[SWE-Paddle] Add task PaddlePaddle__Paddle-74594

swe-paddle/tasks/PaddlePaddle__Paddle-74594/README.md

@@ -0,0 +1,44 @@
+# PaddlePaddle__Paddle-74594
+
+This directory converts Paddle PR #74594 into a SWE-Paddle community task candidate.
+
+## Source
+
+| Field | Value |
+| --- | --- |
+| Repo | `PaddlePaddle/Paddle` |
+| PR | [74594](https://github.com/PaddlePaddle/Paddle/pull/74594) |
+| PR title | `[API compatibility] add broadcast_shapes api` |
+| Base commit | `20e50fb447d8e34bafb337c180ca28d77dfb82ca` |
+| Merged at | `2025-08-15` |
+| Task type | `new_feature` / `api_compatibility` |
+| Resource | CPU |
+
+## Summary
+
+Add the public `paddle.broadcast_shapes` API so callers can compute the common broadcasted shape of zero, one, or multiple input shapes while preserving the existing two-shape `broadcast_shape` behavior.
+
+## Why This Is A Good SWE-Paddle Candidate
+
+- It comes from a merged Paddle API-compatibility PR and represents a real user-facing compatibility gap.
+- The expected behavior is directly observable through returned shapes and exceptions rather than implementation-specific source structure.
+- The production change is small and Python-only, with a clear separation between public exports and the multi-shape wrapper behavior.
+- Stable P2P and F2P tests can run on CPU without a Paddle native source build or GPU.
+
+## Files
+
+- `proposal.md`: candidate proposal for maintainer triage.
+- `instruction.md`: self-contained task statement.
+- `solution/code.patch`: production-only Gold patch.
+- `tests/test.patch`: independent regression tests.
+- `tests/test.sh`: minimal target test command.
+- `environment/README.md`: reproduction environment notes.
+- `README.md`: task overview and verification entrypoint.
+
+## Verification
+
+```bash
+bash tests/test.sh
+```
+
+Expected behavior: with only `tests/test.patch` applied to the Base commit, the existing `broadcast_shape` P2P test passes while the new multi-shape API tests fail. After applying `solution/code.patch`, all target tests pass.

swe-paddle/tasks/PaddlePaddle__Paddle-74594/environment/README.md

@@ -0,0 +1,27 @@
+# Environment Notes
+
+This candidate is part of the SWE-Paddle community task set.
+
+## Expected Environment
+
+- Repository: `PaddlePaddle/Paddle`
+- Base commit: `20e50fb447d8e34bafb337c180ca28d77dfb82ca`
+- Resource: CPU
+- GPU required: no
+- Build path: Paddle source checkout at the base commit. This Python-only task may be verified with an AST overlay and a controlled broadcasting double; a Paddle native source build is not required.
+
+## Run Order
+
+1. Check out `PaddlePaddle/Paddle` at the base commit.
+2. Apply `tests/test.patch`.
+3. Run `bash tests/test.sh`; the target multi-shape API behavior should fail before the change.
+4. Apply `solution/code.patch`.
+5. Run `bash tests/test.sh` again; all target tests should pass.
+
+## Minimal Test Command
+
+```bash
+bash tests/test.sh
+```
+
+The verifier is responsible for deriving stable F2P and P2P node IDs from repeated runs.

swe-paddle/tasks/PaddlePaddle__Paddle-74594/instruction.md

@@ -0,0 +1,27 @@
+# 新增 `paddle.broadcast_shapes` API
+
+## 详细描述
+
+`paddle.broadcast_shape` 只能计算两个 shape 的广播结果。如果要同时处理三个或更多 shape 时，调用方只能重复调用该接口。
+
+需要新增 `paddle.broadcast_shapes(*shapes)`，一次计算所有输入 shape 广播后的结果。每个 shape 可以使用 list 或 tuple 表示，返回值为 list。
+
+未传入 shape 时返回 `[]`；只传入一个 shape 时，返回该 shape 对应的 list。传入多个 shape 时，按照现有广播规则计算结果，无法广播时抛出 `ValueError`。
+
+## 验收说明
+
+- 可以通过 `paddle.broadcast_shapes` 和 `paddle.tensor.broadcast_shapes` 调用该接口
+- 支持传入任意数量的 list 或 tuple 形式的 shape
+- 多个可广播的 shape 返回正确结果
+- 未传入 shape 时返回 `[]`
+- 只传入一个 shape 时返回对应的 list
+- 空 shape 以及包含 `0` 或 `-1` 的 shape 按照现有广播规则处理
+- 输入无法广播时抛出 `ValueError`
+- 返回值为 list
+- 现有 `paddle.broadcast_shape` 的行为保持不变
+
+## 技术要求
+
+- 熟悉 Python
+- 了解 Tensor 的广播规则
+- 了解 Paddle Python API 的定义和导出方式

