test(licensing): make the outage-is-not-a-cancellation test actually test that

[Coding-Dev-Tools]

The defect

test_a_transport_failure_is_not_mistaken_for_a_billing_denial raised:

cloud_session.CloudSessionError("...", status=503, transient=True)

CloudSessionError.__init__ accepts only message and status — transient belongs to CloudFeatureError. So the construction raised TypeError, _fetch_authoritative_entitlement's broad except Exception swallowed it, and getattr(exc, "status", None) returned None instead of 503.

The assertion passed while never exercising a transport failure at all. It was proving "a TypeError doesn't revoke access" — not "a 503 doesn't".

This is mine: I introduced it earlier in this session while adding the 402 billing-denial handling.

Why it matters

That test guards a revenue-visible behaviour: only a 402 may clear a paying customer's cached access; an outage must not. The gate is:

if getattr(exc, "status", None) == 402:
    record_billing_denial()
    _deny_entitlement_cache()

If that were ever loosened to if getattr(exc, "status", None):, a 503 would revoke a paying customer mid-outage. The test existed to catch exactly that — and could not, because TypeError carries no status attribute at all, so the mutation was invisible to it.

Verification

Mutation-tested both ways:

• Gate loosened to any-truthy-status, after this fix → test fails: AssertionError: an outage revoked a paying customer, with cloud_access_active: False.
• Same mutation, before this fix → test passed.

Full offline gate on this branch: 1,397 passed / 0 failed (1,411 collected, 14 skipped), ruff check . clean.

One-line change plus a comment recording why, so the conflation is not reintroduced.

🤖 Generated with Claude Code

[greptile-apps]

Coding-Dev-Tools has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.

[Coding-Dev-Tools]

Review — `test(licensing): make the outage-is-not-a-cancellation test actually test that`

✅ LGTM

Good catch on the `CloudSessionError`/`CloudFeatureError` conflation. The mutation-test evidence is exactly the right proof shape: before the fix the test passed even with the gate loosened to any-truthy-status, which is the regression this assertion was meant to prevent. A test that cannot fail on the regression it guards against is worse than no test — it creates the appearance of coverage while providing none.

The comment recording why `transient` is wrong here is worth its weight: the next person to touch this will not make the same mistake, and the reason is specific enough to be actionable (`transient belongs to CloudFeatureError`, not a generic note).

CI green across 3.10/3.11/3.12 + core floor + browser smoke. +6/-1 is proportional to the fix.

One minor note for the future: consider a lint rule or type narrowing that flags `CloudSessionError(..., transient=...)` at static-analysis time rather than relying on runtime `TypeError` discovery. Not blocking — the comment and mutation test close the loop for now.

[Coding-Dev-Tools]

🔍 Pre-PR Code Review — ✅ LGTM (approve blocked: own PR)

Reviewer: Pre-PR Code Analyzer (automated)
Scope: +6/-1 in 1 file

Test Quality

Critical fix: the test was previously a no-op (TypeError from invalid transient=True swallowed by broad except). Now correctly raises CloudSessionError(status=503) and actually exercises the outage code path. Mutation testing confirms the test now catches the regression it was designed to prevent.

Verdict: LGTM — ready to merge

Minimal, surgical fix that transforms a false-assurance test into a real guard. CI green.