[rustjava-2026-09-19-adoption-audit-prune] chore: 09-19 예외 클래스 사슬 군살 제거 — 출력순서 검사기 삭제 · 블라인드 스팟 보고 축소 · 낡은 주석 정정

.github/workflows/rust.yml

@@ -81,18 +81,9 @@ jobs:
       - uses: actions/checkout@v7
       - run: python3 scripts/check-dod-ci-parity.py
 
-  # a checker that iterates a set prints its findings in a different order on different runs, so two
-  # rounds cannot diff their output — and every other axis stays green while it does. This reads the
-  # checkers' AST rather than re-running them. One runner, not the matrix.
-  script_output_order:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v7
-      - run: python3 scripts/check-script-output-order.py
-
-  # Jvm::exception unwraps new_class(), so naming a class the loader cannot resolve panics instead
-  # of throwing. This compares the names against the registered protos (see the script's docstring
-  # for what it cannot see). One runner, not the matrix.
+  # Jvm::exception returns whatever new_class() failed with, so naming a class the loader cannot
+  # resolve raises NoClassDefFoundError instead of the intended exception. This compares the literal
+  # names against the registered protos. One runner, not the matrix.
   named_exception_classes:
     runs-on: ubuntu-latest
     steps:

CLAUDE.md

@@ -33,7 +33,6 @@
   python3 scripts/check-dod-ci-parity.py
   python3 scripts/check-named-exception-classes-are-loadable.py
   python3 scripts/check-merge-dropped-symbols.py
-  python3 scripts/check-script-output-order.py
   ```
   ★★**이 블록은 이제 «기계가 지킨다» — `scripts/check-dod-ci-parity.py`(CI job `dod_parity`)가
   이 코드블록과 `rust.yml` 을 «각각 파싱해» 대칭차를 낸다.** 어긋나면 그 자리에서 red 다.

docs/worklog/2026-09-25-adoption-audit-prune.md

@@ -0,0 +1,19 @@
+## [2026-09-25] 09-19 예외 클래스 사슬 군살 제거 — 검사기 1개 삭제 · 보고 1개 축소 · 낡은 주석 정정 (rustjava-2026-09-19-adoption-audit-prune)
+
+- **무엇을**: `scripts/check-script-output-order.py` 와 그 CI 잡 `script_output_order`, DoD 줄을 지웠다.
+  `check-named-exception-classes-are-loadable.py` 에서 블라인드 스팟 보고(비리터럴 호출처 계수)를 빼고
+  스캔을 `NAMED.finditer` 한 번으로 줄였다. `rust.yml`·스크립트 머리의 «unwrap → panic» 서술(#76 이후 거짓)과
+  `jvm.rs` 의 회차 이력 주석을 불변식만 남기고 줄였다. `test_exception_fallback_recursion.rs` 의 `asked == 2` 를
+  `asked < 20` 으로 풀었다.
+- **왜**: 사슬 8 PR 에서 검사기가 잡은 것 0(rust.yml 최근 100회 전부 success) · 출력순서 검사기는 현 위반 0에
+  후속 제안도 기각됐고, 블라인드 스팟 보고가 가리키는 1곳은 테스트다. 주석의 회차 이력은 git/worklog 가 이미 가진다.
+- **사용자 영향**: 없음 — 제품 동작 변경 0. 검사기 출력은 블라인드 스팟 3~4줄이 빠지고 판정 줄(43 / 852 / 268)은 같다.
+
+### 판정 메모
+- `asked == 1`(두 곳)은 **남겼다**: 카운터는 숨긴 이름에 대한 질의만 센다(`test-utils` `HidesOneClass`) —
+  «첫 질의에서 멈추고 다시 묻지 않는다»가 불변식 자체다. `asked == 2` 는 반복 가드가 «몇 번째에» 걸리는지를
+  고정하므로 «상한(20) 전에 멈췄다»로 완화했다.
+- 후속 제안 접기(계약 6): 이 사슬(`2026-09-19-*`·`2026-09-20-*`)의 제안은 **전건 이미 처분**돼 있다
+  (12건 = 채택 10 · 09-21 기각 2) ⇒ 접을 카드 0, `.json` 을 쓰지 않았다.
+- 회차 기록은 이 파일에만 적었다 — `REPORT.md`·`STATE.md` 를 동결하는 PR #97 이 열려 있어, 거기 적으면
+  어느 쪽이 먼저 착지하든 충돌 또는 해시 잠금 red 가 된다.

jvm/src/jvm.rs

@@ -93,15 +93,8 @@ impl Jvm {
             "java/lang/Class",
         ];
         for class_name in bootstrap_classes.iter() {
-            // Fails like the closure walk below, and for the same reason -- nothing can be *raised*
-            // yet, this is what loads the classes an exception is made of -- so the error is
-            // `Unraisable` rather than a Java exception. It has to say which name it was, which the
-            // `unwrap` this once was did not: a host with a gap in its class set read
-            // `called Option::unwrap() on a None value` and had to bisect this list to find out
-            // which of six. Measured by last round's sweep: 5 of the 12 named refusals were this
-            // line, and two of those names (`java/lang/Object`, `java/io/Serializable`) are also in
-            // the error path's closure, so the same gap got a good message or a useless one
-            // depending only on which loop reached it first.
+            // Nothing can be *raised* yet -- these are the classes an exception is made of -- so a gap
+            // is `Unraisable`, and it names the class so the host does not have to bisect this list.
             let Some(class_definition) = jvm.inner.bootstrap_class_loader.load_class(&jvm, class_name).await? else {
                 return Err(JavaError::Unraisable(format!(
                     "the class set has no {class_name}, which is one of the {} classes loaded before \
@@ -130,35 +123,14 @@ impl Jvm {
 
         // Everything the error path needs before anything at all can be raised.
         //
-        // `Jvm::exception` builds a message with `java/lang/String` and then an instance of
-        // `java/lang/NoClassDefFoundError`, so a class set missing either cannot report even its own
