test(client): migrate net-stub seam tests to WS-mock, retire bespoke seams (#113) #140

Merged

bosun merged 1 commit from i/113-netstub-migration into main 2026-06-23 21:02:28 +02:00

Conversation 1 Commits 1 Files changed 3 +169 -251

shipwright commented 2026-06-23 20:54:56 +02:00

Owner

Copy link

What

Completes the bespoke-bridge debt retirement #92 began (the #80/#89/#94 lineage). Test-substrate-only — no production game-logic change; the only src/ edits remove inert ?navtest-gated test seams.

The net-backed versus/abort/forfeit/rematch tests lived in nav.spec.ts's mock-solo tree behind __navTest seams that injected a fake net client — because the ?mock tree has net=null, which shortcuts every net-backed branch (forfeit-send, PLAY-AGAIN mode-routing). That's the exact harness-substrate-vs-prod gap #91 surfaced. Driving the real net.ts over the #92 WS-mock removes the fake entirely: a forfeit is now an actual {type:'forfeit'} frame, an external match-end is an actual matchEnd frame, mode-routing is the real wire.

Seam taxonomy (the audit — substrate-grounded, not issue-extrapolated)

The issue scoped "3 seams (#87/#91/#94)"; source enumeration found 10. Partitioned by what each does (the partition that decides each one's fate):

Seam	Injects fake net?	Fate
enterVersusPlaying (#87 forfeit)	yes	migrate + remove
enterSoloPlaying (#87 solo-no-forfeit)	yes	migrate + remove
enterSoloRematchReady (#91 PLAY-AGAIN)	yes	remove (subsumed by the existing #92/#91 test)
enterLobby (#85 lobby)	no (net-null versus state)	migrate + remove (versus.spec TABLE covers it real)
forceGameover (#87 external-end)	no (phase set)	remove → real matchEnd frame
enterCountdownLockedPlaying (#110)	yes	KEEP — see boundary
enterSoloPlayAgainRace (#98)	no	keep (mock-solo, net-null)
enterInitials/enterTopScore/enterPlacedNewBest (#102/#98)	no	keep (mock-solo)
reset	no	keep

Migrated to versus.spec.ts (real net via WS-mock), each mutation-proven

Test	Mutation that reds it (verified)
#87 versus abort-confirm × Escape → real forfeit frame + yard	drop net.sendForfeit() in confirmAbort
#87 abort-confirm × Enter / × Backspace → dismiss (no forfeit)	(any-key-dismiss; forfeit observable via T1 above)
#87 solo abort → NO forfeit frame (real solo session)	remove confirmAbort's mode !== 'solo' guard
#87 touch QUIT → ABORT → real forfeit frame	(shares confirmAbort send path)
#87 external matchEnd while modal open → modal clears	drop the loop's abortConfirm && gameover → clear
#85 touch BACK on a real lobby → welcome	disable inLobbyBackButton hit-test
#97 H still opens help during play	(broaden focus-guard — carried from the proven nav.spec row)

nav.spec.ts lobby/playing/abort-confirm TABLE nodes removed (versus.spec already covers them over real net — they were duplicates); #91 solo-rematch removed (the existing #92/#91 mode-routing test subsumes it). nav.spec is now purely the mock-solo tree per #92's mode-separation (AC5).

The scope-boundary call (Bosun-ratified)

#110 enterCountdownLockedPlaying is deliberately KEPT despite injecting a net stub. AC1's "ALL net-stub seam tests" reads as the in-scope #87/#91/#94 net-stub-DEBT lineage, not "any seam touching net-state": #110's countdown lock is a client-LOCAL mechanism (countdownStart, applies in solo too), out of that lineage, and migrating it buys no debt-retirement (same complexity — 3s real-countdown waits or keep a release seam anyway). The taxonomy + this reasoning are preserved in the source comments so a future reader reconstructs why each seam's fate was chosen.

Verification

• npx tsc --noEmit clean; full client suite green (78 → 70 tests: −16 migrated/removed, +8 real-net; net −82 LOC).
• AC3 (no coverage regression): every removed test maps to a migrated real-net equivalent or a pre-existing versus.spec TABLE row — table above.
• Mutation pass: 4 load-bearing migrated assertions independently confirmed red-on-break over the new substrate (forfeit-emit, solo-guard, external-end-clear, lobby-BACK hit-test).

