@ Powerful, personal AI should work for everyone, on any device, anywhere. That me
 
 To sponsor localcode, [reach out](https://github.com/mjwsolo/localcode).
 
+Installation diagnostics: `localcode doctor`. See [compatibility tests and local hardware verification](docs/src/content/docs/development/verification.md).
+
 ## Contributing
 
 See [CONTRIBUTING.md](CONTRIBUTING.md).

{localcode-0.5.8 → localcode-0.5.9}/README.md

@@ -104,6 +104,8 @@ Powerful, personal AI should work for everyone, on any device, anywhere. That me
 
 To sponsor localcode, [reach out](https://github.com/mjwsolo/localcode).
 
+Installation diagnostics: `localcode doctor`. See [compatibility tests and local hardware verification](docs/src/content/docs/development/verification.md).
+
 ## Contributing
 
 See [CONTRIBUTING.md](CONTRIBUTING.md).

{localcode-0.5.8 → localcode-0.5.9}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "localcode"
-version = "0.5.8"
+version = "0.5.9"
 description = "High-performance AI coding on consumer hardware."
 readme = "README.md"
 requires-python = ">=3.10"

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/__init__.py

@@ -8,4 +8,4 @@ from importlib.metadata import PackageNotFoundError, version as _pkg_version
 try:
     __version__ = _pkg_version("localcode")
 except PackageNotFoundError:  # not installed (e.g. raw source checkout)
-    __version__ = "0.5.8"
+    __version__ = "0.5.9"

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/bin/localcode-ui

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
-name: localcode-0.5.8/src/localcode/bin/localcode-ui
+name: localcode-0.5.9/src/localcode/bin/localcode-ui
 size: 102460976 bytes
 sha256: a09d854109bdb8e9b346777e8fa306afcae9edb32ade0780ccd15a0e15fe3c05

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/bootstrap.py

@@ -719,6 +719,10 @@ def _classify_download_error(e: Exception) -> str:
     et = type(e).__name__
     if "no space" in msg or "disk full" in msg or "enospc" in msg or et == "OSError" and "28" in msg:
         return "disk_full"
+    if "ssl" in msg or "certificate" in msg:
+        return "ssl"
+    if "proxy" in msg or "407" in msg:
+        return "proxy"
     if "401" in msg or "403" in msg or "gated" in msg or "permission" in msg:
         return "auth"
     if "404" in msg or "not found" in msg:
@@ -727,8 +731,6 @@ def _classify_download_error(e: Exception) -> str:
         return "timeout"
     if "connection" in msg or "network" in msg or "dns" in msg or "resolve" in msg:
         return "network"
-    if "ssl" in msg or "certificate" in msg:
-        return "ssl"
     return "unknown"
 
 
@@ -740,8 +742,9 @@ def _format_download_error(category: str, e: Exception) -> str:
         "auth":       f"Hugging Face rejected the request — model may be gated or require a token.\n[{raw}]",
         "not_found":  f"Model file not found on Hugging Face — the catalog entry's filename may be stale.\n[{raw}]",
         "timeout":    f"Download timed out. Check your connection and retry.\n[{raw}]",
+        "proxy":      "Proxy blocked the download. Ask IT for the approved proxy configuration; do not share proxy credentials.",
         "network":    f"Network error during download. Check your connection and retry.\n[{raw}]",
-        "ssl":        f"SSL/TLS error during download — your system trust store may be out of date.\n[{raw}]",
+        "ssl":        f"SSL/TLS error during download — ask your IT team to configure the corporate CA in your Python trust store. Do not disable TLS verification.\n[{raw}]",
         "unknown":    f"Model download failed: {raw}",
     }
     return by_cat.get(category, by_cat["unknown"])

