comfy-env 0.1.20__py3-none-any.whl → 0.1.22__py3-none-any.whl

comfy_env/__init__.py

@@ -108,7 +108,6 @@ from .isolation import (
     # Workers
     Worker,
     WorkerError,
-    MPWorker,
     SubprocessWorker,
     # Tensor utilities
     TensorKeeper,
@@ -168,7 +167,6 @@ __all__ = [
     # Workers
     "Worker",
     "WorkerError",
-    "MPWorker",
     "SubprocessWorker",
     "TensorKeeper",
 ]

comfy_env/cli.py

@@ -44,6 +44,9 @@ def main(args: Optional[List[str]] = None) -> int:
     p.add_argument("--config", "-c", type=str, help="Config path")
     p.add_argument("--dry-run", action="store_true", help="Preview only")
 
+    # cleanup
+    sub.add_parser("cleanup", help="Remove orphaned environments")
+
     parsed = parser.parse_args(args)
     if not parsed.command:
         parser.print_help()
@@ -52,6 +55,7 @@ def main(args: Optional[List[str]] = None) -> int:
     commands = {
         "init": cmd_init, "generate": cmd_generate, "install": cmd_install,
         "info": cmd_info, "doctor": cmd_doctor, "apt-install": cmd_apt_install,
+        "cleanup": cmd_cleanup,
     }
 
     try:
@@ -237,5 +241,16 @@ def cmd_apt_install(args) -> int:
     return result.returncode
 
 
+def cmd_cleanup(args) -> int:
+    from .environment.cache import cleanup_orphaned_envs
+    print("Cleaning orphaned environments...")
+    cleaned = cleanup_orphaned_envs()
+    if cleaned:
+        print(f"Removed {cleaned} orphaned environment(s)")
+    else:
+        print("No orphaned environments found")
+    return 0
+
+
 if __name__ == "__main__":
     sys.exit(main())

comfy_env/install.py

@@ -116,9 +116,12 @@ def _install_via_pixi(cfg: ComfyEnvConfig, node_dir: Path, log: Callable[[str],
     from .packages.toml_generator import write_pixi_toml
     from .packages.cuda_wheels import get_wheel_url, CUDA_TORCH_MAP
     from .detection import get_recommended_cuda_version
-    from .environment.cache import get_central_env_path, write_marker, write_env_metadata, MARKER_FILE, get_cache_dir
+    from .environment.cache import get_central_env_path, write_marker, write_env_metadata, MARKER_FILE, get_cache_dir, cleanup_orphaned_envs
     import shutil, subprocess, sys
 
+    # Clean up orphaned environments before installing
+    cleanup_orphaned_envs(log)
+
     deps = cfg.pixi_passthrough.get("dependencies", {})
     pypi_deps = cfg.pixi_passthrough.get("pypi-dependencies", {})
     if not cfg.cuda_packages and not deps and not pypi_deps:

comfy_env/isolation/__init__.py

@@ -11,7 +11,6 @@ from .wrap import (
 from .workers import (
     Worker,
     WorkerError,
-    MPWorker,
     SubprocessWorker,
 )
 from .tensor_utils import (
@@ -29,7 +28,6 @@ __all__ = [
     # Workers
     "Worker",
     "WorkerError",
-    "MPWorker",
     "SubprocessWorker",
     # Tensor utilities
     "TensorKeeper",

comfy_env/isolation/workers/__init__.py

@@ -1,16 +1,14 @@
 """
 Workers - Process isolation implementations.
 