swe-paddle/tasks/PaddlePaddle__Paddle-74594/proposal.md

@@ -0,0 +1,55 @@
+# Task Proposal: PaddlePaddle__Paddle-74594
+
+## 1. 来源信息
+
+* Instance ID：`PaddlePaddle__Paddle-74594`
+* PR 链接：https://github.com/PaddlePaddle/Paddle/pull/74594
+* PR 标题：`[API compatibility] add broadcast_shapes api`
+* `base_commit`：`20e50fb447d8e34bafb337c180ca28d77dfb82ca`
+* merged 时间：`2025-08-15`
+* 你的身份：熟悉该模块的 contributor
+* 后续联系人：TBD
+
+## 2. 问题一句话
+
+新增 `paddle.broadcast_shapes` 公共 API，使调用方能够一次计算任意数量 shape 的共同广播结果，并在输入无法广播时得到明确的异常。
+
+## 3. 为什么适合作为 SWE-Paddle 样本
+
+* **真实性**：该任务来自已合入的 Paddle PR #74594，不是人工构造的需求。
+* **代表性**：该任务涉及 API compatibility、公共 API 导出以及多个 shape 的广播规则。
+* **边界清楚**：production change 集中在 `python/paddle/__init__.py`、`python/paddle/tensor/__init__.py` 和 `python/paddle/tensor/math.py`，测试代码可以与 Gold production patch 清晰分离。
+* **非平凡性**：任务需要正确处理多个 shape、无输入、单输入和空 shape，并在输入不兼容时保持明确的异常行为，同时保证现有二元 `broadcast_shape` 不受影响。
+
+## 4. 任务类型和标签
+
+* 任务类型：`feature_enhancement`
+* 执行后端：`cpu`
+* 设备范围：`cpu_only`
+* 模块标签：`[api_compatibility, tensor, math, broadcasting, shape]`
+
+## 5. 验证思路
+
+* 目标测试命令：`bash tests/test.sh`
+* 目标测试文件：`test/swe_paddle/test_pr74594_broadcast_shapes.py`
+* 修复前预期：已有 `broadcast_shape` delegation P2P 应 pass；`broadcast_shapes` 的多输入、零输入和单输入测试应因 API 尚不存在而 fail。
+* 修复后预期：应用 production Gold patch 后，P2P 与全部 F2P 均应 pass；多个可广播 shape 应得到正确结果，不兼容输入应继续抛出 `ValueError`。
+* P2P 候选：已有 `broadcast_shape` 继续按原有方式接收两个 shape，并返回底层广播结果。
+* F2P 候选：三个及以上 shape 的共同广播；零输入、单输入和 empty shape；不兼容输入的异常行为。
+
+## 6. 环境与资源
+
+* 资源需求：CPU
+* Paddle 来源：`PaddlePaddle/Paddle` source checkout at `base_commit`
+* 是否能提供 Docker：暂无
+* patch 类型：Python-only
+* 环境建议：有Paddle 环境，使用能够运行 `pytest` 的 Python 环境
+* 最小测试命令：`bash tests/test.sh`
+* 是否有 oracle 日志：由 SWE-Paddle verifier 结果另行维护
+
+## 7. 风险自查
+
+* 泄露风险：`instruction.md` 只描述公共 API 的输入、输出、异常和兼容性要求，不透露 Gold patch 的循环结构、内部函数调用方式或具体实现步骤。
+* 环境风险：测试不依赖历史 Paddle wheel、GPU、CUDA、native extension 或网络环境。
+* flaky 风险：测试使用固定 shape 和确定性的异常断言，不涉及随机数、并发或异步执行。
+* 拆分风险：公共命名空间导出和多 shape 广播行为共同构成 `broadcast_shapes` 的完整能力，适合作为单个任务。

