Closed

CHANGELOG-ZH.md

@@ -1,3 +1,20 @@
+## 0.6.9
+
+- Matlab `sparse(A)` 现可接受静态 shape 的 logical rank-2 数组，并产生类型化 logical CSC 矩阵。
+- 推断 shape、显式 shape 与 `nzmax` 三类 R2024 triplet 构造形式现均支持 logical value。
+- 重复 logical triplet 坐标按照 Matlab 语义使用 `any` 聚合，不再执行数值加法。
+- false logical value 不会作为显式 entry 保留在 canonical CSC 存储中。
+- 稀疏普通转置与共轭转置现可保持 logical 值类型。
+- scalar、linear 与 Cartesian submatrix selection 会在适用结果上保持 logical sparse 存储和类型。
+- indexed assignment、扩容与删除会保持 logical CSC；写入 false 会删除对应存储项。
+- sparse reshape 会保持 logical value、Matlab 列主序、推断维度和零 extent。
+- logical sparse 输入经 `full` 转换后返回 logical dense 数组，`issparse` 与 `nnz` 保持预期行为。
+- 生成的 JavaScript 为 sparse value domain 携带显式 tag，并在使用前拒绝损坏的 logical CSC value。
+- 生成的 C++17 使用 `sparse_matrix<bool>` 表示 logical CSC，并安全处理压缩 logical vector 存储。
+- JavaScript 与 C++17 使用彼此独立的类型化 sparse runtime，不要求先生成另一目标。
+- 修复引入类型化 sparse conversion ABI 后，sparse RHS 方阵求解的转换错误。
+- sparse assignment 失败现在会在生成的 JavaScript 与 C++17 中使用一致的 real-or-logical 诊断。
+
 ## 0.6.8
 
 - Matlab 稀疏矩阵现可在已支持的静态 finite-real rank-2 能力中完整保持零 extent。
@@ -13,8 +30,6 @@
 - 双目标 LIR v34 新增显式 runtime shape 调用 ABI，使 renderer 只序列化已验证调用，不再恢复 semantic storage 或 shape policy。
 - source map 会保留全部新增稀疏操作以及零维左除、右除的原始调用位置。
 - 新增零 extent sparse 的双目标差分执行、生成计划拒错、跨层损坏、fuzz 与架构覆盖。
-- 验证基线现包含 268 项 C++ 测试、98 个差分 case、127 项 CTest 和 18 项生成 runtime 拒错测试。
-- 生产源码覆盖率为 91.69%（37,154/40,520 行），既有 sparse-solve 性能预算无需放宽即可通过。
 
 ## 0.6.7
 
@@ -29,10 +44,8 @@
 - source map 会为全部稀疏 operand 组合保留原始 `.*` 表达式位置。
 - complex、零 extent、动态 shape 或不兼容的稀疏逐元素运算继续失败关闭，不会静默改变语义。
 - 新增可执行 Matlab 示例，在两个输出目标间比较数值、shape 与稀疏存储。
-- 验证基线现包含 258 项 C++ 测试、97 个差分 case、124 项 CTest 和 16 项生成 runtime 拒错测试。
 - Matlab fuzz corpus 现覆盖稀疏逐元素语法、操作数方向、广播与 runtime 验证边界。
 - 新增独立 sparse element-wise benchmark，约束编译延迟、吞吐、arena 峰值与生成代码大小。
-- 生产源码覆盖率为 91.59%（36,732/40,103 行），高于 85% 硬门槛。
 
 ## 0.6.6
 
@@ -110,7 +123,7 @@
 - 跨层 verifier 会在发射前拒绝损坏的 sparse-index identity、arity、shape、type、storage 或 inactive-state fact。
 - 当 storage representation 已知时，`full` 现可接受结果 extent 为动态值或零值的 rank-two dense/CSC selection。
 - sparse assignment、超过两个 selector、complex/sparse selector、N 维 linear result，以及动态、空或 complex sparse source 继续以稳定诊断失败关闭。
-- 新增可执行 sparse-indexing 示例，覆盖双目标行为、source-map 保留、越界拒绝与 fuzz 回归；生产代码行覆盖率为 91.24%（33,780/37,023）。
+- 新增可执行 sparse-indexing 示例，覆盖双目标行为、source-map 保留、越界拒绝与 fuzz 回归。
 