localcode-0.5.9/src/localcode/diagnostics.py

@@ -0,0 +1,58 @@
+"""Offline installation/platform diagnostics; never prints credential values."""
+from __future__ import annotations
+
+import json
+import os
+import platform
+import shutil
+import subprocess
+import sys
+from importlib.metadata import version
+
+
+def sysctl(name: str) -> str:
+    try:
+        return subprocess.check_output(['sysctl', '-n', name], stderr=subprocess.DEVNULL,
+                                       text=True, timeout=3).strip()
+    except (OSError, subprocess.SubprocessError):
+        return ''
+
+
+def platform_problem() -> str | None:
+    if platform.system() != 'Darwin':
+        return 'This release requires an Apple silicon Mac with macOS 13 or newer.'
+    if platform.machine().lower() != 'arm64':
+        if sysctl('sysctl.proc_translated') == '1' or sysctl('hw.optional.arm64') == '1':
+            return ('Python is running under Rosetta. Open a native Apple silicon terminal, '
+                    'install an arm64 Python or uv, then reinstall LocalCode there.')
+        return 'Intel Macs are not supported by the bundled Apple silicon binaries.'
+    release = platform.mac_ver()[0]
+    if not release or not release.split('.')[0].isdigit():
+        return 'Could not determine macOS version. LocalCode requires macOS 13 or newer.'
+    if int(release.split('.')[0]) < 13:
+        return f'macOS {release} is too old. Upgrade to macOS 13 or newer.'
+    return None
+
+
+def report() -> dict:
+    from localcode import __version__
+    try:
+        installed = version('localcode')
+    except Exception:
+        installed = __version__
+    return {
+        'version': installed, 'python': sys.version.split()[0],
+        'python_executable': sys.executable, 'localcode_on_path': shutil.which('localcode'),
+        'system': platform.system(), 'architecture': platform.machine(),
+        'macos': platform.mac_ver()[0], 'chip': sysctl('machdep.cpu.brand_string') if platform.system() == 'Darwin' else '',
+        'ram_bytes': sysctl('hw.memsize') if platform.system() == 'Darwin' else '',
+        'platform_problem': platform_problem(),
+        'network_configuration_present': {name: bool(os.environ.get(name)) for name in (
+            'HTTPS_PROXY', 'HTTP_PROXY', 'https_proxy', 'http_proxy', 'SSL_CERT_FILE',
+            'REQUESTS_CA_BUNDLE', 'PIP_INDEX_URL', 'UV_INDEX_URL', 'UV_DEFAULT_INDEX')},
+    }
+
+
+def main() -> int:
+    print(json.dumps(report(), indent=2))
+    return 0

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/entrypoint.py

@@ -11,7 +11,7 @@ def build_parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(
         prog="localcode",
         description="LocalCode — AI coding assistant running entirely on your machine",
-        epilog="Other commands: localcode api shows connection details; localcode decide asks a typed question.",
+        epilog="Other commands: localcode doctor diagnoses installation; localcode api shows connection details; localcode decide asks a typed question.",
     )
     parser.add_argument("--version", action="store_true", help="Print the version and exit")
     parser.add_argument("--mode", choices=["chat", "decisions"], default="chat", help="Start in normal chat or typed decision mode")
@@ -26,9 +26,12 @@ def build_parser() -> argparse.ArgumentParser:
 
 def _unsupported_platform_message() -> str:
     import platform
+    from .diagnostics import platform_problem
+    detail = platform_problem() or ""
     return (
         f"localcode: this release runs on Apple silicon Macs (macOS 13 or newer); "
         f"this machine reports {platform.system()} {platform.machine()}.\n"
+        f"{detail}\n"
         "Earlier releases installed on other platforms but bundled an Apple silicon "
         "model server, so they did not work there either."
     )
@@ -36,6 +39,9 @@ def _unsupported_platform_message() -> str:
 
 def main(argv: list[str] | None = None) -> None:
     argv = list(sys.argv[1:] if argv is None else argv)
+    if argv == ["doctor"]:
+        from .diagnostics import main as doctor_main
+        sys.exit(doctor_main())
     if argv and argv[0] == "decide":
         from .decision import main as decide_main
         sys.exit(decide_main(argv[1:]))

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/hf_quants.py

@@ -100,7 +100,7 @@ def _quant_from_entry(entry: dict) -> Quant | None:
         return None
     # Vision projectors and speculative drafters (MTP heads, DFlash) are
     # sidecars, not pickable models.
-    if lowered.startswith(("mmproj", "mtp-", "mtp_", "dflash-", "dflash_")):
+    if lowered.startswith(("mmproj", "mtp-", "mtp_", "dflash")):
         return None
     label = _parse_label(path)
     # Real weight quants carry a quant code (Q4_K_M, IQ3_S, BF16, …). Anything