swe-paddle/tasks/PaddlePaddle__Paddle-74594/solution/code.patch

@@ -0,0 +1,98 @@
+diff --git a/python/paddle/__init__.py b/python/paddle/__init__.py
+index b1bdc05813d8c0dc810eeb5deb49696457f06e2c..6c220f760cf5d03911348cbe94b8f7185058901a 100644
+--- a/python/paddle/__init__.py
++++ b/python/paddle/__init__.py
+@@ -406,6 +406,7 @@ from .tensor.math import (  # noqa: F401
+     bitwise_right_shift,
+     bitwise_right_shift_,
+     broadcast_shape,
++    broadcast_shapes,
+     cartesian_prod,
+     ceil,
+     clip,
+@@ -1024,6 +1025,7 @@ __all__ = [
+     'DataParallel',
+     'argmin',
+     'prod',
++    'broadcast_shapes',
+     'broadcast_shape',
+     'conj',
+     'neg',
+diff --git a/python/paddle/tensor/__init__.py b/python/paddle/tensor/__init__.py
+index 824d8d681f4e594193da345e9c0bb34d8e3dd09e..6c6a77df9be0465a5c8c4dbca38ea223c1e8f0d6 100644
+--- a/python/paddle/tensor/__init__.py
++++ b/python/paddle/tensor/__init__.py
+@@ -269,6 +269,7 @@ from .math import (  # noqa: F401
+     bitwise_right_shift,
+     bitwise_right_shift_,
+     broadcast_shape,
++    broadcast_shapes,
+     cartesian_prod,
+     ceil,
+     ceil_,
+@@ -638,6 +639,7 @@ tensor_method_func = [
+     'isneginf',
+     'isposinf',
+     'isreal',
++    'broadcast_shapes',
+     'broadcast_shape',
+     'conj',
+     'neg',
+diff --git a/python/paddle/tensor/math.py b/python/paddle/tensor/math.py
+index 72a73ab931ab0024e5e672c7713039f25d7ebc60..ad845d5c1fb4220b2fe968086940d9be593e53ca 100644
+--- a/python/paddle/tensor/math.py
++++ b/python/paddle/tensor/math.py
+@@ -5448,6 +5448,53 @@ def any(
+         return out
+
+
++def broadcast_shapes(*shapes: Sequence[int]) -> list[int]:
++    """
++    The function returns the shape of doing operation with broadcasting on tensors of shape list.
++
++    Note:
++        If you want know more about broadcasting, please refer to `Introduction to Tensor`_ .
++
++        .. _Introduction to Tensor: ../../guides/beginner/tensor_en.html#chapter5-broadcasting-of-tensor
++
++    Args:
++        *shapes (list[int]|tuple[int]): A shape list of multiple tensors.
++
++
++    Returns:
++        list[int], the result shape.
++
++    Examples:
++        .. code-block:: python
++
++            >>> import paddle
++
++            >>> shape = paddle.broadcast_shapes([2, 1, 3], [1, 3, 1])
++            >>> shape
++            [2, 3, 3]
++
++            >>> # shape = paddle.broadcast_shapes([2, 1, 3], [3, 3, 1])
++            >>> # ValueError (terminated with error message).
++
++            >>> shape = paddle.broadcast_shapes([5, 1, 3], [1, 4, 1], [1, 1, 3])
++            >>> shape
++            [5, 4, 3]
++
++            >>> # shape = paddle.broadcast_shapes([5, 1, 3], [1, 4, 1], [1, 2, 3])
++            >>> # ValueError (terminated with error message).
++
++    """
++    if len(shapes) == 0:
++        return []
++    elif len(shapes) == 1:
++        return list(shapes[0])
++    else:
++        current_shape = list(shapes[0])
++        for next_shape in shapes[1:]:
++            current_shape = broadcast_shape(current_shape, next_shape)
++        return current_shape
++
++
+ def broadcast_shape(
+     x_shape: Sequence[int], y_shape: Sequence[int]
+ ) -> list[int]:

