docs(6205): forge comments state the reason, not the instance that taught it
CI & Build / Python lint (push) Successful in 2s
CI & Build / Plugin hooks (push) Successful in 13s
CI & Build / TypeScript typecheck (push) Successful in 55s
CI & Build / integration (push) Successful in 1m22s
CI & Build / Python tests (push) Successful in 2m11s
CI & Build / Build & push image (push) Successful in 25s
CI & Build / Python lint (push) Successful in 2s
CI & Build / Plugin hooks (push) Successful in 13s
CI & Build / TypeScript typecheck (push) Successful in 55s
CI & Build / integration (push) Successful in 1m22s
CI & Build / Python tests (push) Successful in 2m11s
CI & Build / Build & push image (push) Successful in 25s
The module header drops the milestone/step map and record citations; the per-user optionality, the keyring model and the empty-selector baseline are stated as present-tense facts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
@@ -1,25 +1,24 @@
|
||||
"""Forge adapter — optional server-side READ access to the operator's git forge.
|
||||
|
||||
Step 4 of milestone 288 (#2689, decision #2686). The recorded location of a
|
||||
snippet is the source of truth for its code and the stored body is a cache;
|
||||
this module is the seam that lets the SERVER read that source of truth, so the
|
||||
cache can be refreshed at pull time (step 5), drift can be flagged from push
|
||||
webhooks (step 6), and coverage can be measured (step 7).
|
||||
The recorded location of a snippet is the source of truth for its code and
|
||||
the stored body is a cache; this module is the seam that lets the SERVER read
|
||||
that source of truth, so the cache can be refreshed at pull time, drift can be
|
||||
flagged from push webhooks, and coverage can be measured.
|
||||
|
||||
Design constraints, in force everywhere below:
|
||||
|
||||
- OPTIONAL per user (rule #115, sharpened by #2778). Connections are
|
||||
- OPTIONAL per user. Connections are
|
||||
per-user keyring rows resolved by repo host on the PROJECT OWNER's
|
||||
keyring; `get_forges()` returns an empty selector when the owner has
|
||||
nothing configured, and every consumer must treat that as "keep today's
|
||||
behavior". A user who never configures a forge is not degraded — that is
|
||||
nothing configured, and every consumer must treat that as "behave as if
|
||||
there were no forge". A user who never configures a forge is not degraded — that is
|
||||
the baseline.
|
||||
- READ-ONLY by construction. The adapter exposes reads; there is no write
|
||||
method to misuse. The token an operator mints for it only ever needs read
|
||||
scope, and the docs say so.
|
||||
- The contract stays as small as its consumers (steps 5-7): read_file /
|
||||
- The contract stays as small as its consumers: read_file /
|
||||
latest_commit / archive / default_branch / resolve_repo / check. Two
|
||||
implementations (Gitea, GitHub — step 8) keep it honest; resist widening
|
||||
implementations (Gitea, GitHub) keep it honest; resist widening
|
||||
it speculatively.
|
||||
- Repo identity is the repo-binding key — `normalize_repo_key`'s
|
||||
host/owner/repo — so the join between a snippet's recorded repo and the
|
||||
@@ -53,12 +52,12 @@ logger = logging.getLogger(__name__)
|
||||
FORGE_KINDS = ("gitea", "github")
|
||||
|
||||
# Total budget per forge call. Consumers either have a cache to fall back to
|
||||
# (step 5) or a user watching a button (the test probe) — neither tolerates a
|
||||
# or a user watching a button (the test probe) — neither tolerates a
|
||||
# hung socket, and there is no retry: the fallback IS the retry policy.
|
||||
_TIMEOUT = httpx.Timeout(5.0)
|
||||
|
||||
# Archive downloads move a whole-repo tarball and only ever run off the
|
||||
# request path (coverage recompute, step 7), so they get a bigger budget than
|
||||
# request path (coverage recompute), so they get a bigger budget than
|
||||
# the per-file reads — but still a bound, because a hung background task
|
||||
# holds a connection slot as surely as a foreground one.
|
||||
_ARCHIVE_TIMEOUT = httpx.Timeout(60.0)
|
||||
@@ -79,13 +78,13 @@ class ForgeFile:
|
||||
"""One file read from the forge at a specific point in history."""
