feat(deploy): hm.deploy + hm.dev local-deployment DSL + CI#1
Merged
Conversation
Driver-agnostic top-level decorator (@hm.deploy) + driver-specific factory (hm.dev.deploy) so future cloud drivers slot in without touching the registry. v1 ships local Docker only via `hm dev up`, foreground, with per-session bridge networks and OS-assigned host ports so multiple worktrees can coexist. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
12 tasks, TDD-shaped: scaffold abstract Deployment, port sentinel, LocalDeployment, deploy() factory, Dep[T] marker + call_with_deps extension, @hm.deploy decorator, topo sort, dump_registry_json, python -m harmont.dev CLI shim, CLAUDE.md update, end-to-end canonical example, PR-readiness pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sets the driver-agnostic foundation for hm.deploy. Concrete LocalDeployment (Task 3) subclasses Deployment; the DEPLOYMENTS registry stores polymorphic entries. Test-only reset fixture covers DEPLOYMENTS plus the existing TARGETS/REGISTRATIONS registries so all three are wiped between tests. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught that the raw _TARGETS_BY_NAME / _RESOLVING clears miss _TARGET_CACHE (the per-render target memo in harmont/_target.py), so future tests using @hm.target would leak memoized values across tests. Switch to clear_registry() + clear_target_cache(), matching the pattern in tests/test_registry.py and tests/test_target.py. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Singleton with stable repr and hash. Misuse outside port_mapping is detected by deploy()'s field validation (Task 4), not at the port() call site, so the error points at the exact misuse location. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught two minor issues:
1. `_PortSentinel` in __all__ violates the project's underscore-prefix
private-name convention (cf. harmont/_typing.py — _TargetMarker stays
private, Target is the public alias). Downstream isinstance checks
import directly from harmont.dev._port.
2. The bare {port(): 1} expression would trip ruff B018 (useless
expression) at Task 11's lint gate. Switched to an explicit
assert {port(): 1}[port()] == 1 — still tests hashability AND
key-equality, lint-clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Concrete subclass of Deployment for the local Docker driver. __post_init__ enforces driver=='local'; everything else is a plain dataclass field. The deploy(...) factory in Task 4 is the sanctioned constructor. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught it as F401 dead-import (Step is imported alongside scratch but never referenced). Trimmed to just scratch. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Strict, fix-directed validation per PRINCIPLES § 5: every error message points at the misuse and states the fix. The factory leaves name="" so the @hm.deploy decorator can stamp the slug in via dataclasses.replace after deciding the slug from its arg or fn name. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught the Pyright "unreachable" on the `isinstance(from_, Step)` guard — Pyright correctly narrows from_ to Step after the preceding XOR check, so the runtime defensive branch is statically dead. Type system already enforces the Step | None contract; drop the check. Step stays imported (still in the type annotation). Plus two minor cleanups: - volumes host-path error now includes type name (consistency with the env / cmd / port_mapping messages). - test rename: test_volumes_keys_resolve_relative_to_worktree... → test_volumes_preserves_host_path_verbatim (the old name implied behavior that lives in _registry_dump, not the factory). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Old name implied "resolves relative to worktree at dump time" — that behavior lives in _registry_dump.py, not the factory. The factory preserves the host-path string verbatim. Renamed to test_volumes_preserves_host_path_verbatim. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Dep[T] resolves a parameter against harmont._deploy.DEPLOYMENTS by the parameter name (same shape as Target[T] vs _TARGETS_BY_NAME). Cycle detection reuses the existing _RESOLVING stack so dep cycles between deployments and dep cycles between targets share one detector. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Implementer chose isinstance(marker, _DepMarker) over the identity-check pattern (consistent with how _BaseImageMarker is checked), so the _DEP_MARKER instance import is unused and Pyright flags it. Drop it. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught three integration gaps with the existing `_deps.py` machinery: 1. Target "not registered" raises TypeError; the new Dep branch was raising ValueError. Aligned to TypeError. 2. Error-message shape: existing errors are `"hm: <kind> 'name' not found\n → fix"`. The Dep error was inline `"hm.Dep parameter ..."` with an em-dash inline fix and a trailing period. Reformatted to match the in-file convention exactly. 3. _DepMarker docstring missing the second-sentence singleton hint that _TargetMarker carries — added. Test updated to expect TypeError + the new message, plus a marker assertion in test_dep_marker_alias_subscripts_to_annotated. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ection Decorator validates the slug regex (Docker container-name rules), rejects duplicates, validates the function signature via the existing validate_target_signature, and wraps the function so call_with_deps resolves Target/Dep/BaseImage markers at registry-walk time. dataclasses.replace stamps the resolved slug into the returned Deployment so the value seen by callers and the registry has name=<slug> (the factory leaves name=""). