magogi/prod/: mgf-hrb-0.1.10 metadata and description
Magogi Foundation hardware resource broker — userspace arbitration of all hardware (PCIe, USB, GPU, CPU/memory) for any client. Zig below the socket; Python above.
| author | Bassam Alsanie, mgf-hrb contributors |
| description_content_type | text/markdown |
| license | Apache-2.0 |
| license_file |
|
| metadata_version | 2.5 |
| requires_dist |
|
| requires_python | >=3.11 |
Because this project isn't in the mirror_whitelist,
no releases from root/pypi are included.
| File | Tox results | History |
|---|---|---|
mgf_hrb-0.1.10-py3-none-any.whl
|
|
|
mgf_hrb-0.1.10.tar.gz
|
|
mgf-hrb — Magogi Foundation Hardware Resource Broker
Replaces:
mgf-usb(deleted 2026-05-27). SeeMIGRATION_FROM_MGF_USB.mdfor the supersession statement + what was carried forward. Shape. Federation Python library (sibling ofmgf-common); pinsmgf-common>=0.38,<0.39. The Zig broker ships alongside the SDK in this same repo (hrb-code/) so the two sides of the no-FFI boundary stay in lockstep. Conformance level. L2 — Standard (flipped 2026-05-24 after the loggers +HrbSettings+ LG-rationale pass; seedocs/inprogress/MGF_STANDARDS_CONFORMANCE.md§4 for the audit log). Status. Phases 1–7 of the design are in code; Part 13 (self-aware introspection) shipped 21/21 sub-features in May 2026; Tier-0 survivability #1/#2/#3/#4 all landed (only #5 — asymmetric shm trust — remains). Gates green: 308 Zig tests (broker; ~44 more in the carved leaf repos) + 466 Python tests (default suite); marker-gated stress/perf/e2e add ~94 more. Zero-FFI Python↔Zig over a Unix socket; S1 trust gate; S2 deadlines on every kabi op; runtime identity-continuity comparator (MODEL-01); privileged-helper subprocess (D-031) opt-in viaMGF_HRB_USB_USE_HELPER=1.
A userspace hardware-resource broker that arbitrates read/write access to all hardware (PCIe, USB, GPU, CPU/memory) for any client — reliable, fast, complete, impossible to need to bypass. Designed kernel-agnostic so it lifts onto a from-scratch capability-based OS by writing one backend.
Priority order: Reliability · Performance · Fidelity · Flexibility · Simplicity. Languages: Zig below the socket (enforcement, hardware, hot path) · Python above it (ergonomics, policy, tooling).
Repo shape
mgf-hrb/
├── README.md this file
├── STARTHERE.md procedural how-to
├── INDEX.md deep entry point (links to design, code, decisions, recipes)
├── LICENSE CHANGELOG.md SECURITY.md FEEDBACK.md USERS.md PUBLIC_API.{md,json}
├── pyproject.toml Python library packaging
├── src/mgf/hrb/ the Python SDK + embedded CLI
│ ├── __init__.py public surface (see PUBLIC_API.md)
│ ├── _client.py · domain.py · identity.py · exceptions.py
│ └── cli/ mgf-hrb console script (scan / info / doctor / explain)
├── tests/
│ ├── unit/mgf/hrb/ pytest tree mirroring src/
│ └── e2e/ Python ↔ Zig over a real socket
├── examples/ runnable consumer snippets
├── hrb-code/ the Zig broker (kernel side; 308 Zig tests)
├── hrb-design/ the design corpus (spine + 4 deepenings + survivability review)
├── recipes/ canonical consumer patterns
├── docs/
│ ├── claude/CLAUDE.md project-specific Claude reference (DOC-07)
│ ├── inprogress/MGF_STANDARDS_CONFORMANCE.md
│ ├── standards/README.md pointer-only → mgf-common (DOC-02 §2.0.4)
│ └── cutover/ (empty until first breaking change)
├── DECISIONS.md ADR catalog (35 D-NNN decisions)
└── STANDARDS_NOTE.md the *argument* behind the conformance posture
Quick start
Install the Python SDK (editable, with dev tools):
git clone <repo-url> mgf-hrb && cd mgf-hrb
uv venv && uv pip install -e '.[dev]'
Run the broker (Zig, from this same repo):
cd hrb-code
zig build && ./zig-out/bin/mgf-hrb-server
# → mgf-hrb broker listening on /tmp/mgf-hrb.sock
(Zig 0.16 — no zig on PATH? uv pip install ziglang and alias zig to python3 -m ziglang.)
Smallest useful program:
from mgf.hrb import Client, ClaimMode
with Client.connect("/tmp/mgf-hrb.sock") as c:
# Enumerate (read-only, never claims):
for urn in c.list("hw:/**"):
print(urn)
# Peek at one URN's metadata (vid/pid/driver/etc), still read-only:
detail = c.inspect("hw:/usb/bus3/port-9/dev/if0")
print(detail["attrs"])
# Claim exclusively; lease auto-releases at block exit (DP-13):
with c.claim("hw:/usb/bus3/port-9/dev/if0", ClaimMode.EXCLUSIVE) as lease:
ch = lease.raw() # SCM_RIGHTS-passed fd; broker steps out (G4)
... # use `ch["fd"]` directly with USBDEVFS ioctls
Operator CLI:
mgf-hrb scan # enumerate, render the device tree
mgf-hrb doctor # diagnose broker preconditions (socket, IOMMU, sysfs)
mgf-hrb --help
Test suites:
scripts/zig-fetch-deps # cache carved Zig deps (one-time)
cd hrb-code && zig build test --summary all # 308 Zig tests (broker)
zig build && pytest tests/e2e -m e2e # Python↔Zig E2E (marker-gated)
pytest # 466 Python tests (default suite)
scripts/mtest --tier 0-1 --budget 60s # budget-aware supervisor (recommended)
scripts/publish-wheel # build + verify the SDK wheel (dry run)
scripts/publish-wheel --upload # …and publish it (CI / release)
The Python SDK (src/mgf/hrb/) is published as a pip install-able wheel
from this monorepo — the broker and SDK deliberately stay in one repo
so the two sides of the no-FFI wire protocol never skew. See
scripts/publish-wheel.
Where to read next
| You want… | Start at |
|---|---|
| The procedural how-to | STARTHERE.md |
| The architectural story | INDEX.md → hrb-design/mgf-hrb-complete-design.md |
| The runnable examples | examples/ |
| The canonical consumer patterns | recipes/ |
| The public Python API | PUBLIC_API.md |
| The kernel acceptance test (your future OS) | hrb-design/kernel-design-brief.md + hrb-code/src/conformance.zig |
Federation
This project depends on mgf-common and follows the Magogi engineering bar to the extent it applies to a dual-language Zig/Python project. The Python SDK inherits typed exceptions (EH-01 / EH-02), four-gate CI, the AP / DP / TS / DOC rule families. The Zig kernel side is governed by its own conformance contracts — the kabi behavioural suite (hrb-code/src/conformance.zig) and the S1–S5 survivability bar (Part 12). See STANDARDS_NOTE.md for the deeper reasoning and docs/inprogress/MGF_STANDARDS_CONFORMANCE.md for the per-rule audit.
For AI agents: run mgf-common agents for the canonical federation briefing, then docs/claude/CLAUDE.md for the project-specific Claude reference.
License
MIT.