-        // gap -- reporting the absent String needs a String, and reporting the absent reporter needs
-        // the reporter. Two earlier rounds each closed one of those names and each found the next by
-        // reading the code and guessing. Then the whole question was measured at once, by hiding every
-        // name construction asks the loader for, one at a time (`jvm/tests/test_error_path_class_sweep.rs`,
-        // 37 non-array names): five *more* recurse with no floor -- `java/lang/Throwable`,
-        // `java/lang/Error`, `java/lang/LinkageError`, `java/lang/CharSequence` and
-        // `java/lang/Comparable` -- and they are exactly the supertype and interface closure of those
-        // two. Of course they are: resolving a class resolves its supertypes, and a gap found *there*
-        // is reported with the very classes still being resolved. The closure is 9 names and the
-        // account of it is complete: 2 were already checked, 5 recursed, and the remaining 2
-        // (`java/lang/Object`, `java/io/Serializable`) are in `bootstrap_classes` above, so they fail
-        // before this runs -- naming themselves there, as of the round that closed that gap.
-        //
-        // So the closure is what is checked, not a list someone has to remember to extend. Asked of
-        // the loader directly and never through `resolve_class`, because the reporting path cannot
-        // report *this* failure: building the report is the thing that is missing. A bare
-        // `resolve_class` here would hand the question to `Jvm::exception`, which is the cycle itself
-        // -- measured, that is still `stack overflow, aborting`, only during construction instead of
-        // later.
-        //
-        // Start-up cost, counted rather than timed (the previous round measured wall clock here and
-        // discarded it as below this host's noise): the walk asks the loader 9 times where the two
-        // asserts it replaces asked twice, so construction goes from 44 loader questions to 51.
-        //
-        // Here rather than in `bootstrap_classes` above, and before the properties loop below: that
-        // loop is the first thing in construction that needs a String, and resolution runs class
-        // initialisation, which needs the thread attached above.
+        // `Jvm::exception` builds a `java/lang/String` message and a `java/lang/NoClassDefFoundError`
+        // instance, and resolving those resolves their supertype and interface closure. A gap anywhere
+        // in that closure cannot be reported -- the report needs the missing class -- so it would
+        // recurse until the stack overflows. The closure is walked here instead of a fixed list, and
+        // asked of the loader directly: `resolve_class` would hand the failure to `Jvm::exception`,
+        // which is the cycle itself. It runs after the thread is attached (initialisation needs it) and
+        // before the properties loop below, the first thing that needs a String.
+        // `jvm/tests/test_error_path_class_sweep.rs` hides each name construction asks for.
         let mut pending = Vec::from(["java/lang/String".to_owned(), "java/lang/NoClassDefFoundError".to_owned()]);
         let mut asked = HashSet::new();
         while let Some(class_name) = pending.pop() {

jvm/tests/test_exception_fallback_recursion.rs

@@ -63,14 +63,15 @@ async fn a_class_set_missing_a_bootstrap_class_is_an_error_naming_it() {
 // that calls `fillInStackTrace` again. Before `Jvm::exception` refused to build an exception it was
 // already building on the same thread, this cap (the sweep's 20) overflowed the default 2 MiB test
 // stack -- `stack overflow, aborting`, rc 134 -- and the run said nothing about the first failure. Now
-// the repeat returns `Unraisable` naming the exception the thread started with, and the loader is
-// asked twice: once for the first failure, once by the construction that repeated it.
+// the repeat returns `Unraisable` naming the exception the thread started with. The invariant is
+// that this happens before the cap: how many questions the guard needs first is its business
+// (today 2 -- the first failure and the repeat), and pinning that number would pin the guard's depth.
 #[tokio::test]
 async fn a_failure_while_raising_is_reported_instead_of_recursing() {
     let (message, asked) = unraisable_hiding("[Ljava/lang/String;", 20).await;
     assert!(
         message.contains("into raising java/lang/NoClassDefFoundError ([Ljava/lang/String;), which is the first failure"),
         "{message}"
     );
-    assert_eq!(asked, 2);
+    assert!(asked < 20, "asked {asked} times: the guard did not stop it before the cap");
 }