[rustjava-bootstrap-argument-diagnostic-names-index-and-tag] fix(classfile): 부트스트랩 인자 거부가 어느 인자·무엇을 찾았는지 말한다 (#73)

[rustjava-bootstrap-argument-diagnostic-names-index-and-tag] fix(classfile): 부트스트랩 인자 거부가 어느 인자·무엇을 찾았는지 말한다

Author: Jun025

REPORT.md

@@ -204,6 +204,19 @@
 - 검증: 검사기 `✓ 43 named … all 268 loadable` rc=0 · `check-worklog-json` rc=0 · `check-dod-ci-parity` rc=0(명령 7개) · `cargo fmt` rc=0 · `mbvar-guard` rc=0(위반 0).
 - ★후속 추천: **loadable 집합을 «파싱으로 재유도»하지 말고 loader 쪽에서 «내보낼» 것인가**(M — 이번 결함 3건이 전부 그 재유도 자리였다). 상세 = `docs/worklog/2026-09-18-named-exception-classes-are-loadable.md`.
 
+## [2026-09-18] 부트스트랩 인자 거부가 «어느 인자·무엇을 찾았는지» 말한다 (rustjava-bootstrap-argument-diagnostic-names-index-and-tag)
+- 무엇을: 채택 제안 `2026-09-18-bootstrap-argument-diagnostic-sequencing#p0`(worklog json 기록). ★**새 기능이 아니라 «이미 계산된 값을 버리지 않는 것»** — 술어가 `false` 를 내는 그 자리에 index 와 찾은 항목이 **손에 있었다**.
+- ★**ⓑ 대안을 만들지 않았다** — 같은 enum 의 `UnsupportedVersion(u16)` 이 **이미 3층을 관통해 경계에서 `format!` 되는 패턴**이라 그것을 **복제**했다(새 진단 체계 **0**).
+- ★**전/후**(같은 픽스처 `test-data/ldc/LdcDynamicBSMArgPastEnd.class` · CLI 실제 출력):
+  `BEFORE  java.lang.ClassFormatError: a bootstrap method argument names nothing or is not a loadable constant`
+  `AFTER   java.lang.ClassFormatError: bootstrap method #0 argument #0 names no constant pool entry`
+- ★**세 요구 3/3**: expected(문면) · index(`argument_index` ★+`method_index` — OpenJDK 는 그걸 안 말해 여럿일 때 모호하다) · actual(가리킨 상수의 **종류 이름**). ★태그를 **번호가 아니라 이름**으로 나른다 — 태그 바이트가 파싱 후 남지 않아 번호는 «아무도 분기 안 하는 두 번째 표»를 만들게 된다.
+- ★★**양방향의 급소** — `StringConcat` 은 인자가 **1개**라 index 0 이 **계산이든 하드코딩이든 통과한다**. 그래서 정적 인자 **3개**짜리 `Lambda.class` 의 **#0 과 #2** 를 각각 망가뜨려 ★**보고된 index 가 0 ↔ 2 로 따라가는 것**을 잠갔다.
+- ★**소비자 전수**: `InvalidFormat` 34사용처 중 **구조적 소비자는 1개**(`jvm-bytecode` 의 `From` match) — **arm 을 더했을 뿐 고치지 않았다**. 나머지 14규칙 무변.
+- ★**대가**(재서 적는다): ★**타입이 «커지지 않았다»** — `Copy` 유지 · 크기 불변(시험으로 잠금) ⇒ 할당 0. ※제안이 경계한 「소유 데이터로 커진다」는 **이 설계에선 일어나지 않았다**. ★**두 경로 공존**의 혼동은 변형 docstring 한 줄로 못박았다 · ★이 규칙의 **메시지 문면이 바뀌어** 그것을 단언하던 시험 2곳과 grep 습관이 깨진다.
+- 검증: `cargo test -p classfile` **16 passed**(전 13) · `--test test_class_format` **22 passed** · `cargo test --all` rc=0.
+- ★후속 추천: ⑴**다른 14규칙 중 «수를 아는데 버리는» 것을 세라**(S — 아직 그 수를 모른다) ⑵`ClassDefinitionError` 의 `&'static str` 두 변형도 같은 한계(M · 이 회차는 나란히 더해 **우회**했다). 상세 = `docs/worklog/2026-09-18-bootstrap-argument-index-and-tag.md`.
+
 ## [2026-09-18] 이름으로 부르는 예외 클래스가 «실을 수 있는» 것인가 — 한 자리에서 대조한다 (rustjava-lock-every-named-exception-class-is-loadable)
 - 무엇을: 채택 제안 `2026-09-17-string-concat-recipe-arity#p0`(worklog json `adoptedProposals` 기록). 산출물 = `scripts/check-named-exception-classes-are-loadable.py` **한 자리** + CI job + DoD 한 줄. ★**런타임 클래스 추가 0 · `.unwrap()` 무접촉.**
 - ★**전제를 코드로 확인했다**(총괄 선실측 없음): `jvm/src/jvm.rs:943-950` 의 `new_class(...).await.`★**`unwrap()`** ⇒ 부트스트랩 로더가 이름을 못 풀면 **Java 예외가 아니라 프로세스가 죽는다**. ★**자기 참조다** — `:842` 가 클래스 부재를 `exception("java/lang/NoClassDefFoundError", …)` 로 보고하므로 **오류 경로 자신의 클래스**가 실려야 한다.