-Provides multiprocessing and subprocess-based workers for isolated execution.
+Provides subprocess-based workers for isolated execution.
 """
 
 from .base import Worker, WorkerError
-from .mp import MPWorker
 from .subprocess import SubprocessWorker
 
 __all__ = [
     "Worker",
     "WorkerError",
-    "MPWorker",
     "SubprocessWorker",
 ]

comfy_env/isolation/workers/base.py

@@ -16,7 +16,7 @@ class Worker(ABC):
 
     Workers should be used as context managers when possible:
 
-        with MPWorker() as worker:
+        with SubprocessWorker(python="/path/to/venv/bin/python") as worker:
             result = worker.call(my_func, arg1, arg2)
     """
 

comfy_env/isolation/workers/subprocess.py

@@ -297,6 +297,45 @@ def _to_shm(obj, registry, visited=None):
     return obj
 
 
+def _deserialize_tensor_ref(data):
+    """Deserialize tensor from PyTorch shared memory (TensorRef format)."""
+    import torch
+    import torch.multiprocessing.reductions as reductions
+
+    dtype_str = data["dtype"]
+    dtype = getattr(torch, dtype_str.split(".")[-1])
+
+    manager_path = data["manager_path"]
+    storage_key = data["storage_key"]
+    storage_size = data["storage_size"]
+
+    # Encode to bytes if needed
+    if isinstance(manager_path, str):
+        manager_path = manager_path.encode("utf-8")
+    if isinstance(storage_key, str):
+        storage_key = storage_key.encode("utf-8")
+
+    # Rebuild storage from shared memory file
+    rebuilt_storage = reductions.rebuild_storage_filename(
+        torch.UntypedStorage, manager_path, storage_key, storage_size
+    )
+
+    # Wrap in TypedStorage
+    typed_storage = torch.storage.TypedStorage(
+        wrap_storage=rebuilt_storage, dtype=dtype, _internal=True
+    )
+
+    # Rebuild tensor
+    metadata = (
+        data["tensor_offset"],
+        tuple(data["tensor_size"]),
+        tuple(data["tensor_stride"]),
+        data["requires_grad"],
+    )
+    tensor = reductions.rebuild_tensor(torch.Tensor, typed_storage, metadata)
+    return tensor
+
+
 def _from_shm(obj, unlink=True):
     """Reconstruct object from shared memory metadata."""
     if not isinstance(obj, dict):
@@ -304,7 +343,15 @@ def _from_shm(obj, unlink=True):
             return [_from_shm(v, unlink) for v in obj]
         return obj
 
-    # numpy array (or tensor that was converted to numpy)
+    # TensorRef -> use PyTorch's native deserialization (new format)
+    if obj.get("__type__") == "TensorRef":
+        tensor = _deserialize_tensor_ref(obj)
+        # Convert back to numpy if it was originally numpy
+        if obj.get("__was_numpy__"):
+            return tensor.numpy()
+        return tensor
+
+    # numpy array (or tensor that was converted to numpy) - legacy format
     if "__shm_np__" in obj:
         block = shm.SharedMemory(name=obj["__shm_np__"])
         arr = np.ndarray(tuple(obj["shape"]), dtype=np.dtype(obj["dtype"]), buffer=block.buf).copy()
@@ -362,10 +409,10 @@ def _serialize_for_ipc(obj, visited=None):
     if obj_id in visited:
         return visited[obj_id]  # Return cached serialized result
 
-    # Handle Path objects - convert to string for JSON serialization
+    # Handle Path objects - mark for reconstruction
     from pathlib import PurePath
     if isinstance(obj, PurePath):
-        return str(obj)
+        return {"__path__": str(obj)}
 
     # Check if this is a custom object with broken module path
     if (hasattr(obj, '__dict__') and
@@ -434,6 +481,8 @@ import socket
 import struct
 import traceback
 import faulthandler
+import collections
+import time
 from types import SimpleNamespace
 
 # Enable faulthandler to dump traceback on SIGSEGV/SIGABRT/etc
@@ -533,6 +582,34 @@ if sys.platform == "win32":
 from multiprocessing import shared_memory as shm
 import numpy as np
 
