[python-package] `build-python.sh --precompile --no-isolation` still uses build isolation, and logs some warnings about license data

[Manamama]

Summary:

pip install -v lightgbm kinda works, but then the installer compiles the package from scratch via ninja and thus causes the out of memory errors, even on a Droid with about 16 GB virtual (swap) plus real RAM, only in Termux. (It works fine in the Prooted OSes, out of the box, as nothing gets compiled there). I tried adding .toml switches to make the compiler use some -j1 like args, but I failed. The below method only works, some of the below was written by ChatGPT:

Description

One needs to install LightGBM's Python package from a precompiled lib_lightgbm.so build using:

bash build-python.sh --precompile --no-isolation install

The script still triggers:

• a temporary isolated virtual environment
• execution of python -m build --sdist
• multiple warnings about deprecated project.license syntax

The behavior is harmless, but unexpected. The invocation also indirectly uses setuptools’ isolated environment to generate the sdist even though no compilation happens. The script succeeds — but a warning-free, venv-free flow would be ideal when all build artifacts are already present.

---

Reproducible example

# Prepare and compile manually
cmake -S. -Bbuild-lightgbm -DCMAKE_BUILD_TYPE=Release \
  -D__BUILD_FOR_PYTHON=ON -D__INTEGRATE_OPENCL=OFF -DUSE_OPENMP=OFF -G"Unix Makefiles"
make -C build-lightgbm -j1

# Then:
bash build-python.sh --precompile --no-isolation install

---

Environment info

LightGBM version or commit hash:
master, commit near LightGBM v4.6.0.99

Command(s) you used to install LightGBM

bash build-python.sh --precompile --no-isolation install

Platform:
Termux on Android 11, using Clang and Make

---

My tweaks to CmakeList, for good measure:

~/downloads/LightGBM $ head CMakeLists.txt -n 30
option(USE_MPI "Enable MPI-based distributed learning" OFF)
option(USE_OPENMP "Enable OpenMP" ON)
option(USE_GPU "Enable GPU-accelerated training" OFF)
option(USE_SWIG "Enable SWIG to generate Java API" OFF)
option(USE_TIMETAG "Set to ON to output time costs" ON)
option(USE_CUDA "Enable CUDA-accelerated training " OFF)
option(USE_DEBUG "Set to ON for Debug mode" OFF)
option(USE_SANITIZER "Use sanitizer flags" OFF)
set(
  ENABLED_SANITIZERS
  "address" "leak" "undefined"
  CACHE
  STRING
  "Semicolon separated list of sanitizer names, e.g., 'address;leak'. \
Supported sanitizers are address, leak, undefined and thread."
)
option(USE_HOMEBREW_FALLBACK "(macOS-only) also look in 'brew --prefix' for libraries (e.g. OpenMP)" ON)
option(BUILD_CLI "Build the 'lightgbm' command-line interface in addition to lib_lightgbm" ON)
option(BUILD_CPP_TEST "Build C++ tests with Google Test" OFF)
option(BUILD_STATIC_LIB "Build static library" OFF)
option(INSTALL_HEADERS "Install headers to CMAKE_INSTALL_PREFIX (e.g. '/usr/local/include')" ON)
option(__BUILD_FOR_PYTHON "Set to ON if building lib_lightgbm for use with the Python-package" ON)
option(__BUILD_FOR_R "Set to ON if building lib_lightgbm for use with the R-package" OFF)
option(__INTEGRATE_OPENCL "Set to ON if building LightGBM with the OpenCL ICD Loader and its dependencies included" OFF)

cmake_minimum_required(VERSION 3.28)

# If using Visual Studio generators, always target v10.x of the Windows SDK.
# Doing this avoids lookups that could fall back to very old versions, e.g. by finding
# outdated registry entries.
~/downloads/LightGBM $

Result:

~/downloads/LightGBM $ pip show lightgbm
Name: lightgbm
Version: 4.6.0.99
Summary: LightGBM Python-package
Home-page: https://github.com/microsoft/LightGBM
Author:
Author-email:
License: The MIT License (MIT)

under:

~/downloads/LightGBM $ uname -a
Linux localhost 4.14.186+ #1 SMP PREEMPT Thu Mar 17 16:28:22 CST 2022 aarch64 Android
~/downloads/LightGBM $

Ver. 1.3, typos removed and meaning hopefully cleared up.