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Spec reviewer caught two duplications in the prior commit: 1. _validate_deploy_signature was a deploy-specific re-implementation of the shared validate_target_signature from _deps.py. The plan mandated calling the shared validator. The variant existed only to raise ValueError instead of TypeError to match a test expectation that was itself inconsistent with Task 5's TypeError precedent for marker errors. 2. _marker_for was copied verbatim from _deps.py to support the variant validator. Both deleted. validate_target_signature is now imported from _deps and called directly. test_deploy_requires_marker_on_param updated to expect TypeError + the actual validator's "has no marker" message. `name=` kwarg is now explicitly `del name`'d to silence Pyright's "not accessed" hint without losing the future-API kwarg shape. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer noted the duplicate-slug message suggested passing `name="..."` to disambiguate, but `name=` is reserved-for-future (currently a no-op via `del name`). Pointing users at it is misleading. The actual fix is to pass an explicit `slug=` — that's now what the error says. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
dep_graph walks the registry, introspects each wrapped function for Dep[T] params, and emits slug -> tuple of upstream slugs in parameter order. topo_order runs Kahn's algorithm with stable level ordering (insertion order within a level) so the registry-dump output is deterministic. Cycle detection raises RuntimeError listing the unresolved slugs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Walks DEPLOYMENTS in topo order, lowering LocalDeployment values to
the schema documented in the spec (§ 1). Step-chain from_= values are
lowered via the existing harmont.pipeline() + keygen pipeline so the
Rust executor can run the chain and use the terminal key as the
build-image tag. Non-local drivers are passed through as
{"driver": X, "_unhandled": true} for hm dev ls.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reviewer caught the leftover `if TYPE_CHECKING: pass` scaffolding (with the TYPE_CHECKING import) — dead code from the plan-quoted template. None of the imports under TYPE_CHECKING were needed. Removed both the import and the block. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Walks .harmont/*.py, imports each by file path so @hm.deploy registrations land in harmont._deploy.DEPLOYMENTS, then prints the deployment registry JSON to stdout. The Rust CLI invokes this and deserializes via serde (see harmont-cli plan). Missing .harmont/ exits 1 with a fix-directed stderr. Argparse handles usage errors with exit 2. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
parser.error() is typed NoReturn (calls sys.exit(2) internally), so the trailing return was unreachable. Removed; added a one-line comment noting the NoReturn so future readers don't reintroduce it. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds the deployments section to the agent-facing doc with the canonical example and the full public surface. Cross-links the design spec for engineers who need the wire-format details. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mirrors the spec's worked example. Asserts topo order, dep edges, cross-deploy f-string env values, and that from_=Step lowers through the existing v0 IR pipeline. This is the "vibe check" gate before the CLI plan can start consuming the JSON output. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous canonical example used postgres:16 + a Step-chain api build + node:20 — three heavy images and a build path that v1 cli stubs out. The hello+greeter pair now runs `python -m http.server` from `python:3.12-alpine` (the Python stdlib's built-in HTTP server; no third-party image dependency). Same surface coverage (@hm.deploy, hm.Dep[T], cross-deploy env interpolation), much smaller footprint. Updated the design spec § 1 + § 6, CLAUDE.md, the py plan's Task 11, and tests/dev/test_canonical_example.py to match. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
PR + push-to-main gate. Matrix over python 3.11 / 3.12 (match the package's requires-python = ">=3.11"). Existing release.yml (tag-driven PyPI publish) is untouched. Excludes tests/test_gradle.py and tests/test_haskell.py via --deselect — those have pre-existing failures unrelated to the hm.deploy work and would block PRs unrelated to them. Pre-flight (Step 3): ruff was 49 errors red across harmont/ and tests/dev/ — all pre-existing. Fixed them before adding the workflow so it doesn't land already-broken: moved type-only imports to TYPE_CHECKING blocks (TC001/TC003), ValueError→TypeError for pure type checks (TRY004), noqa suppressions where rename would be lossy (S108 deliberate /tmp sentinel, SLF001 test-isolation internals), split compound assertions (PT018), re-ordered imports (E402), used sys.stderr.write/sys.stdout.write instead of print (T201), extracted f-string exception messages (EM102/TRY003). Tests updated to match TypeError raises. mypy: clean. pytest: 373 passed. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8 tasks
Plans are scratchpads for implementation; once code lands the spec + tests + commit messages carry the load. Remove the plans dir entirely (including the pre-existing pypi-tag-release-cd plan from prior work). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Spec was implementation-time scaffolding; once the code lands the
public surface lives in CLAUDE.md (the agent-facing doc) and the
code itself. The empty docs/superpowers/{specs,} dirs are also
gone (git auto-prunes empty trees).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.