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/models_catalog.py

@@ -84,12 +84,15 @@ DRAFTER_QWEN36 = Drafter(
     sha256="339030c9dad4f410c7ef2d6cbd6e38251a84fabb59c44db992538c2451a0efee",
     revision="baec3ebee244827cda0f4557eafa8b28f7545fa6",
 )
+# DFlash 2 from the DFlash authors (z-lab mirror of incoai). Measured against
+# ggml-org's DFlash v1 on the same Q4_K_XL Muse, greedy, 600 tokens: accept
+# rate 0.74 vs 0.60, decode +12%, identical output, vision unaffected.
 DRAFTER_MUSE = Drafter(
-    filename="dflash-Muse-Glimmer-30B-Q4_0.gguf", hf_repo="ggml-org/Muse-Glimmer-30B-GGUF",
-    hf_filename="dflash-Muse-Glimmer-30B-Q4_0.gguf", spec_type="draft-dflash",
-    size_gb=1.45, size_bytes=1451096896,
-    sha256="048242c69254627fc11226c9f68dc34fbc82eac1c026ef148c9dbd2f90686db2",
-    revision="c286f180ef572621f5395b55da2d3e3d53d5972e",
+    filename="dflash2-Muse-Glimmer-30B-Q4_K_M.gguf", hf_repo="z-lab/Muse-Glimmer-30B-DFlash2-GGUF",
+    hf_filename="Muse-Glimmer-30B-DFlash2-Q4_K_M.gguf", spec_type="draft-dflash",
+    size_gb=1.65, size_bytes=1645657280,
+    sha256="93dbfb6f88e4645dec1347cf93f9d6fc80b90d413038722385b2a8e53565c949",
+    revision="880882627431093d99d3b2368efb4a6fcf12d4cb",
 )
 DRAFTERS: list[Drafter] = [DRAFTER_GEMMA_12B, DRAFTER_GEMMA_26B, DRAFTER_QWEN36, DRAFTER_MUSE]
 

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/ui/__init__.py

@@ -51,7 +51,8 @@ def ui_binary_path() -> Path | None:
 
 
 def platform_supported() -> bool:
-    return platform.system() == "Darwin" and platform.machine().lower() == "arm64"
+    from localcode.diagnostics import platform_problem
+    return platform_problem() is None
 
 
 def plugin_path() -> Path:

{localcode-0.5.8 → localcode-0.5.9}/src/localcode/ui/server_cmd.py

@@ -18,6 +18,7 @@ from localcode.runtime import LocalCodeRuntimeGateway
 def server_command(gguf: str, port: int, alias: str | None = None) -> list[str]:
     cfg = load_config()
     g = LocalCodeRuntimeGateway(cfg.runtime)
+    validate_model_memory(gguf, g._system_ram_gb())
     cmd = g.llama_server_command(gguf, port)
     cmd[1:1] = ["--host", "127.0.0.1"]
     # Any web page can reach a localhost port; the key makes the model and its
@@ -114,6 +115,33 @@ def context_size(gguf: str) -> int:
     return int(cmd[cmd.index("--ctx-size") + 1])
 
 
+def validate_model_memory(gguf: str, ram_gb: float) -> None:
+    """Reject definite over-capacity loads; passing is not a fit guarantee.
+
+    Include installed projector/drafter weights and OS reserve. Actual KV and
+    Metal workspace peaks are measured by the local hardware gate.
+    """
+    paths = {Path(gguf)}
+    projector = mmproj_for(gguf)
+    if projector:
+        paths.add(projector)
+    choice = catalog_choice(gguf)
+    drafter = getattr(choice, "drafter", None)
+    if drafter:
+        paths.add(Path(drafter.local_path))
+    weights = sum(p.stat().st_size for p in paths if p.is_file())
+    total = int(ram_gb * 1024 ** 3)
+    if total <= 0:
+        raise ValueError("Could not determine RAM; run `localcode doctor` before loading a model.")
+    reserve = max(3 * 1024 ** 3, int(total * 0.15))
+    if weights and weights + reserve >= total:
+        raise ValueError(
+            f"Insufficient RAM: model weights require {weights / 1024**3:.1f} GiB "
+            f"on this {ram_gb:g} GiB Mac before KV cache and Metal workspace. "
+            "Choose a smaller model or quantization in /models; close other apps."
+        )
+
+
 if __name__ == "__main__":
     if sys.argv[1] == "--ctx":
         print(context_size(sys.argv[2]))

{localcode-0.5.8 → localcode-0.5.9/src/localcode.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: localcode
-Version: 0.5.8
+Version: 0.5.9
 Summary: High-performance AI coding on consumer hardware.
 Author: LocalCode contributors
 License-Expression: Apache-2.0
@@ -152,6 +152,8 @@ Powerful, personal AI should work for everyone, on any device, anywhere. That me
 
 To sponsor localcode, [reach out](https://github.com/mjwsolo/localcode).
 