[jameslamb]

Thanks for using LightGBM.

Below only works, some written by ChatGPT

I found this report very difficult to understand.

For example

• the "My tweaks to CmakeList," section doesn't appear to show any changes to the project's code
• I don't understand what you mean by "adding some .toml switches"
• does "below only works" mean "here are some commands that worked"? If so, what exactly are you asking us to change?
• the title "Termux OS build trick" does not at all describe what this report is about

You are responsible for every word attached to your name when you post here... if you use something like ChatGPT, please edit it for clarity before posting.

Please also review https://docs.github.com/en/get-started/writing-on-github/getting-started-with-writing-and-formatting-on-github/basic-writing-and-formatting-syntax, for some help with editing markdown. I slightly edited your post for formatting (for example, the entire first paragraph here was formatted in large text like an h1 heading until I edited it).

The script still triggers: ...

This was difficult to understand because you omitted some details (for example, version of build and pip? the exact warning messages about licenses? evidence that build isolation was being ignored?).

So I tried to reproduce it by running the following code, from the latest commit of LightGBM (e7c6c43).

cmake -B build -S . -DUSE_OPENMP=OFF
cmake --build build -j1

# pre-install build dependencies, to avoid build isolation
python -m pip install \
  build \
  scikit-build-core

# build and install 'lightgbm' Python package
sh build-python.sh install --precompile --no-isolation

That worked successfully for me (on Ubuntu 24.04 with gcc, but nothing in your report points to a problem that's specific to Android).

But you're right, there are some warnings about project.license and build isolation is being used despite --no-isolation being passed... both things we should fix.

[INFO] building lightgbm
Requirement already satisfied: build>=0.10.0 in /home/jlamb/miniforge3/lib/python3.12/site-packages (1.2.2.post1)
Requirement already satisfied: packaging>=19.1 in /home/jlamb/miniforge3/lib/python3.12/site-packages (from build>=0.10.0) (24.2)
Requirement already satisfied: pyproject_hooks in /home/jlamb/miniforge3/lib/python3.12/site-packages (from build>=0.10.0) (1.2.0)
[INFO] found pre-compiled lib_lightgbm.so
[INFO] --- building sdist ---
* Creating isolated environment: venv+pip...
* Installing packages in isolated environment:
  - setuptools
* Getting build dependencies for sdist...
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/config/_apply_pyprojecttoml.py:82: SetuptoolsDeprecationWarning: `project.license` as a TOML table is deprecated
!!

        ********************************************************************************
        Please use a simple string containing a SPDX expression for `project.license`. You can also use `project.license-files`. (Both options available on setuptools>=77.0.0).

        By 2026-Feb-18, you need to update your project and remove deprecated calls
        or your builds will no longer be supported.

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  corresp(dist, value, root_dir)
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/config/_apply_pyprojecttoml.py:61: SetuptoolsDeprecationWarning: License classifiers are deprecated.
!!

        ********************************************************************************
        Please consider removing the following classifiers in favor of a SPDX license expression:

        License :: OSI Approved :: MIT License

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  dist._finalize_license_expression()
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/dist.py:759: SetuptoolsDeprecationWarning: License classifiers are deprecated.
!!