+# Set PyTorch to use file_system sharing (uses /dev/shm, no resource_tracker)
+try:
+    import torch
+    import torch.multiprocessing as mp
+    mp.set_sharing_strategy("file_system")
+    wlog("[worker] PyTorch sharing strategy set to file_system")
+except ImportError:
+    wlog("[worker] PyTorch not available")
+
+
+# Tensor keeper - holds tensor references to prevent GC before parent reads shared memory
+class TensorKeeper:
+    """Keep tensors alive for a retention period to prevent shared memory deletion."""
+    def __init__(self, retention_seconds=30.0):
+        self.retention_seconds = retention_seconds
+        self._keeper = collections.deque()
+        self._lock = threading.Lock()
+
+    def keep(self, t):
+        now = time.time()
+        with self._lock:
+            self._keeper.append((now, t))
+            # Cleanup old entries
+            while self._keeper and now - self._keeper[0][0] > self.retention_seconds:
+                self._keeper.popleft()
+
+_tensor_keeper = TensorKeeper()
+
 
 def _prepare_trimesh_for_pickle(mesh):
     """
@@ -548,6 +625,43 @@ def _prepare_trimesh_for_pickle(mesh):
     return mesh
 
 
+def _serialize_tensor_native(t, registry):
+    """Serialize tensor using PyTorch's native shared memory (no resource_tracker)."""
+    import torch
+    import torch.multiprocessing.reductions as reductions
+
+    # Keep tensor alive until parent reads it
+    _tensor_keeper.keep(t)
+
+    # Put tensor in shared memory via PyTorch's manager
+    if not t.is_shared():
+        t.share_memory_()
+
+    storage = t.untyped_storage()
+    sfunc, sargs = reductions.reduce_storage(storage)
+
+    if sfunc.__name__ == "rebuild_storage_filename":
+        # sargs: (cls, manager_path, storage_key, size)
+        return {
+            "__type__": "TensorRef",
+            "strategy": "file_system",
+            "manager_path": sargs[1].decode("utf-8") if isinstance(sargs[1], bytes) else sargs[1],
+            "storage_key": sargs[2].decode("utf-8") if isinstance(sargs[2], bytes) else sargs[2],
+            "storage_size": sargs[3],
+            "dtype": str(t.dtype),
+            "tensor_size": list(t.size()),
+            "tensor_stride": list(t.stride()),
+            "tensor_offset": t.storage_offset(),
+            "requires_grad": t.requires_grad,
+        }
+    else:
+        # Fallback: force file_system strategy
+        import torch.multiprocessing as mp
+        mp.set_sharing_strategy("file_system")
+        t.share_memory_()
+        return _serialize_tensor_native(t, registry)
+
+
 def _to_shm(obj, registry, visited=None):
     """Serialize to shared memory. Returns JSON-safe metadata."""
     if visited is None:
@@ -557,22 +671,26 @@ def _to_shm(obj, registry, visited=None):
         return visited[obj_id]
     t = type(obj).__name__
 
-    if t == 'ndarray':
-        arr = np.ascontiguousarray(obj)
-        block = shm.SharedMemory(create=True, size=arr.nbytes)
-        np.ndarray(arr.shape, arr.dtype, buffer=block.buf)[:] = arr
-        registry.append(block)
-        result = {"__shm_np__": block.name, "shape": list(arr.shape), "dtype": str(arr.dtype)}
+    # Tensor -> use PyTorch's native shared memory (bypasses resource_tracker)
+    if t == 'Tensor':
+        import torch
+        tensor = obj.detach().cpu().contiguous()
+        result = _serialize_tensor_native(tensor, registry)
         visited[obj_id] = result
         return result
 
-    if t == 'Tensor':
-        arr = obj.detach().cpu().numpy()
-        result = _to_shm(arr, registry, visited)
-        result["__was_tensor__"] = True
+    # ndarray -> convert to tensor, use PyTorch's native shared memory
+    if t == 'ndarray':
+        import torch
+        arr = np.ascontiguousarray(obj)
+        tensor = torch.from_numpy(arr)
+        result = _serialize_tensor_native(tensor, registry)
+        result["__was_numpy__"] = True
+        result["numpy_dtype"] = str(arr.dtype)
+        visited[obj_id] = result
         return result
 
