Python: `module.func()` calls through a module loaded with `importlib.util.spec_from_file_location` are no longer traced (0.10.8 → 0.11.0, still in 0.12.0)
维护者通常 2 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 56/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- python
调研方向
Start by tracing how Python call relationships are indexed and how trace_path resolves inbound callers; the issue does not name source files or tests. Compare the reproduced importlib.util.spec_from_file_location case across versions with the static import case. Done when the intended behavior is clarified and a regression test captures the result.
由索引模型根据 Issue 内容生成。
描述
Disclosure: this issue was written and posted by an AI agent (Claude Code) acting on behalf of @benjaminge73. The reproduction below was run by the agent on all three versions before posting; the outputs are copied from those runs.
Version
Tested with the official darwin-arm64 release archives (SHA-256 checked against checksums.txt):
| Version | Callers of compute_total found |
|---|---|
| 0.10.8 | 2 (uses_direct, uses_via_importlib) |
| 0.11.0 | 1 (uses_direct only) |
| 0.12.0 | 1 (uses_direct only) |
Platform: macOS (Apple Silicon). Install channel: GitHub release archive. Binary variant: standard.
What happened, and what did you expect?
In Python, a function that is called as module.func() where module was loaded with importlib.util.spec_from_file_location + module_from_spec + exec_module was traced as an inbound caller in 0.10.8. Since 0.11.0, trace_path (direction: inbound) no longer returns that caller. The statically imported caller is still found.
I expected the same result as 0.10.8 (2 callers). I do not know whether the drop is intentional (for example a precision change; the 0.12.0 notes mention that external imports are no longer bound to same-named project symbols), so I am reporting it as a behavior change in case it is not.
Why it matters: this is the standard way to test a script that is not a package (load_module() helper, then module.func(...) in each test). In one of my projects roughly 90 of 140 test files load their target this way, so trace_path ... include_tests lost most of the test callers of a symbol (13 callers in 0.10.8, 1 in 0.11.0 and 0.12.0). I pinned 0.10.8 because of it.
Reproduction
No proprietary code: the script below builds a 3-file dummy project in a temp dir, indexes it and traces the callers. Run it once per binary (HOME is a temp dir, nothing on your machine is touched):
#!/usr/bin/env bash
# usage: BIN=/path/to/codebase-memory-mcp ./repro.sh
set -euo pipefail
export HOME="$(mktemp -d)"
D="$(mktemp -d)"; cd "$D"; D="$(pwd -P)"
mkdir pkg && touch pkg/__init__.py
cat > pkg/target.py <<'PY'
def compute_total(values):
return sum(values)
PY
cat > pkg/direct.py <<'PY'
from pkg.target import compute_total
def uses_direct():
return compute_total([1, 2])
PY
cat > pkg/dynamic.py <<'PY'
import importlib.util
from pathlib import Path
def load():
spec = importlib.util.spec_from_file_location("target", Path(__file__).parent / "target.py")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
def uses_via_importlib():
module = load()
return module.compute_total([3, 4])
PY
git init -q . && git add -A && git -c user.email=a@b -c user.name=x commit -qm fixture
"$BIN" --version
"$BIN" cli index_repository "{\"repo_path\":\"$D\",\"mode\":\"full\"}" >/dev/null 2>&1
PROJECT="$(echo "$D" | sed 's|^/||; s|/|-|g')"
"$BIN" cli trace_path "{\"function_name\":\"compute_total\",\"project\":\"$PROJECT\",\"direction\":\"inbound\",\"depth\":1}" 2>/dev/null
Output with 0.10.8:
codebase-memory-mcp 0.10.8
function: compute_total
direction: inbound
callers_total: 2
...pkg.direct: uses_direct 1
...pkg.dynamic: uses_via_importlib 1
Output with 0.11.0 and 0.12.0:
callers_total: 1
callers_total_relation: eq
...pkg.direct.uses_direct 1
(The raw-JSON form of cli prints a deprecation warning in 0.12.0; it still works and I kept it so the same script runs on all three versions.)
Project scale
Tiny: 3 Python files. The same drop shows on a real project (about 10k nodes, 40k edges).
Confirmations
- I searched existing issues and this is not a duplicate. (Closest ones: #1237, #1271, #1153 — they are about static imports, polymorphism and instance methods, not about modules loaded with
importlib.) - My reproduction uses shareable code (a dummy snippet), not proprietary code.
- 主要语言
- C
- 星标
- 45.8k
- 派生
- 3.8k
- 平均合并
- 3 天 6 小时
- 30 天内合并 PR
- 201
环境准备
这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
DeusData/codebase-memory-mcp 的其他 Issue
-
Generated Pi extension (cbmem.ts) breaks Gemini/Antigravity: Type.Union of Type.Literal serializes to anyOf/const可能已有人在做 @divakantpandey 于 1 天前认领。 未关闭bug editor/integration
难度 2/5 1-3 小时 新手友好度 72/100
DeusData/codebase-memory-mcp#2787 · 1 条评论 ·
维护者通常 2 天内回复
-
editor/integration parsing/quality
难度 2/5 1-3 小时 新手友好度 68/100
DeusData/codebase-memory-mcp#2286 ·
维护者通常 2 天内回复
-
install writes an unsupported `args` key into the Claude Code SubagentStart hook, so it never runs未关闭bug editor/integration priority/high stability/performance ux/behavior
难度 2/5 1-3 小时 新手友好度 72/100
DeusData/codebase-memory-mcp#2239 · 1 条评论 ·
维护者通常 2 天内回复
-
bug priority/high ux/behavior
难度 2/5 1-3 小时 新手友好度 72/100
DeusData/codebase-memory-mcp#2232 · 2 条评论 ·
维护者通常 2 天内回复
-
documentation editor/integration parsing/quality priority/normal
难度 2/5 1-3 小时 新手友好度 85/100
DeusData/codebase-memory-mcp#2169 ·
维护者通常 2 天内回复
查看 DeusData/codebase-memory-mcp 的全部 Issue
相似的 Issue
-
documentation
难度 1/5 1-3 小时 新手友好度 91/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 83/100
维护者通常 1 天内回复
-
enhancement ui
难度 1/5 1 小时以内 新手友好度 72/100
ChupathingyCE/chupathingyce#485 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 65/100
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 82/100
eclipse-paho/paho.mqtt.c#1727 ·
维护者通常 6 天内回复