 ## 0.6.1
 

CHANGELOG.md

@@ -1,3 +1,20 @@
+## 0.6.9
+
+- Matlab `sparse(A)` now accepts statically shaped logical rank-two arrays and produces typed logical CSC matrices.
+- Logical triplet constructors are supported across the inferred, explicitly sized, and `nzmax` R2024 call forms.
+- Duplicate logical triplets now follow Matlab semantics by combining values with `any` instead of numeric addition.
+- False logical values are omitted from canonical CSC storage rather than retained as explicit entries.
+- Ordinary and conjugating sparse transpose preserve the logical value class.
+- Scalar, linear, and Cartesian submatrix selection preserve logical sparse storage and result type where applicable.
+- Indexed assignment, growth, and deletion preserve logical CSC matrices, with false assignments removing stored entries.
+- Sparse reshape preserves logical values, Matlab column-major order, inferred dimensions, and zero extents.
+- `full` returns logical dense arrays for logical sparse inputs, while `issparse` and `nnz` retain their expected behavior.
+- Generated JavaScript carries an explicit sparse value-domain tag and rejects malformed logical CSC values before use.
+- Generated C++17 uses `sparse_matrix<bool>` for logical CSC values and safely handles packed logical-vector storage.
+- JavaScript and C++17 use independent typed sparse runtimes and do not require one target to be generated before the other.
+- Fixed sparse square solves with sparse right-hand sides after introducing the typed sparse conversion ABI.
+- Sparse assignment failures now use consistent real-or-logical diagnostics across generated JavaScript and C++17.
+
 ## 0.6.8
 
 - Matlab sparse matrices now preserve zero extents throughout the supported static, finite-real rank-two feature set.
@@ -13,8 +30,6 @@
 - Target LIR v34 now carries an explicit runtime shape-call ABI, allowing renderers to serialize validated calls without recovering semantic storage or shape policy.
 - Source maps preserve every new sparse operation and zero-dimensional left- or right-division call site.
 - Added dual-target differential execution, generated-plan rejection, cross-layer corruption, fuzz, and architecture coverage for zero-extent sparse behavior.
-- The validation baseline now contains 268 C++ tests, 98 differential cases, 127 CTest entries, and 18 generated-runtime rejection tests.
-- Production source coverage is 91.69% (37,154 of 40,520 lines), and the existing sparse-solve performance budget passes without being widened.
 
 ## 0.6.7
 
@@ -29,10 +44,8 @@
 - Source maps preserve the original `.*` expression location for every sparse operand arrangement.
 - Complex, zero-extent, dynamically shaped, and incompatible sparse element-wise operations continue to fail closed with diagnostics instead of changing semantics.
 - Added an executable Matlab example that compares values, shape, and sparse storage across both output targets.
-- The validation baseline now contains 258 C++ tests, 97 differential cases, 124 CTest entries, and 16 generated-runtime rejection tests.
 - The Matlab fuzz corpus now exercises sparse element-wise syntax, operand directions, broadcasting, and runtime validation boundaries.
 - A dedicated sparse element-wise benchmark constrains compilation latency, throughput, peak arena memory, and generated-code size.
-- Production source coverage is 91.59% (36,732 of 40,103 lines), above the 85% required threshold.
 
 ## 0.6.6
 
@@ -110,7 +123,7 @@
 - Cross-layer verifiers reject corrupted sparse-index identity, arity, shape, type, storage, or inactive-state facts before emission.
 - `full` now accepts rank-two dense or CSC selections with dynamic or zero result extents when their storage representation is known.
 - Sparse assignment, more than two selectors, complex or sparse selectors, N-dimensional linear results, and dynamic, empty, or complex sparse sources continue to fail closed with stable diagnostics.
-- Added an executable sparse-indexing example with dual-target behavior, source-map preservation, out-of-bounds rejection, and fuzz regression coverage; production line coverage is 91.24% (33,780/37,023).
+- Added an executable sparse-indexing example with dual-target behavior, source-map preservation, out-of-bounds rejection, and fuzz regression coverage.
 
 ## 0.6.1
 

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.20)
 
 project(
   mpf
-  VERSION 0.6.8
+  VERSION 0.6.9
   DESCRIPTION "Modern, high-performance multilingual translation framework."
   LANGUAGES C CXX)
 