-    # trimesh.Trimesh -> pickle -> shared memory (preserves visual, metadata, normals)
+    # trimesh.Trimesh -> pickle -> shared memory
     if t == 'Trimesh':
         import pickle
         obj = _prepare_trimesh_for_pickle(obj)
@@ -601,12 +719,62 @@ def _to_shm(obj, registry, visited=None):
 
     return obj
 
+
+def _deserialize_tensor_native(data):
+    """Deserialize tensor from PyTorch shared memory."""
+    import torch
+    import torch.multiprocessing.reductions as reductions
+
+    dtype_str = data["dtype"]
+    dtype = getattr(torch, dtype_str.split(".")[-1])
+
+    manager_path = data["manager_path"]
+    storage_key = data["storage_key"]
+    storage_size = data["storage_size"]
+
+    # Encode to bytes if needed
+    if isinstance(manager_path, str):
+        manager_path = manager_path.encode("utf-8")
+    if isinstance(storage_key, str):
+        storage_key = storage_key.encode("utf-8")
+
+    # Rebuild storage from shared memory file
+    rebuilt_storage = reductions.rebuild_storage_filename(
+        torch.UntypedStorage, manager_path, storage_key, storage_size
+    )
+
+    # Wrap in TypedStorage
+    typed_storage = torch.storage.TypedStorage(
+        wrap_storage=rebuilt_storage, dtype=dtype, _internal=True
+    )
+
+    # Rebuild tensor
+    metadata = (
+        data["tensor_offset"],
+        tuple(data["tensor_size"]),
+        tuple(data["tensor_stride"]),
+        data["requires_grad"],
+    )
+    tensor = reductions.rebuild_tensor(torch.Tensor, typed_storage, metadata)
+    return tensor
+
+
 def _from_shm(obj):
     """Reconstruct from shared memory metadata. Does NOT unlink - caller handles that."""
     if not isinstance(obj, dict):
         if isinstance(obj, list):
             return [_from_shm(v) for v in obj]
         return obj
+
+    # TensorRef -> use PyTorch's native deserialization (new format, worker->parent)
+    if obj.get("__type__") == "TensorRef":
+        tensor = _deserialize_tensor_native(obj)
+        # Convert back to numpy if it was originally numpy
+        if obj.get("__was_numpy__"):
+            return tensor.numpy()
+        return tensor
+
+    # __shm_np__ -> legacy format (parent->worker, uses Python SharedMemory)
     if "__shm_np__" in obj:
         block = shm.SharedMemory(name=obj["__shm_np__"])
         arr = np.ndarray(tuple(obj["shape"]), dtype=np.dtype(obj["dtype"]), buffer=block.buf).copy()
@@ -616,13 +784,16 @@ def _from_shm(obj):
             import torch
             return torch.from_numpy(arr)
         return arr
-    # trimesh (pickled to preserve visual, metadata, normals)
+
+    # trimesh (pickled)
     if "__shm_trimesh__" in obj:
         import pickle
         block = shm.SharedMemory(name=obj["name"])
         mesh_bytes = bytes(block.buf[:obj["size"]])
         block.close()
+        block.unlink()
         return pickle.loads(mesh_bytes)
+
     return {k: _from_shm(v) for k, v in obj.items()}
 
 def _cleanup_shm(registry):
@@ -634,6 +805,25 @@ def _cleanup_shm(registry):
             pass
     registry.clear()
 