STATE.md

@@ -122,6 +122,14 @@
   ★**수 전/후**: 이름 41→**43** · 호출부 812→**846** · loadable 263→**268**(왜인지 기재).
   ★**ⓒ 편집 «전»에 새 red 위험 측정** — 늘어나는 2이름 모두 등재 ⇒ **새 red 0**.
   ★**시간 유의차 없음**(전 1.42/1.67/1.46 ↔ 후 2.01/1.69/1.19 · 구간 겹침). ★런타임 클래스 추가 0 · `protos` 무접촉.
+- [rustjava-bootstrap-argument-diagnostic-names-index-and-tag] ★★**부트스트랩 인자 거부가 «어느 인자·무엇을» 말한다.** 채택 제안 `2026-09-18-bootstrap-argument-diagnostic-sequencing#p0`.
+  ★**전제 확인(코드)**: 술어가 `false` 를 내는 자리에 `index` 와 `constant_pool.get(index)` 가 **손에 있었다** — `bool` 이 둘을 버렸다.
+  ★**새 체계 0** — 같은 enum 의 `UnsupportedVersion(u16)` 이 이미 3층을 관통하는 패턴이라 **복제**했다.
+  ★**전/후**: `ClassFormatError: a bootstrap method argument names nothing or is not a loadable constant` → ★`ClassFormatError: bootstrap method #0 argument #0 names no constant pool entry`(CLI 실측).
+  ★**세 요구 3/3**(expected·index+method_index·actual) · 태그는 **이름**으로(번호는 표가 하나 더 는다).
+  ★**양방향 급소**: `StringConcat` 은 인자 1개라 index 0 이 하드코딩이어도 통과 ⇒ **인자 3개 `Lambda.class` 의 #0↔#2** 로 **index 가 따라가는 것**을 잠갔다.
+  ★**소비자 1개**(`jvm-bytecode` `From`) — arm 추가뿐, 14규칙 무변. ★**타입 «안 커졌다»**(`Copy`·크기 불변 시험) · ★메시지 문면이 바뀌어 단언 2곳이 바뀐다.
+  ★`-p classfile` **16 passed**(전 13) · `test_class_format` **22** · `--all` rc=0.
 - [rustjava-lock-every-named-exception-class-is-loadable] ★★**이름으로 부르는 예외 클래스가 «실을 수 있는» 것인가 — 대조 한 자리.** 채택 제안 `2026-09-17-string-concat-recipe-arity#p0`(worklog json 기록).
   ★**전제 확인(코드)**: `jvm/src/jvm.rs:943-950` `new_class(...).await.`**`unwrap()`** ⇒ 못 싣는 이름은 **throw 가 아니라 패닉**. ★자기 참조 — `:842` 가 부재를 `exception("java/lang/NoClassDefFoundError")` 로 보고한다.
   ★**베이스라인 0**(★게이트² 정정 후): `exception(` 리터럴 **43 고유 / 846 호출부** ↔ 등재 **268**(`as_proto` 265 + `list_proto` 3). ★초판의 41/812/263 은 **전부 과소**였다(다중 줄 34 · `list_proto` 3 · 짧은 이름 충돌). ★제안의 「72」는 여전히 **재현 안 됨**.