        ********************************************************************************
        Please consider removing the following classifiers in favor of a SPDX license expression:

        License :: OSI Approved :: MIT License

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  self._finalize_license_expression()
running egg_info
creating lightgbm.egg-info
writing lightgbm.egg-info/PKG-INFO
writing dependency_links to lightgbm.egg-info/dependency_links.txt
writing requirements to lightgbm.egg-info/requires.txt
writing top-level names to lightgbm.egg-info/top_level.txt
writing manifest file 'lightgbm.egg-info/SOURCES.txt'
reading manifest file 'lightgbm.egg-info/SOURCES.txt'
reading manifest template 'MANIFEST.in'
warning: no files found matching '*.dll' under directory 'lightgbm'
warning: no files found matching '*.dylib' under directory 'lightgbm'
adding license file 'LICENSE'
writing manifest file 'lightgbm.egg-info/SOURCES.txt'
* Building sdist...
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/config/_apply_pyprojecttoml.py:82: SetuptoolsDeprecationWarning: `project.license` as a TOML table is deprecated
!!

        ********************************************************************************
        Please use a simple string containing a SPDX expression for `project.license`. You can also use `project.license-files`. (Both options available on setuptools>=77.0.0).

        By 2026-Feb-18, you need to update your project and remove deprecated calls
        or your builds will no longer be supported.

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  corresp(dist, value, root_dir)
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/config/_apply_pyprojecttoml.py:61: SetuptoolsDeprecationWarning: License classifiers are deprecated.
!!

        ********************************************************************************
        Please consider removing the following classifiers in favor of a SPDX license expression:

        License :: OSI Approved :: MIT License

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  dist._finalize_license_expression()
/tmp/build-env-jk1pjbz7/lib/python3.12/site-packages/setuptools/dist.py:759: SetuptoolsDeprecationWarning: License classifiers are deprecated.
!!

        ********************************************************************************
        Please consider removing the following classifiers in favor of a SPDX license expression:

        License :: OSI Approved :: MIT License

        See https://packaging.python.org/en/latest/guides/writing-pyproject-toml/#license for details.
        ********************************************************************************

!!
  self._finalize_license_expression()
running sdist
running egg_info
writing lightgbm.egg-info/PKG-INFO
writing dependency_links to lightgbm.egg-info/dependency_links.txt
writing requirements to lightgbm.egg-info/requires.txt
writing top-level names to lightgbm.egg-info/top_level.txt
reading manifest file 'lightgbm.egg-info/SOURCES.txt'
reading manifest template 'MANIFEST.in'
warning: no files found matching '*.dll' under directory 'lightgbm'
warning: no files found matching '*.dylib' under directory 'lightgbm'
adding license file 'LICENSE'
writing manifest file 'lightgbm.egg-info/SOURCES.txt'
running check
creating lightgbm-4.6.0.99
creating lightgbm-4.6.0.99/lightgbm
creating lightgbm-4.6.0.99/lightgbm.egg-info
creating lightgbm-4.6.0.99/lightgbm/lib
copying files to lightgbm-4.6.0.99...
copying LICENSE -> lightgbm-4.6.0.99
copying MANIFEST.in -> lightgbm-4.6.0.99
copying README.rst -> lightgbm-4.6.0.99
copying pyproject.toml -> lightgbm-4.6.0.99
copying lightgbm/__init__.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/basic.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/callback.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/compat.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/dask.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/engine.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/libpath.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/plotting.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/py.typed -> lightgbm-4.6.0.99/lightgbm
copying lightgbm/sklearn.py -> lightgbm-4.6.0.99/lightgbm
copying lightgbm.egg-info/PKG-INFO -> lightgbm-4.6.0.99/lightgbm.egg-info
copying lightgbm.egg-info/SOURCES.txt -> lightgbm-4.6.0.99/lightgbm.egg-info
copying lightgbm.egg-info/dependency_links.txt -> lightgbm-4.6.0.99/lightgbm.egg-info
copying lightgbm.egg-info/requires.txt -> lightgbm-4.6.0.99/lightgbm.egg-info
copying lightgbm.egg-info/top_level.txt -> lightgbm-4.6.0.99/lightgbm.egg-info
copying lightgbm/lib/lib_lightgbm.so -> lightgbm-4.6.0.99/lightgbm/lib
copying lightgbm.egg-info/SOURCES.txt -> lightgbm-4.6.0.99/lightgbm.egg-info
Writing lightgbm-4.6.0.99/setup.cfg
Creating tar archive
removing 'lightgbm-4.6.0.99' (and everything under it)
Successfully built lightgbm-4.6.0.99.tar.gz
[INFO] --- installing lightgbm ---
Looking in links: .
Processing ./lightgbm-4.6.0.99.tar.gz
  Preparing metadata (pyproject.toml) ... done
Building wheels for collected packages: lightgbm
  Building wheel for lightgbm (pyproject.toml) ... done
  Created wheel for lightgbm: filename=lightgbm-4.6.0.99-py3-none-any.whl size=3527602 sha256=032403961f04de7a3c032fc21651c952c0d128fb1d414e434e57d7e41f4a9956
  Stored in directory: /tmp/pip-ephem-wheel-cache-rxb22hr5/wheels/8f/bf/e1/b02b4567f05be1181880c443cdd3f42464160add5fe9730cd6
Successfully built lightgbm
Installing collected packages: lightgbm
Successfully installed lightgbm-4.6.0.99
[INFO] cleaning up