swe-paddle/tasks/PaddlePaddle__Paddle-74594/tests/test.patch

@@ -0,0 +1,107 @@
+diff --git a/test/swe_paddle/test_pr74594_broadcast_shapes.py b/test/swe_paddle/test_pr74594_broadcast_shapes.py
+new file mode 100644
+index 0000000..c2d8800
+--- /dev/null
++++ b/test/swe_paddle/test_pr74594_broadcast_shapes.py
+@@ -0,0 +1,101 @@
++from __future__ import annotations
++
++import ast
++import copy
++from pathlib import Path
++from types import SimpleNamespace
++from typing import Sequence
++
++import pytest
++
++
++TARGET = Path("python/paddle/tensor/math.py")
++
++
++def _extract_function(name, namespace):
++    tree = ast.parse(TARGET.read_text(encoding="utf-8"), filename=str(TARGET))
++    node = next(
++        (
++            copy.deepcopy(item)
++            for item in tree.body
++            if isinstance(item, (ast.FunctionDef, ast.AsyncFunctionDef))
++            and item.name == name
++        ),
++        None,
++    )
++    assert node is not None, f"paddle.tensor.math must provide {name}"
++    module = ast.Module(body=[node], type_ignores=[])
++    ast.fix_missing_locations(module)
++    scope = {"Sequence": Sequence, **namespace}
++    exec(compile(module, str(TARGET), "exec"), scope)
++    return scope[name]
++
++
++def _broadcast_two(x_shape, y_shape):
++    left = list(x_shape)
++    right = list(y_shape)
++    width = max(len(left), len(right))
++    left = [1] * (width - len(left)) + left
++    right = [1] * (width - len(right)) + right
++    result = []
++    for x_dim, y_dim in zip(left, right):
++        if x_dim == y_dim:
++            result.append(x_dim)
++        elif x_dim == 1:
++            result.append(y_dim)
++        elif y_dim == 1:
++            result.append(x_dim)
++        else:
++            raise ValueError(
++                f"shapes {x_shape!r} and {y_shape!r} are not broadcastable"
++            )
++    return result
++
++
++def _load_broadcast_shapes():
++    core = SimpleNamespace(broadcast_shape=_broadcast_two)
++    broadcast_shape = _extract_function("broadcast_shape", {"core": core})
++    return _extract_function(
++        "broadcast_shapes", {"broadcast_shape": broadcast_shape}
++    )
++
++
++def test_existing_broadcast_shape_delegation_remains_compatible():
++    calls = []
++    marker = [2, 3, 3]
++
++    def broadcast_shape(x_shape, y_shape):
++        calls.append((x_shape, y_shape))
++        return marker
++
++    function = _extract_function(
++        "broadcast_shape",
++        {"core": SimpleNamespace(broadcast_shape=broadcast_shape)},
++    )
++    x_shape = [2, 1, 3]
++    y_shape = (1, 3, 1)
++
++    result = function(x_shape, y_shape)
++
++    assert result is marker
++    assert calls == [(x_shape, y_shape)]
++
++
++def test_broadcast_shapes_combines_multiple_shapes_and_rejects_mismatch():
++    function = _load_broadcast_shapes()
++
++    assert function([5, 1, 3], [1, 4, 1], [1, 1, 3]) == [5, 4, 3]
++    assert function([2, 3, 1], [6], [3, 1]) == [2, 3, 6]
++    assert function([0, 3], [3], []) == [0, 3]
++
++    with pytest.raises(ValueError):
++        function([5, 1, 3], [1, 4, 1], [1, 2, 3])
++
++
++def test_broadcast_shapes_supports_empty_and_single_shape_inputs():
++    function = _load_broadcast_shapes()
++
++    assert function() == []
++    assert function((8, 3)) == [8, 3]
++    assert function([]) == []
++    assert function([], [2, 3, 4]) == [2, 3, 4]

swe-paddle/tasks/PaddlePaddle__Paddle-74594/tests/test.sh

@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+python -m pytest test/swe_paddle/test_pr74594_broadcast_shapes.py -q