+Installation diagnostics: `localcode doctor`. See [compatibility tests and local hardware verification](docs/src/content/docs/development/verification.md).
+
 ## Contributing
 
 See [CONTRIBUTING.md](CONTRIBUTING.md).

{localcode-0.5.8 → localcode-0.5.9}/src/localcode.egg-info/SOURCES.txt

@@ -1154,6 +1154,7 @@ src/localcode/_subproc_env.py
 src/localcode/bootstrap.py
 src/localcode/config.py
 src/localcode/decision.py
+src/localcode/diagnostics.py
 src/localcode/entrypoint.py
 src/localcode/errors.py
 src/localcode/events.py
@@ -1208,6 +1209,7 @@ src/localcode/ui/plugin/modes.ts
 tests/test_architecture.py
 tests/test_bootstrap.py
 tests/test_cancel_download.py
+tests/test_compatibility.py
 tests/test_comprehensive_bg_download.py
 tests/test_comprehensive_binary.py
 tests/test_comprehensive_cli.py

{localcode-0.5.8 → localcode-0.5.9}/tests/test_architecture.py

@@ -360,6 +360,7 @@ def test_agent_submodules_declare_all():
 # allowlist with a reason than silently let printfs leak into the
 # agent loop.
 PRINT_ALLOWLIST = {
+    "diagnostics.py",  # Offline `localcode doctor` CLI emits its JSON report.
     "launch.py",  # UI launcher: prints to the terminal BEFORE the UI runtime starts (no event bus yet)
     "ports.py",  # UI launcher: prints to the terminal BEFORE the UI runtime starts (no event bus yet)
     "server_cmd.py",  # UI launcher: prints to the terminal BEFORE the UI runtime starts (no event bus yet)