classfile/src/error.rs

@@ -10,5 +10,26 @@
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
 pub enum ClassFileError {
     InvalidFormat(&'static str),
+    /// A bootstrap method argument that is not a loadable constant, carrying the two things the
+    /// predicate already knows at the moment it refuses: *which* argument, and what it found there.
+    ///
+    /// ★ Which variant to use: `InvalidFormat` is for a rule that has nothing to point at, and that
+    /// is still most of them. Use this one only when a number is already in hand — folding it into
+    /// prose is the thing this variant exists to stop. `UnsupportedVersion` is the same shape and
+    /// predates it, so this is the established way here rather than a second scheme.
+    ///
+    /// It stays `Copy`: two `u16`s and a `&'static str`, no owned data.
+    InvalidBootstrapArgument {
+        /// Position of the `BootstrapMethods` entry, zero-based, as the attribute stores them.
+        method_index: u16,
+        /// Position within that entry's argument list, zero-based. OpenJDK calls this
+        /// `argument_index` and prints it without the method, which is ambiguous when a class has
+        /// more than one bootstrap method; both are carried here for that reason.
+        argument_index: u16,
+        /// The kind of constant the argument actually names, or `None` when the index names no
+        /// pool entry at all — "names nothing" and "names the wrong kind" are different failures
+        /// and the message says which.
+        actual: Option<&'static str>,
+    },
     UnsupportedVersion(u16),
 }