CMake configure logs (click me)

-- The C compiler identification is GNU 13.3.0
-- The CXX compiler identification is GNU 13.3.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Performing Test MM_PREFETCH
-- Performing Test MM_PREFETCH - Success
-- Using _mm_prefetch
-- Performing Test MM_MALLOC
-- Performing Test MM_MALLOC - Success
-- Using _mm_malloc
-- Configuring done (0.4s)
-- Generating done (0.0s)
-- Build files have been written to: /home/jlamb/repos/LightGBM/build

CMake build logs (click me)

[  2%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/boosting.cpp.o
[  5%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_model_text.cpp.o
[  7%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_prediction.cpp.o
[ 10%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt.cpp.o
[ 12%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/prediction_early_stop.cpp.o
[ 15%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/sample_strategy.cpp.o
[ 17%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/bin.cpp.o
[ 20%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config_auto.cpp.o
[ 23%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config.cpp.o
[ 25%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset_loader.cpp.o
[ 28%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset.cpp.o
[ 30%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/file_io.cpp.o
[ 33%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/json11.cpp.o
[ 35%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/metadata.cpp.o
[ 38%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/parser.cpp.o
[ 41%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o
[ 43%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/tree.cpp.o
[ 46%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/dcg_calculator.cpp.o
[ 48%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/metric.cpp.o
[ 51%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linker_topo.cpp.o
[ 53%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_mpi.cpp.o
[ 56%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_socket.cpp.o
[ 58%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/network.cpp.o
[ 61%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/objective/objective_function.cpp.o
[ 64%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/data_parallel_tree_learner.cpp.o
[ 66%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/feature_histogram.cpp.o
[ 69%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/feature_parallel_tree_learner.cpp.o
[ 71%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/gpu_tree_learner.cpp.o
[ 74%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/gradient_discretizer.cpp.o
[ 76%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/linear_tree_learner.cpp.o
[ 79%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/serial_tree_learner.cpp.o
[ 82%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/tree_learner.cpp.o
[ 84%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/voting_parallel_tree_learner.cpp.o
[ 87%] Building CXX object CMakeFiles/lightgbm_objs.dir/src/utils/openmp_wrapper.cpp.o
[ 87%] Built target lightgbm_objs
[ 89%] Building CXX object CMakeFiles/lightgbm.dir/src/main.cpp.o
[ 92%] Building CXX object CMakeFiles/lightgbm.dir/src/application/application.cpp.o
[ 94%] Linking CXX executable /home/jlamb/repos/LightGBM/lightgbm
[ 94%] Built target lightgbm
[ 97%] Building CXX object CMakeFiles/lightgbm_capi_objs.dir/src/c_api.cpp.o
[ 97%] Built target lightgbm_capi_objs
[100%] Linking CXX shared library /home/jlamb/repos/LightGBM/lib_lightgbm.so
[100%] Built target _lightgbm

output of 'pip list' (click me)