+# Shared memory keeper - holds references to prevent premature GC
+class ShmKeeper:
+    """Keep shm blocks alive for a retention period to prevent race conditions."""
+    def __init__(self, retention_seconds=30.0):
+        self.retention_seconds = retention_seconds
+        self._keeper = collections.deque()
+        self._lock = threading.Lock()
+
+    def keep(self, blocks):
+        now = time.time()
+        with self._lock:
+            self._keeper.append((now, list(blocks)))  # Copy the list
+            # Cleanup old entries
+            while self._keeper and now - self._keeper[0][0] > self.retention_seconds:
+                old_time, old_blocks = self._keeper.popleft()
+                _cleanup_shm(old_blocks)
+
+_shm_keeper = ShmKeeper()
+
 # =============================================================================
 # Object Reference System - keep complex objects in worker, pass refs to host
 # =============================================================================
@@ -796,6 +986,9 @@ def _connect(addr):
 def _deserialize_isolated_objects(obj):
     """Reconstruct objects serialized with __isolated_object__ marker."""
     if isinstance(obj, dict):
+        if obj.get("__path__"):
+            from pathlib import Path
+            return Path(obj["__path__"])
         if obj.get("__isolated_object__"):
             attrs = {k: _deserialize_isolated_objects(v) for k, v in obj.get("__attrs__", {}).items()}
             ns = SimpleNamespace(**attrs)
@@ -941,6 +1134,7 @@ def main():
                 wlog(f"[worker] Creating instance...")
                 instance = object.__new__(cls)
                 if self_state:
+                    self_state = _deserialize_isolated_objects(self_state)
                     instance.__dict__.update(self_state)
                 wlog(f"[worker] Calling {method_name}...")
                 method = getattr(instance, method_name)
@@ -957,7 +1151,7 @@ def main():
             wlog(f"[worker] Created {len(shm_registry)} shm blocks for result")
 
             transport.send({"status": "ok", "result": result_meta})
-            # Note: don't cleanup shm_registry here - host needs to read it
+            _shm_keeper.keep(shm_registry)  # Keep alive for 30s until host reads
 
         except Exception as e:
             # Cleanup shm on error since host won't read it
@@ -1134,7 +1328,9 @@ class SubprocessWorker(Worker):
         lib_dir = self.python.parent.parent / "lib"
         if lib_dir.is_dir():
             existing = env.get("LD_LIBRARY_PATH", "")
-            env["LD_LIBRARY_PATH"] = f"{lib_dir}:{existing}" if existing else str(lib_dir)
+            # Also include system library paths so apt-installed libs (OpenGL, etc.) are found
+            system_libs = "/usr/lib/x86_64-linux-gnu:/usr/lib:/lib/x86_64-linux-gnu"
+            env["LD_LIBRARY_PATH"] = f"{lib_dir}:{system_libs}:{existing}" if existing else f"{lib_dir}:{system_libs}"
 
         # On Windows, pass host Python directory so worker can add it via os.add_dll_directory()
         # This fixes "DLL load failed" errors for packages like opencv-python-headless

comfy_env/isolation/wrap.py

@@ -72,22 +72,14 @@ def _get_worker(env_dir: Path, working_dir: Path, sys_path: list[str],
         if cache_key in _workers and _workers[cache_key].is_alive():
             return _workers[cache_key]
 
-        host_ver = f"{sys.version_info.major}.{sys.version_info.minor}"
-        iso_ver = _get_python_version(env_dir)
         python = env_dir / ("python.exe" if sys.platform == "win32" else "bin/python")
 
-        if iso_ver and iso_ver != host_ver:
-            # Different Python version - must use SubprocessWorker
-            from .workers.subprocess import SubprocessWorker
-            print(f"[comfy-env] SubprocessWorker: {python} ({iso_ver} vs {host_ver})")
-            worker = SubprocessWorker(python=str(python), working_dir=working_dir, sys_path=sys_path, name=working_dir.name)
-        else:
-            # Same version - use MPWorker with venv Python for true isolation (like pyisolate)
-            # This fixes Windows where spawn would otherwise re-import main.py
-            from .workers.mp import MPWorker
-            print(f"[comfy-env] MPWorker: {python}")
-            worker = MPWorker(name=working_dir.name, sys_path=sys_path, lib_path=lib_path,
-                            env_vars=env_vars, python=str(python))
+        # Always use SubprocessWorker - MPWorker's spawn mechanism tries to re-import
+        # the parent's __main__ (ComfyUI's main.py), which fails with import errors.
+        # SubprocessWorker uses a clean entry script that avoids this issue.
+        from .workers.subprocess import SubprocessWorker
+        print(f"[comfy-env] SubprocessWorker: {python}")
+        worker = SubprocessWorker(python=str(python), working_dir=working_dir, sys_path=sys_path, name=working_dir.name)
 
