molecule-ai/molecule-core
41
0
Fork 2
Code Issues 46 Pull Requests 23 Actions 36 Packages Projects Releases Wiki Activity

fix(workspace): register plugins_registry as sys.modules shim before loading adapters #301

Closed
infra-runtime-be wants to merge 2 commits from runtime/fix-plugin-registry-import-path into main
pull from: runtime/fix-plugin-registry-import-path
merge into: molecule-ai:main
Conversation 5 Commits 2 Files Changed 8 +86 -9

2 Commits

Author SHA1 Message Date
Molecule AI Infra-Runtime-BE
2e0080fb0b fix(workspace): register plugins_registry as sys.modules shim before loading adapters
Some checks failed
Secret scan / Scan diff for credential-shaped strings (pull_request) Failing after 1s
Details
sop-tier-check / tier-check (pull_request) Failing after 1s
Details
audit-force-merge / audit (pull_request) Has been skipped
Details 
KI-296 fix: when the PyPI-installed runtime (molecule-ai-workspace-runtime
0.1.129+) ships plugins_registry as molecule_runtime.plugins_registry (a
subpackage), plugin adapter files that do ``from plugins_registry import ...``
as a top-level name fail with ModuleNotFoundError because Python's import
system cannot find a top-level ``plugins_registry`` package.

The fix in plugins_registry/__init__.py:_load_module_from_path() registers
molecule_runtime.plugins_registry as ``plugins_registry`` in sys.modules
before exec'ing any plugin adapter file, so the top-level import resolves
correctly in both environments:
- PyPI wheel (molecule_runtime.plugins_registry → sys.modules["plugins_registry"])
- molecule-core workspace source (top-level workspace/plugins_registry already
  on sys.path; the setdefault is a no-op)

Submodules (builtins, protocol, raw_drop) are also registered so adapters
that import ``from plugins_registry.builtins import ...`` work without error.

Added test_load_module_from_path_registers_plugins_registry_sys_modules.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
 2026-05-10 14:55:05 +00:00
Molecule AI Infra-Runtime-BE
75b5901d57 fix(workspace): default PLATFORM_URL to host.docker.internal in all modules 
KI-014 follow-on: inside a workspace container, localhost refers to the
container itself, not the platform. Four files had the Docker-aware
if-branch correct but fell through to localhost:8080 as the non-Docker
fallback — effectively making the Docker path the ONLY path that works,
since local dev on Mac/Linux can also resolve host.docker.internal via
the Docker daemon's built-in resolver.

Fix: unify the default to host.docker.internal in both branches, so
the env-var override always works and no caller ever silently falls
back to the wrong address.

- a2a_cli.py: else branch hardcoded localhost → host.docker.internal
- consolidation.py: same
- coordinator.py: same
- builtin_tools/temporal_workflow.py: two inline os.environ.get defaults
  replaced with a _platform_url() helper for DRY + consistent detection

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
 2026-05-10 14:55:05 +00:00