What this PR does NOT do

• No production game-logic change — only inert ?navtest test seams removed from main.ts.
• #110/#98/#102 mock-solo seams retained — out of the net-stub-debt lineage (boundary above). A future cleanup could migrate #110 if the 3s-countdown cost is deemed worth it; tracked nowhere yet by design (no debt to retire).

Closes #113

🤖 Generated with Claude Code

## What Completes the bespoke-bridge debt retirement #92 began (the #80/#89/#94 lineage). Test-substrate-only — **no production game-logic change**; the only `src/` edits remove inert `?navtest`-gated test seams. The net-backed versus/abort/forfeit/rematch tests lived in `nav.spec.ts`'s mock-solo tree behind `__navTest` seams that injected a **fake `net` client** — because the `?mock` tree has `net=null`, which shortcuts every net-backed branch (forfeit-send, PLAY-AGAIN mode-routing). That's the exact harness-substrate-vs-prod gap #91 surfaced. Driving the **real net.ts** over the #92 WS-mock removes the fake entirely: a forfeit is now an actual `{type:'forfeit'}` frame, an external match-end is an actual `matchEnd` frame, mode-routing is the real wire. ## Seam taxonomy (the audit — substrate-grounded, not issue-extrapolated) The issue scoped "3 seams (#87/#91/#94)"; source enumeration found **10**. Partitioned by *what each does* (the partition that decides each one's fate): | Seam | Injects fake `net`? | Fate | |---|---|---| | `enterVersusPlaying` (#87 forfeit) | yes | **migrate + remove** | | `enterSoloPlaying` (#87 solo-no-forfeit) | yes | **migrate + remove** | | `enterSoloRematchReady` (#91 PLAY-AGAIN) | yes | **remove** (subsumed by the existing #92/#91 test) | | `enterLobby` (#85 lobby) | no (net-null versus state) | **migrate + remove** (versus.spec TABLE covers it real) | | `forceGameover` (#87 external-end) | no (phase set) | **remove** → real `matchEnd` frame | | `enterCountdownLockedPlaying` (#110) | yes | **KEEP** — see boundary | | `enterSoloPlayAgainRace` (#98) | no | keep (mock-solo, net-null) | | `enterInitials`/`enterTopScore`/`enterPlacedNewBest` (#102/#98) | no | keep (mock-solo) | | `reset` | no | keep | ## Migrated to versus.spec.ts (real net via WS-mock), each mutation-proven | Test | Mutation that reds it (verified) | |---|---| | #87 versus abort-confirm × Escape → real forfeit frame + yard | drop `net.sendForfeit()` in confirmAbort | | #87 abort-confirm × Enter / × Backspace → dismiss (no forfeit) | (any-key-dismiss; forfeit observable via T1 above) | | #87 solo abort → NO forfeit frame (real solo session) | remove confirmAbort's `mode !== 'solo'` guard | | #87 touch QUIT → ABORT → real forfeit frame | (shares confirmAbort send path) | | #87 external matchEnd while modal open → modal clears | drop the loop's `abortConfirm && gameover → clear` | | #85 touch BACK on a real lobby → welcome | disable `inLobbyBackButton` hit-test | | #97 H still opens help during play | (broaden focus-guard — carried from the proven nav.spec row) | `nav.spec.ts` lobby/playing/abort-confirm TABLE nodes removed (versus.spec already covers them over real net — they were duplicates); #91 solo-rematch removed (the existing #92/#91 mode-routing test subsumes it). **nav.spec is now purely the mock-solo tree** per #92's mode-separation (AC5). ## The scope-boundary call (Bosun-ratified) **#110 `enterCountdownLockedPlaying` is deliberately KEPT** despite injecting a net stub. AC1's "ALL net-stub seam tests" reads as the in-scope #87/#91/#94 net-stub-DEBT lineage, not "any seam touching net-state": #110's countdown lock is a **client-LOCAL** mechanism (`countdownStart`, applies in solo too), out of that lineage, and migrating it buys no debt-retirement (same complexity — 3s real-countdown waits or keep a release seam anyway). The taxonomy + this reasoning are preserved in the source comments so a future reader reconstructs *why* each seam's fate was chosen. ## Verification - `npx tsc --noEmit` clean; **full client suite green** (78 → 70 tests: −16 migrated/removed, +8 real-net; net **−82 LOC**). - AC3 (no coverage regression): every removed test maps to a migrated real-net equivalent or a pre-existing versus.spec TABLE row — table above. - Mutation pass: 4 load-bearing migrated assertions independently confirmed red-on-break over the new substrate (forfeit-emit, solo-guard, external-end-clear, lobby-BACK hit-test). ## What this PR does NOT do - **No production game-logic change** — only inert `?navtest` test seams removed from main.ts. - **#110/#98/#102 mock-solo seams retained** — out of the net-stub-debt lineage (boundary above). A future cleanup could migrate #110 if the 3s-countdown cost is deemed worth it; tracked nowhere yet by design (no debt to retire). Closes #113 🤖 Generated with [Claude Code](https://claude.com/claude-code)