         _workers[cache_key] = worker
         return worker

{comfy_env-0.1.20.dist-info → comfy_env-0.1.22.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: comfy-env
-Version: 0.1.20
+Version: 0.1.22
 Summary: Environment management for ComfyUI custom nodes - CUDA wheel resolution and process isolation
 Project-URL: Homepage, https://github.com/PozzettiAndrea/comfy-env
 Project-URL: Repository, https://github.com/PozzettiAndrea/comfy-env
@@ -17,6 +17,7 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Requires-Python: >=3.10
+Requires-Dist: numpy
 Requires-Dist: pip>=21.0
 Requires-Dist: tomli-w>=1.0.0
 Requires-Dist: tomli>=2.0.0

{comfy_env-0.1.20.dist-info → comfy_env-0.1.22.dist-info}/RECORD

@@ -1,6 +1,6 @@
-comfy_env/__init__.py,sha256=7R1WnVv6Rmq97bbK2Fvp1A50jkFwxIgcmfiILit9k8E,4666
-comfy_env/cli.py,sha256=SWErVa3lB2ZHc7aNJJgWElWCRMsCzVLP_CrupuOk_zw,7684
-comfy_env/install.py,sha256=CT3A3kIVQqXGA5nfp7jxTU_gHdsood4NsjW0tbXv7_8,10781
+comfy_env/__init__.py,sha256=xm2aZVHfUct136_yGkO-Z0IHH5g8-MMMzJPvgWZxSYg,4634
+comfy_env/cli.py,sha256=yO5BvNhZgRy59QVv6b0a0kzxG44BCcg9GYh0qMRbuvc,8134
+comfy_env/install.py,sha256=jSfUH3f55MwUwIItkQnLRWqhd-Dk4xH2GNJB8X9SBu0,10894
 comfy_env/config/__init__.py,sha256=QlxIc5Hdghje6cm4FutReMO6fQK5rBu-zr36V2fjcLE,474
 comfy_env/config/parser.py,sha256=bNSy8Jn4VZg3xw8pldlkEyGaaI_BmNsUEYQz-hKiXgE,2221
 comfy_env/config/types.py,sha256=vFgWeEl_p26OmcoUv0wAOlHe9GBio2isjIWl7kACKFY,1096
@@ -14,13 +14,12 @@ comfy_env/environment/cache.py,sha256=RGfVW2caMO0Dd1nX2otUQP0xW3pVS7iSOP4vIUAMdE
 comfy_env/environment/libomp.py,sha256=nzr3kDnRLgcf9CZ_WF4ItWskqEDS2S0geqZS43XoKig,1319
 comfy_env/environment/paths.py,sha256=5TFFAkOZXa8R3cHfVHDEFnwy6_JcHilVBOHJuy-yqR0,1129
 comfy_env/environment/setup.py,sha256=KQgeqlEaqB_tOVhsR2RQF76-LuPud2EPtkQWUM3AJ5Y,3231
-comfy_env/isolation/__init__.py,sha256=XfMLEiBIcEzHG_k2vk9fT9GvFfmOsfbpM26czuxbdRI,800
+comfy_env/isolation/__init__.py,sha256=6VtzWaY0B_Cpk6TQP2Lr4QUrDjBfguuhVpwCi_GByhs,768
 comfy_env/isolation/tensor_utils.py,sha256=2_f4jjylqCPaPldD1Jw-es5CyOtuF5I1ROdyEIxsg-U,2951
-comfy_env/isolation/wrap.py,sha256=UO5tVK-Bzp-8AVxkRd9WNU8m9lOfy_oZFkA6trHOcbM,8876