README-ZH.md

@@ -24,7 +24,7 @@
 
 | 输入语言 | 自动识别扩展名 | 当前能力摘要 |
 |---|---|---|
-| Matlab | `.m` | 脚本与局部函数、条件/循环/标量 `switch`、N 维逐元素 `~`/`&`/`|`、标量短路 `&&`/`||`、数组 condition truthiness、保持 shape 的 `all`/`any` 逻辑归约、实数与复数标量/数组、保持 shape 的 `0×0` 与零 extent 数组、静态 N 维及 local-function runtime-shape 隐式扩展、数组比较、静态及运行时 extent 的 `end`、保序/重复/空 numeric selector、线性/逐维 logical selector、vector/matrix/N 维扩容与单轴删除、普通/共轭转置、section、`reshape`、复数逐元素算术、二维实数/复数矩阵乘法、结构感知实数方阵求解、Hermitian 正定/稠密复数方阵求解、rank-aware 实数/复数矩形求解、R2024 全部 `sparse` 调用形式在静态 real rank-2 CSC contract（包含零 extent）内的实现、稀疏 scalar/linear/submatrix selection、indexed assignment、扩容、删除、reshape、转置/查询/计数、含 `0×0` 系数及 shaped-empty 结果的稀疏方阵求解，以及保持 Matlab 结果 storage 的 finite-real sparse×sparse/sparse×dense/dense×sparse 矩阵乘法、双向 sparse/scalar 缩放和 compatible-size sparse `.*` scalar/dense/sparse 乘法、实数/复数方阵 safe-integer 整数幂与多输出函数 |
+| Matlab | `.m` | 脚本与局部函数、条件/循环/标量 `switch`、N 维逐元素 `~`/`&`/`|`、标量短路 `&&`/`||`、数组 condition truthiness、保持 shape 的 `all`/`any` 逻辑归约、实数与复数标量/数组、保持 shape 的 `0×0` 与零 extent 数组、静态 N 维及 local-function runtime-shape 隐式扩展、数组比较、静态及运行时 extent 的 `end`、保序/重复/空 numeric selector、线性/逐维 logical selector、vector/matrix/N 维扩容与单轴删除、普通/共轭转置、section、`reshape`、复数逐元素算术、二维实数/复数矩阵乘法、结构感知实数方阵求解、Hermitian 正定/稠密复数方阵求解、rank-aware 实数/复数矩形求解、R2024 全部 `sparse` 调用形式对静态 real/logical rank-2 输入的实现（全零构造仍返回 double）、logical triplet 重复坐标按 `any` 合并，以及保持类型的稀疏 scalar/linear/submatrix selection、indexed assignment、扩容、删除、reshape、转置/`full`/查询/计数、含 `0×0` 系数及 shaped-empty 结果的稀疏方阵求解，以及保持 Matlab 结果 storage 的 finite-real sparse×sparse/sparse×dense/dense×sparse 矩阵乘法、双向 sparse/scalar 缩放和 compatible-size sparse `.*` scalar/dense/sparse 乘法、实数/复数方阵 safe-integer 整数幂与多输出函数 |
 | Python | `.py`、`.pyw` | 函数与参数、条件和循环、list/tuple、解包、比较链、多维数组、索引和切片 |
 | Fortran | `.f`、`.for`、`.ftn`、`.f77`、`.f90` 等 | free/fixed form、function/subroutine、`INTENT`/`OPTIONAL`、数组与 section、`SELECT CASE` |
 | TypeScript | `.ts`、`.mts`、`.cts` | 类型化标量和数组、函数、块作用域、条件、`while`、标准 C 风格 `for` |
@@ -124,7 +124,7 @@ cmake --install build/release --prefix build/stage
 在项目中查找当前精确版本：
 
 ```cmake
-find_package(mpf 0.6.8 EXACT CONFIG REQUIRED COMPONENTS core cpp)
+find_package(mpf 0.6.9 EXACT CONFIG REQUIRED COMPONENTS core cpp)
 target_link_libraries(my_application PRIVATE mpf::mpf)
 ```
 
