FakeGh.api routes inline comments on an exact suffix, so every real call falls through to the default nf-8d5 ← Beads

Twinned from https://github.com/cfsh/ninefold/issues/725 by tools/beads/import_github.py's reconcile pass.

Filed unclaimed and unlabelled from a strategic-uplift window, working #702. Backlog.

nflib.fakes.FakeGh.api() routes the inline-comment read on an exact suffix (tools/nflib/fakes.py:268):

```python if "/pulls/" in path and path.endswith("/comments"): ```

Every real caller passes a query string, so the suffix never matches:

```python >>> p = 'repos/cfsh/ninefold/pulls/7/comments?per_page=100' >>> p.endswith('/comments') False ```

The call falls through to return default, which callers pass as []. A test that seeds inline comments into the fake and asserts something about them gets an empty list and passes by reading nothing — the "tests assert nothing" class this repo files under #542/#481/#482.

The linked-issue read is worse: repos/{repo}/issues/{ref}/comments?per_page=100 (dispatch.py:1331) has no branch at all in the fake, so the whole linked_refs trigger path (#340 — comments on a linked issue are trigger activity) cannot be exercised through the shared fake.

Scope, stated honestly

⚠️ No test currently passes vacuously because of this, and I checked rather than assuming:

• merge_pr.py:243 and poll.py:463,807 read inline comments through the inline_comments() method, which the fake implements correctly. Their inline tests are sound.
• dispatch.py:1315,1331 are the only readers going through api() with a query string, and the only tests seeding inline= for the dispatcher are ones I added in #713 — where I hit this, diagnosed it, and worked around it with a local Soft subclass rather than touching the shared fake mid-chunk.

So this is a trap for the next person, not a live wrong result. It is worth filing because the failure mode is silent and the correct-looking code is the buggy one: you seed a fixture, write the assertion, it goes green, and nothing indicates the fixture was never read.

Shapes

• Cheapest: match on the path *before* the query string — urlsplit(path).path.endswith("/comments") — which the fake already imports urlsplit for (_search_issues).
• Also worth it: add the issues/{n}/comments branch, seeded from the existing issues dict, so the #340 linked-issue trigger is reachable in tests at all.
• ⚠️ Whatever the fix, it wants a test that the fake itself serves the path its real caller passes. A fake whose routing is asserted only by the tests that use it has the same hole one level down — and a shared fake is exactly the place where one silent miss propagates to every tool.

Related: #340 (the linked-issue trigger this makes untestable), #713 (where I hit it), #542/#481/#482 (the assert-nothing class).