-comfy_env/isolation/workers/__init__.py,sha256=Zp6sZSRBcb5Negqgzqs3jPjfO9T1u3nNrQhp6WqTAuc,325
-comfy_env/isolation/workers/base.py,sha256=4ZYTaQ4J0kBHCoO_OfZnsowm4rJCoqinZUaOtgkOPbw,2307
-comfy_env/isolation/workers/mp.py,sha256=0zGDo5RxbuiSGTbTxf5qYE0tD-Sistzl48CsIOkYCdg,34777
-comfy_env/isolation/workers/subprocess.py,sha256=15G4rQA7bO-mmVkkeeR3we9ytHIw6kHLNdjc9XWXCu8,57650
+comfy_env/isolation/wrap.py,sha256=d-JLGHHBfb6NerAtETMZouqqCSmQD3CsPpgsN8FVCac,8412
+comfy_env/isolation/workers/__init__.py,sha256=rmfSuH03_BbFFy0TSBTBL9UthhzP69o3Ou8p5R0WBrw,264
+comfy_env/isolation/workers/base.py,sha256=feRCKtjVtTVh1r3efZJ1cJ3rH9D1xKBCbbwLlckcBJk,2348
+comfy_env/isolation/workers/subprocess.py,sha256=mvhuc2kvt3g3FP_moLHJ407IkOoZrGn-pmyk7PL95FU,64609
 comfy_env/packages/__init__.py,sha256=4pRCUnfcVFVgy7hkbPz9BPVXELtSFHha6L7n-hqNuZA,1155
 comfy_env/packages/apt.py,sha256=pxy3A5ZHv3X8ExCVyohODY8Fcy9ji4izIVPfYoxhqT4,1027
 comfy_env/packages/cuda_wheels.py,sha256=G_CnlwNcfeWlEU24aCVBpeqQQ05y8_02dDLBwBFNwII,3980
@@ -29,8 +28,8 @@ comfy_env/packages/pixi.py,sha256=RPu8x5sSOLE1CYAhWMMjoQrbFGGt00fdsbqtRcTz7LQ,38
 comfy_env/packages/toml_generator.py,sha256=Vhc8F9euHhMTwH1TV6t96-D9Pjrn9jIN4e9WXrCIFE8,3414
 comfy_env/templates/comfy-env-instructions.txt,sha256=ve1RAthW7ouumU9h6DM7mIRX1MS8_Tyonq2U4tcrFu8,1031
 comfy_env/templates/comfy-env.toml,sha256=ROIqi4BlPL1MEdL1VgebfTHpdwPNYGHwWeigI9Kw-1I,4831
-comfy_env-0.1.20.dist-info/METADATA,sha256=TnybPs-8qbU-3NjJxnAadtdp2mj7Pj6fjTfMA2e7PqQ,4808
-comfy_env-0.1.20.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-comfy_env-0.1.20.dist-info/entry_points.txt,sha256=J4fXeqgxU_YenuW_Zxn_pEL7J-3R0--b6MS5t0QmAr0,49
-comfy_env-0.1.20.dist-info/licenses/LICENSE,sha256=E68QZMMpW4P2YKstTZ3QU54HRQO8ecew09XZ4_Vn870,1093
-comfy_env-0.1.20.dist-info/RECORD,,
+comfy_env-0.1.22.dist-info/METADATA,sha256=0WOa_D45kdxrIhNe8cdZLGIH3_9ZrCWyRxS5Zmz9MNA,4829
+comfy_env-0.1.22.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+comfy_env-0.1.22.dist-info/entry_points.txt,sha256=J4fXeqgxU_YenuW_Zxn_pEL7J-3R0--b6MS5t0QmAr0,49
+comfy_env-0.1.22.dist-info/licenses/LICENSE,sha256=E68QZMMpW4P2YKstTZ3QU54HRQO8ecew09XZ4_Vn870,1093
+comfy_env-0.1.22.dist-info/RECORD,,