localcode-0.5.9/tests/test_compatibility.py

@@ -0,0 +1,112 @@
+"""Platform/enterprise errors and the default shipping server command."""
+from pathlib import Path
+
+import pytest
+
+from localcode import bootstrap, diagnostics, models_catalog
+from localcode.config import AppConfig, RuntimeConfig, UIConfig
+from localcode.runtime import LocalCodeRuntimeGateway
+from localcode.ui import server_cmd
+
+
+@pytest.mark.parametrize('system,arch,mac,translated,fragment', [
+    ('Linux', 'x86_64', '', '', 'Apple silicon'),
+    ('Darwin', 'x86_64', '14.0', '0', 'Intel'),
+    ('Darwin', 'x86_64', '14.0', '1', 'Rosetta'),
+    ('Darwin', 'arm64', '12.7', '', 'too old'),
+    ('Darwin', 'arm64', '', '', 'determine'),
+    ('Darwin', 'arm64', '13.0', '', None),
+    ('Darwin', 'arm64', '26.0', '', None),
+])
+def test_platform_diagnosis(monkeypatch, system, arch, mac, translated, fragment):
+    monkeypatch.setattr(diagnostics.platform, 'system', lambda: system)
+    monkeypatch.setattr(diagnostics.platform, 'machine', lambda: arch)
+    monkeypatch.setattr(diagnostics.platform, 'mac_ver', lambda: (mac, (), ''))
+    monkeypatch.setattr(diagnostics, 'sysctl', lambda _: translated)
+    result = diagnostics.platform_problem()
+    assert (result is None) if fragment is None else fragment in result
+
+
+def test_doctor_does_not_expose_proxy_credentials(monkeypatch):
+    monkeypatch.setenv('HTTPS_PROXY', 'https://username:secret@example.com')
+    assert 'secret' not in str(diagnostics.report())
+    assert diagnostics.report()['network_configuration_present']['HTTPS_PROXY']
+
+
+@pytest.mark.parametrize('error,category,hint', [
+    ('connection failed: certificate verify failed', 'ssl', 'IT'),
+    ('407 Proxy Authentication Required https://user:secret@proxy', 'proxy', 'IT'),
+    ('No space left on device', 'disk_full', 'space'),
+    ('403 Forbidden', 'auth', 'token'),
+])
+def test_download_diagnosis(error, category, hint):
+    e = RuntimeError(error)
+    assert bootstrap._classify_download_error(e) == category
+    message = bootstrap._format_download_error(category, e)
+    assert hint in message
+    if category == 'proxy':
+        assert 'secret' not in message
+
+
+RAM = [8, 16, 18, 24, 32, 36, 48, 64, 96, 128, 192]
+
+
+@pytest.mark.parametrize('ram', RAM)
+@pytest.mark.parametrize('choice', models_catalog.CHOICES, ids=lambda c: c.key)
+def test_shipping_command_model_ram_matrix(monkeypatch, tmp_path, ram, choice):
+    config = AppConfig(runtime=RuntimeConfig(), ui=UIConfig())
+    config.runtime.llama_cpp_binary = str(tmp_path / 'llama-server')
+    Path(config.runtime.llama_cpp_binary).touch()
+    config.runtime.quant_preset = 'fastest'
+    config.runtime.kv_cache_type_v = 'turbo4'
+    config.runtime.laptop_26b_runtime_mode = 'turbo'
+    config.runtime.max_context_chars = 400000
+    monkeypatch.setattr(server_cmd, 'load_config', lambda: config)
+    monkeypatch.setattr(LocalCodeRuntimeGateway, '_system_ram_gb', lambda _: ram)
+    monkeypatch.setenv('LOCALCODE_SERVER_KEY', 'test-only-key')
+    monkeypatch.setenv('LOCALCODE_AGENT_RUN_DIR', str(tmp_path / 'run'))
+    monkeypatch.setenv('LOCALCODE_PROMPT_WARMUP', '0')
+    cmd = server_cmd.server_command(str(tmp_path / choice.filename), 8123, choice.key)
+    assert cmd[cmd.index('--host') + 1] == '127.0.0.1'
+    assert cmd[cmd.index('--port') + 1] == '8123'
+    assert cmd[cmd.index('--alias') + 1] == choice.key
+    assert 2048 <= int(cmd[cmd.index('--ctx-size') + 1]) <= LocalCodeRuntimeGateway._ram_ctx_ceiling(ram)
+    assert '--api-key' in cmd
+
+
+def test_oversize_model_rejected_before_spawn(tmp_path, monkeypatch):
+    model = tmp_path / 'large.gguf'
+    with model.open('wb') as f:
+        f.truncate(7 * 1024 ** 3)  # sparse file, no 7 GB allocation
+    monkeypatch.setattr(server_cmd, 'mmproj_for', lambda _: None)
+    monkeypatch.setattr(server_cmd, 'catalog_choice', lambda _: None)
+    with pytest.raises(ValueError, match='Insufficient RAM'):
+        server_cmd.validate_model_memory(str(model), 8)
+    server_cmd.validate_model_memory(str(model), 16)
+
+
+def test_memory_guard_counts_sidecars(monkeypatch, tmp_path):
+    from types import SimpleNamespace
+    paths = [tmp_path / name for name in ('model.gguf', 'mmproj.gguf', 'draft.gguf')]
+    for path in paths:
+        with path.open('wb') as f:
+            f.truncate(2 * 1024**3)
+    monkeypatch.setattr(server_cmd, 'mmproj_for', lambda _: paths[1])
+    monkeypatch.setattr(server_cmd, 'catalog_choice', lambda _: SimpleNamespace(drafter=SimpleNamespace(local_path=paths[2])))
+    with pytest.raises(ValueError, match='Insufficient RAM'):
+        server_cmd.validate_model_memory(str(paths[0]), 8)
+
+
+def test_rejected_load_keeps_existing_server(monkeypatch, tmp_path):
+    from localcode.ui.supervisor import Supervisor
+    sup = object.__new__(Supervisor)
+    sup.models_dir = tmp_path
+    sup.port = 8000
+    stopped = []
+    sup.stop = lambda: stopped.append(True)
+    def refuse(*args):
+        raise ValueError('Insufficient RAM')
+    monkeypatch.setattr(server_cmd, 'server_command', refuse)
+    with pytest.raises(ValueError, match='Insufficient RAM'):
+        sup.start('too-large')
+    assert not stopped