@@ -152,7 +152,7 @@ int main() {
 }
 ```
 
-安装包提供 `core`、`javascript` 和 `cpp` component，以及 `mpf::core`、`mpf::backend-javascript`、`mpf::backend-cpp` 和统一入口 `mpf::mpf`。完整集成示例见 [`examples/embedding`](examples/embedding)；配置时传入 `-DMPF_REQUIRED_VERSION=0.6.8`，以保持精确版本匹配。
+安装包提供 `core`、`javascript` 和 `cpp` component，以及 `mpf::core`、`mpf::backend-javascript`、`mpf::backend-cpp` 和统一入口 `mpf::mpf`。完整集成示例见 [`examples/embedding`](examples/embedding)；配置时传入 `-DMPF_REQUIRED_VERSION=0.6.9`，以保持精确版本匹配。
 
 MPF 0.x 有意只安装静态库。共享库需要先明确符号导出、allocator/所有权和版本协商契约；设置 `BUILD_SHARED_LIBS` 不会把当前内部 C++ ABI 意外暴露为受支持的动态库接口。
 

README.md

@@ -24,7 +24,7 @@ A modern, high-performance multilingual transpilation framework. MPF converts su
 
 | Input language | Recognized extensions | Current capabilities |
 |---|---|---|
-| Matlab | `.m` | Scripts and local functions, conditionals/loops/scalar `switch`, element-wise N-D `~`/`&`/`|`, scalar short-circuit `&&`/`||`, array condition truthiness, shape-aware `all`/`any` logical reductions, real and complex scalars/arrays, shape-preserving `0×0` and zero-extent arrays, static N-D and local-function runtime-shape implicit expansion, array comparisons, static and runtime-sized `end`, ordered/repeated/empty numeric and linear/per-dimension logical selectors, vector/matrix/N-D growth and single-axis deletion, ordinary/conjugate transpose, sections, `reshape`, complex element-wise arithmetic, two-dimensional real/complex matrix multiplication, structure-aware real square solves, Hermitian-positive-definite/dense complex square solves, rank-aware real/complex rectangular solves, all R2024 `sparse` call forms within the static real rank-2 CSC contract including zero extents, sparse scalar/linear/submatrix selection plus indexed assignment, growth, deletion, reshape, transpose/query/count, square solves including `0×0` coefficients and shape-preserving empty results, finite-real sparse×sparse/sparse×dense/dense×sparse matrix multiplication, bidirectional sparse/scalar scaling, and compatible-size sparse `.*` scalar/dense/sparse multiplication with Matlab-compatible canonical CSC results, safe-integer real/complex square matrix power, and multiple-output functions |
+| Matlab | `.m` | Scripts and local functions, conditionals/loops/scalar `switch`, element-wise N-D `~`/`&`/`|`, scalar short-circuit `&&`/`||`, array condition truthiness, shape-aware `all`/`any` logical reductions, real and complex scalars/arrays, shape-preserving `0×0` and zero-extent arrays, static N-D and local-function runtime-shape implicit expansion, array comparisons, static and runtime-sized `end`, ordered/repeated/empty numeric and linear/per-dimension logical selectors, vector/matrix/N-D growth and single-axis deletion, ordinary/conjugate transpose, sections, `reshape`, complex element-wise arithmetic, two-dimensional real/complex matrix multiplication, structure-aware real square solves, Hermitian-positive-definite/dense complex square solves, rank-aware real/complex rectangular solves, all R2024 `sparse` call forms for statically shaped real or logical rank-2 inputs (zero-matrix construction remains double), logical duplicate triplets using `any`, type-preserving sparse scalar/linear/submatrix selection plus indexed assignment, growth, deletion, reshape, transpose/`full`/query/count, square solves including `0×0` coefficients and shape-preserving empty results, finite-real sparse×sparse/sparse×dense/dense×sparse matrix multiplication, bidirectional sparse/scalar scaling, and compatible-size sparse `.*` scalar/dense/sparse multiplication with Matlab-compatible canonical CSC results, safe-integer real/complex square matrix power, and multiple-output functions |
 | Python | `.py`, `.pyw` | Functions and parameters, conditionals and loops, lists/tuples, unpacking, comparison chains, multidimensional arrays, indexing, and slicing |
 | Fortran | `.f`, `.for`, `.ftn`, `.f77`, `.f90`, and others | Free/fixed form, functions/subroutines, `INTENT`/`OPTIONAL`, arrays and sections, and `SELECT CASE` |
 | TypeScript | `.ts`, `.mts`, `.cts` | Typed scalars and arrays, functions, block scope, conditionals, `while`, and standard C-style `for` loops |
@@ -124,7 +124,7 @@ cmake --install build/release --prefix build/stage
 Find the exact current version in another project:
 
 ```cmake
-find_package(mpf 0.6.8 EXACT CONFIG REQUIRED COMPONENTS core cpp)
+find_package(mpf 0.6.9 EXACT CONFIG REQUIRED COMPONENTS core cpp)
 target_link_libraries(my_application PRIVATE mpf::mpf)
 ```
 
@@ -152,7 +152,7 @@ int main() {
 }
 ```
 
-The installed package provides the `core`, `javascript`, and `cpp` components; the `mpf::core`, `mpf::backend-javascript`, and `mpf::backend-cpp` targets; and the unified `mpf::mpf` entry point. See [`examples/embedding`](examples/embedding) for a complete integration example; configure it with `-DMPF_REQUIRED_VERSION=0.6.8` so the consumer keeps exact-version matching.
+The installed package provides the `core`, `javascript`, and `cpp` components; the `mpf::core`, `mpf::backend-javascript`, and `mpf::backend-cpp` targets; and the unified `mpf::mpf` entry point. See [`examples/embedding`](examples/embedding) for a complete integration example; configure it with `-DMPF_REQUIRED_VERSION=0.6.9` so the consumer keeps exact-version matching.
 
 MPF 0.x installs static libraries deliberately. A supported shared-library ABI will require an explicit symbol-export, allocator/ownership, and version-negotiation contract; setting `BUILD_SHARED_LIBS` does not silently expose the current internal C++ ABI.