Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

conda libraries no longer found at runtime (when meson uses system compiler for the python extension)

已关闭
#912 30 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 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
52/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
linux, python

调研方向

Start with meson-python 0.22.0's RPATH handling and reproduce the two conda workflows using the minimal mamba commands in the issue. Compare the generated shapely/lib.cpython-313-x86_64-linux-gnu.so files and their ldd output. Done means a source-built Shapely extension imports successfully without requiring compilers in the conda environment and resolves libgeos_c.so.1 at runtime.

由索引模型根据 Issue 内容生成。

描述

I am noticing this behaviour in our CI since a few days, and given the changelog notes about RPATH handling, I assume this might be a regression (or intentional change in behaviour) with the latest 0.22.0 release.

I also see there is an issue open about expectations in general for various cases (https://github.com/mesonbuild/meson-python/issues/903), but since I am not familiar enough with this to judge if this belongs there, opening a separate issue (cc @rgommers)

Trying to reproduce locally with a more minimal example, it seems to be triggered by having compilers installed in the conda env or not. The specific case is installing shapely from source on Linux (ubuntu), where shapely uses meson-python for its build system now and depends on one "system" library (GEOS). What I observe:

  • If the conda env does not contain compilers (but just libgcc which gets included through python/cython, I think), meson picks up the system gcc, finds the GEOS dependency just fine during build-time, but does not find it at runtime.
  • If the conda env does include compilers, meson picks up the conda env gcc, and then shapely finds the conda GEOS lib at runtime just fine.

Checking the resulting .so files for the python extension modules, I do see that in the first case the RUNPATH is not set, while it is in the second case (to the conda env's /lib/ directory).

When I force an older version of meson-python for the first bullet point case, it also works fine.

Now, if one is installing a package from source that requires a compiler in a conda env, it might certainly be good practice to use compilers from conda as well, and maybe this only worked "by accident" before. But since this worked before, and it gives confusing results now (since building works just fine, and with conda you don't consider this as a "custom install" of the geos dependency), reporting here.


On the geopandas CI where we are installing shapely from source, we did include cython in the conda env (maybe with the assumption that this would also install compilers? I don't remember, we have been doing that for many years), but not actually compilers. This can of course easily be fixed, but installing cython also (accidentally) avoided running into https://github.com/mesonbuild/meson/issues/15740

The minimal CI workflow that reproduces the issue:

$ mamba create -n test-meson-cy python=3.13 geos pkg-config cython
$ mamba activate test-meson-cy
(test-meson-cy) $ python -m pip install git+https://github.com/shapely/shapely.git@main -v
...
  Successfully installed Cython-3.3.0 meson-1.12.1 meson-python-0.22.0 numpy-2.5.3 packaging-26.3 pyproject-metadata-0.12.1
...
  Version: 1.12.1
  Source dir: /tmp/pip-req-build-b9h4dfq7
  Build dir: /tmp/pip-req-build-b9h4dfq7/.mesonpy-f634h_cm
  Build type: native build
  Program python found: YES
  Project name: shapely
  Project version: 2.2.0rc1+1.ge044b13
  C compiler for the host machine: cc (gcc 11.4.0 "cc (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0")
  C linker for the host machine: cc ld.bfd 2.38
  Cython compiler for the host machine: cython (cython 3.3.0)
  Host machine cpu family: x86_64
  Host machine cpu: x86_64
  Compiler for C supports arguments -Wno-cast-function-type: YES
  Compiler for C supports arguments -Wno-unused-parameter: YES
  Program python found: YES (/home/joris/conda/envs/test-meson-cy/bin/python)
  Run-time dependency geos found: YES 3.14.1
  numpy-config found: YES (/tmp/pip-build-env-qy0p3i6y/overlay/bin/numpy-config) 2.5.3
  Run-time dependency numpy found: YES 2.5.3
  Found pkg-config: YES (/home/joris/conda/envs/test-meson-cy/bin/pkg-config) 0.29.2
  Run-time dependency python found: YES 3.13
  Build targets in project: 4

  shapely 2.2.0rc1+1.ge044b13
...
Successfully installed numpy-2.5.3 shapely-2.2.0rc1+1.ge044b13

(test-meson-cy) $ python -c "import shapely; print(shapely.__version__)"
...
  File "/home/joris/conda/envs/test-meson-cy/lib/python3.13/site-packages/shapely/__init__.py", line 20, in <module>
    from shapely.lib import GEOSException
ImportError: libgeos_c.so.1: cannot open shared object file: No such file or directory