{localcode-0.5.8 → localcode-0.5.9}/tests/test_download.py

@@ -13,10 +13,9 @@ sys.path.insert(0, str(Path(__file__).parent.parent / "src"))
 from localcode.bootstrap import _download_parallel, download_model, get_model_path
 from localcode.models_catalog import CHOICES
 
-# Download/socket integration tests are opt-in: restricted CI sandboxes cannot
-# resolve Hugging Face or bind loopback sockets. Unit coverage for download
-# planning lives elsewhere and remains in the default suite.
-pytestmark = pytest.mark.skipif(
+# Only external Hugging Face checks are opt-in. Loopback failure/integrity
+# checks run in ordinary CI.
+external_network = pytest.mark.skipif(
     os.environ.get("LOCALCODE_RUN_NETWORK_TESTS") != "1",
     reason="set LOCALCODE_RUN_NETWORK_TESTS=1 to run network/socket integration tests",
 )
@@ -29,6 +28,7 @@ _MODEL_URL = f"https://huggingface.co/{_CATALOG_ENTRY.hf_repo}/resolve/main/{_CA
 
 
 # ── Test 1: HEAD request to HuggingFace actually works ──
+@external_network
 def test_hf_head_request():
     """Verify HF returns Content-Length and Accept-Ranges."""
     import urllib.request
@@ -48,6 +48,7 @@ def test_hf_head_request():
 
 
 # ── Test 2: Range request actually works (download first 1MB) ──
+@external_network
 def test_hf_range_request():
     """Verify partial content download works."""
     import urllib.request
@@ -68,6 +69,7 @@ def test_hf_range_request():
 
 
 # ── Test 3: Parallel download on a small real file ──
+@external_network
 def test_parallel_small_file():
     """Download a small file from HF in parallel to verify chunk assembly."""
     import urllib.request
@@ -96,6 +98,7 @@ def test_parallel_small_file():
 
 
 # ── Test 4: Parallel download with 8 threads on first 10MB of model ──
+@external_network
 def test_parallel_chunked_model_fragment():
     """Download first 10MB of the actual model using range requests to verify chunking."""
     import urllib.request
@@ -154,6 +157,8 @@ def test_parallel_chunked_model_fragment():
             print("  ✓ PASS — chunks assembled correctly")
         finally:
             server.shutdown()
+            server.server_close()
+            server_thread.join(timeout=5)
 
 
 # ── Test 5: Fallback when server doesn't support ranges ──
@@ -192,6 +197,8 @@ def test_fallback_no_ranges():
             print("  ✓ PASS")
         finally:
             server.shutdown()
+            server.server_close()
+            server_thread.join(timeout=5)
 
 
 # ── Test 6: Error handling — bad URL ──
@@ -199,14 +206,9 @@ def test_bad_url():
     """Verify clean error on unreachable URL."""
     with tempfile.TemporaryDirectory() as tmpdir:
         dest = Path(tmpdir) / "bad.bin"
-        try:
+        with pytest.raises(Exception):
             _download_parallel("http://127.0.0.1:1/nonexistent", dest, num_threads=4)
-            print("  ✗ FAIL — should have raised")
-            assert False
-        except Exception as e:
-            print(f"  Error (expected): {type(e).__name__}: {str(e)[:80]}")
-            assert not dest.exists() or dest.stat().st_size == 0, "Partial file should be cleaned up"
-            print("  ✓ PASS")
+        assert not dest.exists() or dest.stat().st_size == 0
 
 
 # ── Test 7: Partial failure — one chunk fails ──
@@ -259,24 +261,20 @@ def test_partial_chunk_failure():
             print("  ✓ PASS")
         finally:
             server.shutdown()
+            server.server_close()
+            server_thread.join(timeout=5)
 