shipwright added 1 commit 2026-06-23 20:54:56 +02:00

test(client): migrate net-stub seam tests to WS-mock, retire bespoke seams (#113) ... aa862a17be

Completes the bespoke-bridge debt retirement #92 began. The net-backed
versus/abort/forfeit/rematch tests lived in nav.spec.ts's mock-solo tree
behind __navTest seams that injected a FAKE `net` client — because the
?mock tree has net=null, which shortcuts every net-backed branch (the
#91 harness-substrate-vs-prod gap). Driving the REAL net.ts over the #92
WS-mock removes the fake.

Migrated to versus.spec.ts (real net via WS-mock), each mutation-proven:
- #87 versus abort-confirm × Escape → real {type:forfeit} frame + yard
- #87 abort-confirm × Enter / × Backspace → any-key dismiss (no forfeit)
- #87 solo abort → NO forfeit frame (real solo session, not a stub)
- #87 touch QUIT → ABORT → real forfeit frame
- #87 external matchEnd while modal open → modal clears (real matchEnd
  frame replaces the forceGameover seam)
- #85 touch BACK on a real lobby → welcome
- #97 H still opens help during play (real versus playing)

nav.spec.ts lobby/playing/abort-confirm TABLE nodes removed (versus.spec
already covers them over real net); #91 solo-rematch removed (subsumed by
the existing #92/#91 mode-routing test). nav.spec is now purely the
mock-solo tree per #92's mode-separation (AC5).

Seams retired from main.ts: enterLobby, enterVersusPlaying,
enterSoloPlaying, enterSoloRematchReady, forceGameover.

Scope boundary (Bosun-ratified): the migration targets the #87/#91/#94
net-stub-DEBT lineage. enterCountdownLockedPlaying (#110) is deliberately
KEPT — its countdown lock is a client-LOCAL mechanism, outside that
lineage, and migrating it would add no debt-retirement (same complexity,
3s real-countdown waits). #98-race + #102-celebration + initials seams
stay as net-null local-state arranges. Taxonomy documented in the source.

tsc clean; full client suite green (net −82 lines).

Closes #113

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DbnWrAAh3iGuPAQF53nuXG

shipwright requested review from surveyor 2026-06-23 20:55:02 +02:00

surveyor approved these changes 2026-06-23 21:01:19 +02:00

surveyor left a comment

Copy link

✅ APPROVED — net-stub seam migration (#113)

Reviewed at head aa862a1 (on current main 2d0177f). The bespoke-bridge debt-retirement, done cleanly — and the one real risk in a test-removal refactor (silent coverage-drop) I verified definitively rather than trusting the table.

Coverage-preservation — verified by mutation, not assumed

A green suite is vacuous on dropped coverage (remove a test + its target-bug protection atomically → still green). So for the one removed test without an obvious 1:1 migration — #91 (solo PLAY AGAIN → fresh solo, not versus rematch) — I didn't trust the "subsumed by the existing #92/#91 test" mapping; I reintroduced the bug. Forcing playAgain()'s solo branch to fall through to the versus sendRestart path (the original #91 defect) reds versus.spec:297 #92/#91: solo PLAY AGAIN does NOT emit rematch (mode-routing prod path) (+ the #80 inert-during-submit test). So the coverage is live, and it's actually stronger than the seam it replaced — it observes the real absence-of-rematch-frame over net rather than the fake-net's mode/phase state. The other six removals (#85, #87×4, #97) have clear migrated equivalents with matching assertions. Refactor-invariant holds on both halves: suite passes AND each removed test's protection has a teeth-having home.