(test-meson-cy) $ ldd /home/joris/conda/envs/test-meson-cy/lib/python3.13/site-packages/shapely/lib.cpython-313-x86_64-linux-gnu.so 
	linux-vdso.so.1 (0x00007ffd2b8d4000)
	libgeos_c.so.1 => not found
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fdb93e00000)
	/lib64/ld-linux-x86-64.so.2 (0x00007fdb940e2000)

The above installs cython in the env (which does not actually install a compiler in the env?), and based on the logs above, meson actually picks up the system (Ubuntu) C compiler. However, what I initially tried first locally to reproduce was to include compilers in the env. Doing that work fine, and the equivalent logs:

$ mamba create -n test-meson python=3.13 geos pkg-config compilers
$ mamba activate test-meson
(test-meson) $ python -m pip install git+https://github.com/shapely/shapely.git@main -v
...
  Successfully installed Cython-3.3.0 meson-1.12.1 meson-python-0.22.0 numpy-2.5.3 packaging-26.3 pyproject-metadata-0.12.1
 ...
  The Meson build system
  Version: 1.12.1
  Source dir: /tmp/pip-req-build-l88btzqu
  Build dir: /tmp/pip-req-build-l88btzqu/.mesonpy-2n5hyvjs
  Build type: native build
  Program python found: YES
  Project name: shapely
  Project version: 2.2.0rc1+1.ge044b13
  C compiler for the host machine: /home/joris/conda/envs/test-meson/bin/x86_64-conda-linux-gnu-cc (gcc 14.3.0 "x86_64-conda-linux-gnu-cc (conda-forge gcc 14.3.0-19) 14.3.0")
  C linker for the host machine: /home/joris/conda/envs/test-meson/bin/x86_64-conda-linux-gnu-cc ld.bfd 2.45.1
  Cython compiler for the host machine: cython (cython 3.3.0)
  Host machine cpu family: x86_64
  Host machine cpu: x86_64
  Compiler for C supports arguments -Wno-cast-function-type: YES
  Compiler for C supports arguments -Wno-unused-parameter: YES
  Program python found: YES (/home/joris/conda/envs/test-meson/bin/python)
  Did not find CMake 'cmake'
  Found CMake: NO
  Run-time dependency geos found: NO  (tried cmake)
  Run-time dependency geos found: YES 3.14.1
  numpy-config found: YES (/tmp/pip-build-env-l5rdrf5r/overlay/bin/numpy-config) 2.5.3
  Run-time dependency numpy found: YES 2.5.3
  Found pkg-config: YES (/home/joris/conda/envs/test-meson/bin/pkg-config) 0.29.2
  Run-time dependency python found: YES 3.13
  Build targets in project: 4

  shapely 2.2.0rc1+1.ge044b13
...
Successfully installed numpy-2.5.3 shapely-2.2.0rc1+1.ge044b13

(test-meson) $ python -c "import shapely; print(shapely.__version__)"
2.2.0rc1+1.ge044b13

(test-meson) $ ldd /home/joris/conda/envs/test-meson/lib/python3.13/site-packages/shapely/lib.cpython-313-x86_64-linux-gnu.so 
	linux-vdso.so.1 (0x00007fff52550000)
	libgeos_c.so.1 => /home/joris/conda/envs/test-meson/lib/libgeos_c.so.1 (0x00007fb1b344a000)
	libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fb1b3200000)
	libgeos.so.3.14.1 => /home/joris/conda/envs/test-meson/lib/./libgeos.so.3.14.1 (0x00007fb1b2e00000)
	libstdc++.so.6 => /home/joris/conda/envs/test-meson/lib/./libstdc++.so.6 (0x00007fb1b2a00000)
	libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fb1b2d19000)
	libgcc_s.so.1 => /home/joris/conda/envs/test-meson/lib/./libgcc_s.so.1 (0x00007fb1b31d3000)
	/lib64/ld-linux-x86-64.so.2 (0x00007fb1b34de000)
	libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fb1b3430000)
主要语言
Python
星标
181
派生
94
平均合并
18 小时 29 分钟
30 天内合并 PR
14

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

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

mesonbuild/meson-python 的其他 Issue

查看 mesonbuild/meson-python 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。