 
 # ── Test 8: download_model skips if file exists ──
-def test_download_model_skip_existing():
-    """Verify download_model returns immediately if model already exists."""
-    with tempfile.TemporaryDirectory() as tmpdir:
-        model_dir = Path(tmpdir) / "models"
-        model_dir.mkdir()
-        fake_model = model_dir / "gemma-4-26B-A4B-it-UD-IQ3_S.gguf"
-        fake_model.write_bytes(b"fake model data")
-
-        with patch("localcode.bootstrap.Path.home", return_value=Path(tmpdir) / "fake_home"):
-            # This won't match because the path construction uses home()
-            pass
-
-        # Simpler: just test get_model_path with the real path
-        print(f"  Existing model check works: get_model_path returns None for missing = {get_model_path() is not None or 'correct'}")
-        print("  ✓ PASS (skip-existing logic verified by inspection)")
+def test_download_model_skip_existing(tmp_path):
+    """An existing verified model returns without network work."""
+    model = tmp_path / "existing.gguf"
+    model.write_bytes(b"fixture")
+    with patch("localcode.bootstrap.get_model_path", return_value=model), \
+         patch("localcode.bootstrap._download_parallel") as download:
+        ok, path = download_model(_CATALOG_ENTRY)
+    assert ok and path == str(model)
+    download.assert_not_called()
 
 
 # ── Test 9: Verify concurrent file writes don't corrupt ──
@@ -329,6 +327,8 @@ def test_concurrent_writes_integrity():
             print("  ✓ PASS")
         finally:
             server.shutdown()
+            server.server_close()
+            server_thread.join(timeout=5)
 
 
 if __name__ == "__main__":

{localcode-0.5.8 → localcode-0.5.9}/tests/test_models_catalog.py

@@ -323,7 +323,7 @@ def test_drafters_are_pinned_and_attached_to_every_quant_of_their_model():
         "gemma": "mtp-gemma-4-26B-A4B-it-Q8_0.gguf", "gemma-q8": "mtp-gemma-4-26B-A4B-it-Q8_0.gguf",
         "gemma-12b": "mtp-gemma-4-12b-it-Q8_0.gguf", "gemma-12b-bf16": "mtp-gemma-4-12b-it-Q8_0.gguf",
         "qwen": "dflash-Qwen3.6-35B-A3B-Q8_0.gguf", "qwen-q8": "dflash-Qwen3.6-35B-A3B-Q8_0.gguf",
-        "muse-glimmer": "dflash-Muse-Glimmer-30B-Q4_0.gguf",
+        "muse-glimmer": "dflash2-Muse-Glimmer-30B-Q4_K_M.gguf",
     }
     # Qwen 3.8 carries its head in the GGUF; the others have no vendor drafter.
     assert all(c.drafter is None for c in CHOICES if c.key in {"qwen38", "qwen38-q8", "kolibri", "openjev", "north-mini-code", "diffusiongemma"})

{localcode-0.5.8 → localcode-0.5.9}/tests/test_runtime.py

@@ -179,7 +179,8 @@ class TestLlamaServerCommand:
         cmd = gw.llama_server_command("/path/model.gguf")
         assert "--model-draft" in cmd
         assert "/path/draft.gguf" in cmd
-        assert "--draft-max" in cmd
+        assert "--spec-draft-n-max" in cmd
+        assert cmd[cmd.index("--spec-draft-n-max") + 1] == "64"
 
     def test_lookup_cache_adds_flag(self) -> None:
         gw = self._make_gw(llama_cpp_lookup_cache=True)
@@ -219,11 +220,13 @@ class TestLlamaServerCommand:
         ub_idx = cmd.index("-ub")
         assert cmd[ub_idx + 1] == "128"
 
-    def test_turbo_mode_defaults_to_large_batch_for_non_qwen(self) -> None:
+    @pytest.mark.parametrize("system,expected", [("Darwin", "2048"), ("Linux", "128")])
+    def test_turbo_mode_defaults_to_platform_batch_for_non_qwen(self, system, expected) -> None:
         gw = self._make_gw(laptop_26b_runtime_mode="turbo", llama_cpp_batch_size=-1)
-        cmd = gw.llama_server_command("/path/gemma-model.gguf")
+        with patch("platform.system", return_value=system):
+            cmd = gw.llama_server_command("/path/gemma-model.gguf")
         b_idx = cmd.index("-b")
-        assert cmd[b_idx + 1] == "2048"
+        assert cmd[b_idx + 1] == expected
 
     def test_large_qwen_on_16gb_uses_safer_batch(self) -> None:
         gw = self._make_gw(laptop_26b_runtime_mode="turbo", llama_cpp_batch_size=-1)