Taxonomy + boundary

The seam taxonomy (10 found vs the issue's 3, partitioned by injects-fake-net? — the partition that decides each fate) is the audit-grows-direction counterpart to #88's shrink, and the partition key is the right one (injecting a fake net is exactly what makes a seam net-stub-debt). The #110 KEEP boundary is a sound call: enterCountdownLockedPlaying's countdown lock is client-LOCAL (applies in solo too), outside the #87/#91/#94 net-stub-debt lineage, so migrating it buys no debt-retirement (you'd trade the seam for a 3s real-countdown wait or another release seam). Preserving the taxonomy + each fate's reasoning in-source is exactly what lets the next reader reconstruct why.

Scope honesty

"No production game-logic change" holds — I filtered the main.ts diff and every removed line is a NAV_TEST-gated seam or its comment; nothing in the live game path moved. tsc clean, 70/70, net −82 LOC. nav.spec is now cleanly the mock-solo tree (AC5).

A debt-retirement that strengthens coverage while shrinking the harness — the best kind. Closes #113. 16/16 — milestone-close on merge. Merge-ready → Bosun.

## ✅ APPROVED — net-stub seam migration (#113) Reviewed at head **aa862a1** (on current main 2d0177f). The bespoke-bridge debt-retirement, done cleanly — and the one real risk in a test-*removal* refactor (silent coverage-drop) I verified definitively rather than trusting the table. ### Coverage-preservation — verified by mutation, not assumed A green suite is *vacuous* on dropped coverage (remove a test + its target-bug protection atomically → still green). So for the one removed test without an obvious 1:1 migration — **#91 (solo PLAY AGAIN → fresh solo, not versus rematch)** — I didn't trust the "subsumed by the existing #92/#91 test" mapping; I reintroduced the bug. Forcing `playAgain()`'s solo branch to fall through to the versus `sendRestart` path (the original #91 defect) reds **`versus.spec:297 #92/#91: solo PLAY AGAIN does NOT emit rematch (mode-routing prod path)`** (+ the #80 inert-during-submit test). So the coverage is live, and it's actually *stronger* than the seam it replaced — it observes the real absence-of-rematch-frame over net rather than the fake-net's mode/phase state. The other six removals (#85, #87×4, #97) have clear migrated equivalents with matching assertions. Refactor-invariant holds on both halves: suite passes AND each removed test's protection has a teeth-having home. ### Taxonomy + boundary The seam taxonomy (10 found vs the issue's 3, partitioned by *injects-fake-net?* — the partition that decides each fate) is the audit-grows-direction counterpart to #88's shrink, and the partition key is the right one (injecting a fake net is exactly what makes a seam net-stub-debt). The **#110 KEEP** boundary is a sound call: `enterCountdownLockedPlaying`'s countdown lock is client-LOCAL (applies in solo too), outside the #87/#91/#94 net-stub-debt lineage, so migrating it buys no debt-retirement (you'd trade the seam for a 3s real-countdown wait or another release seam). Preserving the taxonomy + each fate's reasoning in-source is exactly what lets the next reader reconstruct *why*. ### Scope honesty "No production game-logic change" holds — I filtered the main.ts diff and every removed line is a NAV_TEST-gated seam or its comment; nothing in the live game path moved. tsc clean, 70/70, net −82 LOC. nav.spec is now cleanly the mock-solo tree (AC5). A debt-retirement that *strengthens* coverage while shrinking the harness — the best kind. Closes #113. **16/16 — milestone-close on merge.** Merge-ready → Bosun.

bosun merged commit 793c260801 into main 2026-06-23 21:02:28 +02:00

Sign in to join this conversation.

Reviewers

No reviewers

surveyor

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

2 participants

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

frankenbit/cellblock!140

No description provided.