Package                 Version
----------------------- -----------
archspec                0.2.5
boltons                 24.0.0
Brotli                  1.1.0
build                   1.2.2.post1
certifi                 2025.4.26
cffi                    1.17.1
cfgv                    3.3.1
charset-normalizer      3.4.1
colorama                0.4.6
conda                   24.11.3
conda-libmamba-solver   24.9.0
conda-package-handling  2.4.0
conda_package_streaming 0.11.0
distlib                 0.3.9
distro                  1.9.0
filelock                3.18.0
frozendict              2.4.6
h2                      4.2.0
hpack                   4.1.0
hyperframe              6.1.0
identify                2.6.10
idna                    3.10
jsonpatch               1.33
jsonpointer             3.0.0
libmambapy              1.5.12
lightgbm                4.6.0.99
mamba                   1.5.12
menuinst                2.2.0
nodeenv                 1.9.1
packaging               24.2
pathspec                0.12.1
pip                     25.0.1
platformdirs            4.3.6
pluggy                  1.5.0
pre_commit              4.2.0
pycosat                 0.6.6
pycparser               2.22
pyproject_hooks         1.2.0
PySocks                 1.7.1
PyYAML                  6.0.2
requests                2.32.3
ruamel.yaml             0.18.10
ruamel.yaml.clib        0.2.8
scikit_build_core       0.11.4
setuptools              75.8.2
tqdm                    4.67.1
truststore              0.10.1
ukkonen                 1.0.1
urllib3                 2.3.0
virtualenv              20.31.1
wheel                   0.45.1
zstandard               0.23.0

I will get those fixed. I've modified the title of this issue to describe those problems.

[Manamama]

My fault, @jameslamb : I solved it and wrote the solution with two fingers on Android, after about 2 hours' struggle, also on the mobile itself, even leaving some typos, now that I see. Here I am typing it on a regular keyboard; hopefully it shall be clearer now. (But then, I am not using any LLM to fix my train of thoughts herein, so it may be less clear, on the other hand... ).

In short: if on a Droid below one installs the package via the regular pip install lightgbm , then the installation crashes due to out of memory in a regular Termux, e.g. this one:

------ 
Operating System: Android 11 aarch64 
Kernel: 4.14.186+ 
Shell: /data/data/com.termux/files/usr/bin/bash 5.2.37 
Python: 3.12.11 
------ 
Display Manager 
Display Server:  
------ 
Central Processing Unit: MT6785V/CD (8) @ 2.0GHz 
CPU Usage: 7% 
Memory: 3914MiB / 5638MiB 
Disk (/system/priv-app/WapiCertManager/oat/arm64/WapiCertManager.vdex): 1.9G / 1.9G (100%) 
Uptime: 14 hours, 45 mins 
------ 
Host: realme RMX3085  
------ 

despite the sizeable swap file:

Filename				Type		Size		Used		Priority
/dev/block/zram0                        partition	3145724		2871064		-2
/data/swapfile                          file		6291452		0		-3

probably because ninja uses all the threads by default, as I wrote:

but [it] compiles from scratch via ninja and [thus] causes [the] out of memory errors

Yes, it works when doing cmake --build build -j1 manually etc. as you and as I wrote.

Re:

I found this report very difficult to understand.

vide supra.

Please also review https://docs.github.com/en/get-started/writing-on-github/getting-started-with-writing-and-formatting-on-github/basic-writing-and-formatting-syntax, for some help with editing markdown.

Hard to follow that with one or two fingers on a mobile GUI, when one is sharing how to solve a problem after two hours of trying to do the same, e.g. in the woods. (Yes, I could have waited some days to report it on a PC, but then I would either forget how I solved it or I would not care: an end user here, happy that it finally worked: not a coder.)