classfile/src/validation.rs

@@ -38,15 +38,10 @@ pub(crate) fn validate_class(class: &ClassInfo) -> Result<(), ClassFileError> {
             "class file version does not support a constant tag it carries",
         ));
     }
-    if !bootstrap_method_static_arguments_are_in_the_pool(class) {
-        return Err(ClassFileError::InvalidFormat(
-            // The rule is wider than the function name: the docstring above says the argument must also
-            // be a loadable constant, and OpenJDK says the same ("bad constant type"). The name stayed
-            // behind when the rule widened; renaming it is not this round's scope, so the cause is what
-            // gets the wording right.
-            "a bootstrap method argument names nothing or is not a loadable constant",
-        ));
-    }
+    // The rule is wider than the function name: the docstring below says the argument must also be a
+    // loadable constant, and OpenJDK says the same ("bad constant type"). The name stayed behind when
+    // the rule widened; renaming it is not this round's scope.
+    bootstrap_method_static_arguments_are_in_the_pool(class)?;
     if !bootstrap_method_indices_resolve(class) {
         return Err(ClassFileError::InvalidFormat("a dynamic constant names no bootstrap method"));
     }
@@ -165,11 +160,18 @@ fn constant_pool_tags_fit_the_class_file_version(class: &ClassInfo) -> bool {
 /// One consequence worth naming: a long or double occupies two pool slots and only the first is
 /// usable (JVMS 4.4.5), so the second has no entry in this map and an argument naming it is
 /// rejected. That is the intended reading of "valid index" rather than an accident of the map.
-fn bootstrap_method_static_arguments_are_in_the_pool(class: &ClassInfo) -> bool {
-    class.attributes.iter().all(|attribute| match attribute {
-        AttributeInfo::BootstrapMethods(methods) => methods.iter().all(|method| {
-            method.arguments.iter().all(|index| {
-                class.constant_pool.get(index).is_some_and(|item| {
+/// ★ It returns `Result` rather than `bool` for one reason: at the moment it refuses, it is holding
+/// the position of the argument and the entry it found, and a `bool` throws both away. The caller
+/// then had to describe the failure in prose it did not measure. Nothing else about the rule moved.
+fn bootstrap_method_static_arguments_are_in_the_pool(class: &ClassInfo) -> Result<(), ClassFileError> {
+    for attribute in &class.attributes {
+        let AttributeInfo::BootstrapMethods(methods) = attribute else {
+            continue;
+        };
+        for (method_index, method) in methods.iter().enumerate() {
+            for (argument_index, index) in method.arguments.iter().enumerate() {
+                let entry = class.constant_pool.get(index);
+                let loadable = entry.is_some_and(|item| {
                     matches!(
                         item,
                         ConstantPoolItem::Integer(_)
@@ -182,11 +184,43 @@ fn bootstrap_method_static_arguments_are_in_the_pool(class: &ClassInfo) -> bool
                             | ConstantPoolItem::MethodType { .. }
                             | ConstantPoolItem::Dynamic { .. }
                     )
-                })
-            })
-        }),
-        _ => true,
-    })
+                });
+                if !loadable {
+                    return Err(ClassFileError::InvalidBootstrapArgument {
+                        method_index: method_index as u16,
+                        argument_index: argument_index as u16,
+                        actual: entry.map(constant_kind_name),
+                    });
+                }
+            }
+        }
+    }
+    Ok(())
+}
+
+/// The JVMS 4.4 name of a constant's kind, for saying what an argument actually named.
+///
+/// ★ A name and not the numeric tag: the tag byte is not kept after parsing, so reporting it would
+/// mean a second table to hold numbers nothing branches on. The name is what a reader of the
+/// message needs, and it is derived from the variant rather than stored, so it cannot drift.
+fn constant_kind_name(item: &ConstantPoolItem) -> &'static str {
+    match item {
+        ConstantPoolItem::Utf8(_) => "Utf8",
+        ConstantPoolItem::Integer(_) => "Integer",
+        ConstantPoolItem::Float(_) => "Float",
+        ConstantPoolItem::Long(_) => "Long",
+        ConstantPoolItem::Double(_) => "Double",
+        ConstantPoolItem::Class { .. } => "Class",
+        ConstantPoolItem::String { .. } => "String",
+        ConstantPoolItem::Fieldref { .. } => "Fieldref",
+        ConstantPoolItem::Methodref { .. } => "Methodref",
+        ConstantPoolItem::InterfaceMethodref { .. } => "InterfaceMethodref",
+        ConstantPoolItem::NameAndType { .. } => "NameAndType",
+        ConstantPoolItem::MethodHandle { .. } => "MethodHandle",
+        ConstantPoolItem::MethodType { .. } => "MethodType",
+        ConstantPoolItem::Dynamic { .. } => "Dynamic",
+        ConstantPoolItem::InvokeDynamic { .. } => "InvokeDynamic",
+    }
 }
 
 /// JVMS 4.4.10 and 4.7.23: `bootstrap_method_attr_index` is an index into the `bootstrap_methods`

classfile/tests/test.rs

@@ -367,9 +367,96 @@ fn test_a_bootstrap_argument_that_is_not_a_loadable_constant_is_rejected() {
 