|
||||
|
||||
content: str
|
||||
# The commit the content was served at — what provenance stores (#2688).
|
||||
# The commit the content was served at — what provenance stores.
|
||||
commit_sha: str
|
||||
path: str
|
||||
|
||||
|
||||
def host_of(url: str) -> str:
|
||||
"""The lowercase hostname of a URL — the keyring's lookup key (#2778)."""
|
||||
"""The lowercase hostname of a URL — the keyring's lookup key."""
|
||||
return (urlsplit(url).hostname or "").lower()
|
||||
|
||||
|
||||
@@ -199,7 +198,7 @@ class ForgeAdapter:
|
||||
async def latest_commit(self, repo: str, path: str, ref: str = "") -> str:
|
||||
"""The newest commit touching ``path`` — "" when it can't be told.
|
||||
|
||||
The cached-SHA short-circuit (#2693): when a snippet's provenance
|
||||
The cached-SHA short-circuit: when a snippet's provenance
|
||||
already names a commit, this one small call can prove the file
|
||||
hasn't moved since — no content transfer, which is what keeps
|
||||
pull-time freshness inside GitHub's rate limits.
|
||||
@@ -213,7 +212,7 @@ class ForgeAdapter:
|
||||
async def archive(self, repo: str, ref: str) -> bytes:
|
||||
"""The repo's content at ``ref`` as a gzipped tarball, in one request.
|
||||
|
||||
Coverage measurement (step 7) needs every source file's text; per-file
|
||||
Coverage measurement needs every source file's text; per-file
|
||||
reads would mean one API call per file, so the archive endpoint is the
|
||||
only shape that scales past toy repos. Callers must never run this in
|
||||
a request path — it moves the whole repo.
|
||||
@@ -287,8 +286,8 @@ class GiteaForge(ForgeAdapter):
|
||||
|
||||
|
||||
class GitHubForge(ForgeAdapter):
|
||||
"""The GitHub implementation — the second one, which is the point (#2693):
|
||||
it proves the seam is a contract rather than a Gitea-shaped hole. Works
|
||||
"""The GitHub implementation — the second one, which is the point: it
|
||||
proves the seam is a contract rather than a Gitea-shaped hole. Works
|
||||
against github.com and GitHub Enterprise; the token is a fine-grained PAT
|
||||
with Contents: Read-only (or a classic token with `repo` read)."""
|
||||
|
||||
@@ -379,8 +378,8 @@ def build_adapter(
|
||||
) -> ForgeAdapter | None:
|
||||
"""One validated adapter from raw connection values, or None.
|
||||
|
||||
None means "this connection cannot serve reads" — the same contract the
|
||||
old instance-wide lookup had, applied per keyring row. Misconfigurations
|
||||
None means "this connection cannot serve reads", decided per keyring
|
||||
row. Misconfigurations
|
||||
are logged, never raised: a bad row must not break the reads the good
|
||||
rows can still serve.
|
||||
"""
|
||||
@@ -403,13 +402,13 @@ def build_adapter(
|
||||
|
||||
@dataclass(frozen=True)
|
||||
class ForgeSelector:
|
||||
"""The forge reads available to one project owner (#2778).
|
||||
"""The forge reads available to one project owner.
|
||||
|
||||
Consumers ask it to serve a REPO, not to hand over "the forge": resolve()
|
||||
walks the owner's adapters and returns the (adapter, api_repo) pair for
|
||||
the first one whose host serves the repo — or None, which every consumer
|
||||
treats exactly as the old "no forge configured" state. An empty selector
|
||||
IS rule #115's baseline.
|
||||
treats exactly as "no forge configured". An empty selector is the
|
||||
baseline every install starts from, not a degraded mode.
|
||||
"""
|
||||
|
||||
adapters: tuple[ForgeAdapter, ...] = ()
|
||||
@@ -431,7 +430,7 @@ async def get_forges(
|
||||
) -> ForgeSelector:
|
||||
"""The forge selector for reads on behalf of ``owner_id``'s records.
|
||||
|
||||
The keyring model (#2778): every server-side forge read for a record runs
|
||||
The keyring model: every server-side forge read for a record runs
|
||||
on the PROJECT OWNER's connections, resolved by repo host — a forge token
|
||||
is a user's credential, and one user's reads must never ride another
|
||||
user's token. Pass the record's ``project_id`` so the per-project pin
|
||||
|
||||
Reference in New Issue
Block a user