For example

the "My tweaks to CmakeList," section doesn't appear to show any changes to the project's code

I had switched some values to ON, as a test, but then I forgot which ones and why. Just this set of values "works here on my end", so I added this info FYI, for any Termux user who may still find the default set to not compile on their Droid for another reason.

I don't understand what you mean by "adding some .toml switches"

I vaguely remember now that I tried to add the compilation directive of -j 1 to the .toml itself. I may have also changed some module versions in the .toml file (not switches indeed) or I may have commented out the requirements, so that the installer uses the system installed ones, see below. I am not keen on uninstalling it and git cloning it again, to see which ones may have failed, so let me see which one may have been in the .toml file itself, as I am typing it now.

https://github.com/microsoft/LightGBM/blob/master/python-package/pyproject.toml
has:

pandas = [
    "pandas>=0.24.0"
]
plotting = [
    "graphviz",
    "matplotlib"
]
scikit-learn = [
    "scikit-learn>=0.24.2"
] 

so I probably had to do something with these to use Termux's:

~/downloads $ apt list | grep pandas

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

python-pandas/tur-packages 2.3.0 aarch64
~/downloads $ apt list | grep scikit

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

python-scikit-image/tur-packages,now 0.25.2 aarch64 [installed]
~/downloads $ 

does "below only works" mean "here are some commands that worked"?

"Below only works" was my mental shortcut for: "only when doing below I made it install on my Droid, Termux", because the default method: pip install lightgbm caused out of memory errors.

If so, what exactly are you asking us to change?

It is for you to decide: you may just ignore me completely. My tip: do test for the the OOM errors and if they do occur, do use some version of -j1 single thread, if indeed multithreading be the culprit. Me no coder, I have just been testing your package on Droid once it got installed and it works now.

the title "Termux OS build trick" does not at all describe what this report is about

Mea culpa again: I could have used ChatGPT to write also the title for me (remember, I wrote it with two fingers back then, on a mobile), but then many GitHubbers detest too nicely formatted reports, too (sic). One can change it to: "A trick how to make it build and install on Termux (aarch64 Android) without Out of Memory errors" indeed.

I've modified the title of this issue to describe those problems.

Thanks. I have fixed a typo or two there too.

You are responsible for every word attached to your name when you post here... if you use something like ChatGPT, please edit it for clarity before posting.

Oh, I had edited it, but indeed with a finger or two only. If I had not edited it, then there would be fewer typos in fact!

I will get those fixed.

Thank you too. In short- do test for OOMs if the end user installs it via a regular pip install which uses multithreading and crashes Termux.

Other than that - it works fine, I tested it with the examples and learnt a lot too, yesterday, that it works in seconds also on Droid.

Ver. 1.2

[jameslamb]

One can change it to: "A trick how to make it build and install on Termux (aarch64 Android) without Out of Memory errors" indeed.

I will leave the title with the way I've re-written it.

My understanding from https://github.com/termux/termux-app is that that's a terminal emulator... the choice of terminal emulator is probably not the cause of any of the issues you've reported here. It seems like there are 3 problems, with other causes:

1. out-of-memory when building from source (solution: you should reduce build parallelism, because you have such limited free memory)
2. warnings about project.lice (solution: I will fix this)
3. sh build-python.sh --no-isolation still creating an isolation virtual env (solution: I will fix this)

do test for OOMs if the end user installs it via a regular pip install which uses multithreading and crashes Termux

We don't really have a way in the source distributions of the Python package to implement something like "try to compile, if you hit an out-of-memory then try again with reduced parallelism"... as you reported, out-of-memory during the build could kill the build process. So I don't think it can/should be lightgbm doing that "test".

It's also quite rare that we see LightGBM being built from source on a system with so little free memory. In that case, building single-threaded as you did is a good choice.

You can do that with pip, using a release of lightgbm (so no need to git clone, invoke cmake directly, etc.), like this:

pip install \
  --no-binary 'lightgbm' \
  -C cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=1 \
  'lightgbm>=4.6.0'

That cmake.define.CMAKE_BUILD_PARALLEL_LEVEL approach is described in the docs of lightgbm's build backend, scikit-build-core (docs link)

it works fine, I tested it with the examples and learnt a lot too, yesterday, that it works in seconds also on Droid.