     assert_eq!(
         ClassInfo::parse(&mutated).err(),
-        Some(ClassFileError::InvalidFormat(
-            "a bootstrap method argument names nothing or is not a loadable constant"
-        )),
-        "a bootstrap argument naming a Utf8 is not a loadable constant"
+        Some(ClassFileError::InvalidBootstrapArgument {
+            method_index: 0,
+            argument_index: 0,
+            actual: Some("Utf8"),
+        }),
+        "a bootstrap argument naming a Utf8 is not a loadable constant, and the refusal says which argument and what it found"
     );
 }
+
+/// ★ The index has to *follow the input*, not be a constant that happens to read correctly.
+///
+/// `StringConcat` above has a single argument, so its `argument_index` is 0 whether the code
+/// computes it or hardcodes it — that test cannot tell the two apart. A lambda's bootstrap method
+/// takes three (samMethodType, implMethod, instantiatedMethodType), so breaking a different one has
+/// to move the number. Both mutations are the same edit at a different offset, which is what makes
+/// the pair a control for each other.
+#[test]
+fn test_the_reported_bootstrap_argument_index_follows_the_broken_argument() {
+    let lambda = include_bytes!("../../test-data/indy/Lambda.class");
+    let class = ClassInfo::parse(lambda).unwrap();
+    let arguments = &bootstrap_methods(&class)[0].arguments;
+    assert_eq!(arguments.len(), 3, "a LambdaMetafactory call site carries three static arguments");
+
+    // Derived from the parse rather than hardcoded, so a regenerated fixture moves the window with it.
+    let mut window = vec![0u8, arguments.len() as u8];
+    for argument in arguments {
+        window.extend_from_slice(&argument.to_be_bytes());
+    }
+    let at = lambda
+        .windows(window.len())
+        .position(|candidate| candidate == window)
+        .expect("test-data/indy/Lambda.class layout changed; the BootstrapMethods argument list moved");
+
+    // Entry #4 is a Utf8 ("java/lang/Object"): a real pool entry that is not a loadable constant.
+    // The kind is not assumed — the assertion below reads it back, so a regenerated fixture whose
+    // #4 is something else fails here rather than passing for the wrong reason.
+    let utf8: u16 = 4;
+    assert!(class.constant_pool.get(&utf8).is_some(), "the replacement index must be in the pool");
+
+    for broken in [0usize, 2usize] {
+        let mut mutated = lambda.to_vec();
+        let offset = at + 2 + broken * 2;
+        mutated[offset..offset + 2].copy_from_slice(&utf8.to_be_bytes());
+
+        assert_eq!(
+            ClassInfo::parse(&mutated).err(),
+            Some(ClassFileError::InvalidBootstrapArgument {
+                method_index: 0,
+                argument_index: broken as u16,
+                actual: Some("Utf8"),
+            }),
+            "breaking argument #{broken} must report argument #{broken}"
+        );
+    }
+}
+
+/// The other half of the rule: an index that names no entry at all reports `actual: None`, and the
+/// message says "names no constant pool entry" rather than guessing a kind.
+#[test]
+fn test_a_bootstrap_argument_naming_nothing_reports_no_kind() {
+    let string_concat = include_bytes!("../../test-data/indy/StringConcat.class");
+    let class = ClassInfo::parse(string_concat).unwrap();
+    let window = [0u8, 1, 0, 34, 0, 1, 0, 32];
+    let at = string_concat
+        .windows(window.len())
+        .position(|candidate| candidate == window)
+        .expect("test-data/indy/StringConcat.class layout changed; the BootstrapMethods entry moved");
+
+    let past_end = u16::MAX;
+    assert!(class.constant_pool.get(&past_end).is_none(), "the index must name nothing");
+
+    let mut mutated = string_concat.to_vec();
+    mutated[at + 6..at + 8].copy_from_slice(&past_end.to_be_bytes());
+
+    assert_eq!(
+        ClassInfo::parse(&mutated).err(),
+        Some(ClassFileError::InvalidBootstrapArgument {
+            method_index: 0,
+            argument_index: 0,
+            actual: None,
+        })
+    );
+}
+
+/// The cost of the variant, measured rather than asserted in prose: it stays `Copy` and the enum
+/// does not grow, because two `u16`s and a `&'static str` fit in the space `InvalidFormat` already
+/// needed for its string.
+#[test]
+fn test_the_structured_variant_does_not_grow_the_error_type() {
+    assert_eq!(size_of::<ClassFileError>(), size_of::<&'static str>() + size_of::<usize>());
+    fn assert_copy<T: Copy>() {}
+    assert_copy::<ClassFileError>();
+}