I'm happy to hear that! 🎉

LightGBM does not officially support mobile operating systems, but we do try to keep the code generally portable across a wide range of other operating systems, compilers, etc.

You might want to subscribe to notifications on #6592 ... it'll probably receive updates if we ever add better Android support in the future. For example, maybe if there is enough demand we'd publish Android Python wheels, so that pip install lightgbm on Android just downloads files instead of compiling LightGBM from source.

[Manamama]

Thank you! Would this be the switch I had been hunting for then all the time a week ago or so then? :
-C cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=1

With some trepidation, let me just test if it works on my live Droid:

We have:

Filename				Type		Size		Used		Priority
/dev/block/zram0                        partition	3145724		2340632		-2
/data/swapfile                          file		6291452		39544		-3

that is: Memory: 3827MiB / 5638MiB

and:

pip show pip
Name: pip
Version: 25.1.1
...
Location: /data/data/com.termux/files/usr/lib/python3.12/site-packages
Requires: 
Required-by: llm, pip-api, pipdeptree

so we do:

pip uninstall LightGBM
Found existing installation: lightgbm 4.6.0.99
Uninstalling lightgbm-4.6.0.99:

We try yours:
pip install -v -C cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=1 lightgbm

monitoring memory usage via gotop, as before (yes it works fine, if with sudo, as root), as we love fancy colourful graphs.

Results:

pip install -v -C cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=1 lightgbm 
Using pip 25.1.1 from /data/data/com.termux/files/usr/lib/python3.12/site-packages/pip (python 3.12)
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cpu
Collecting lightgbm
  Downloading lightgbm-4.6.0.tar.gz (1.7 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1.7/1.7 MB 61.4 kB/s eta 0:00:00
  Running command pip subprocess to install build dependencies
  Using pip 25.1.1 from /data/data/com.termux/files/usr/lib/python3.12/site-packages/pip (python 3.12)
  Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cpu, https://download.pytorch.org/whl/cpu
  Collecting scikit-build-core>=0.10.1
    Obtaining dependency information for scikit-build-core>=0.10.1 from https://files.pythonhosted.org/packages/7d/24/994c9e68be81508270b00e32e797add99b445c7bccfb859eeb1634bce5ca/scikit_build_core-0.11.4-py3-none-any.whl.metadata
    Downloading scikit_build_core-0.11.4-py3-none-any.whl.metadata (18 kB)
  Collecting packaging>=23.2 (from scikit-build-core>=0.10.1)
    Obtaining dependency information for packaging>=23.2 from https://files.pythonhosted.org/packages/20/12/38679034af332785aac8774540895e234f4d07f7545804097de4b666afd8/packaging-25.0-py3-none-any.whl.metadata
    Downloading packaging-25.0-py3-none-any.whl.metadata (3.3 kB)
  Collecting pathspec>=0.10.1 (from scikit-build-core>=0.10.1)
    Obtaining dependency information for pathspec>=0.10.1 from https://files.pythonhosted.org/packages/cc/20/ff623b09d963f88bfde16306a54e12ee5ea43e9b597108672ff3a408aad6/pathspec-0.12.1-py3-none-any.whl.metadata
    Downloading pathspec-0.12.1-py3-none-any.whl.metadata (21 kB)
  Downloading scikit_build_core-0.11.4-py3-none-any.whl (185 kB)
  Downloading packaging-25.0-py3-none-any.whl (66 kB)
  Downloading pathspec-0.12.1-py3-none-any.whl (31 kB)
  Installing collected packages: pathspec, packaging, scikit-build-core

aha, it seems that a single thread kicks in indeed this time, but is it so indeed?:

  Running command Building wheel for lightgbm (pyproject.toml)
  2025-06-23 11:16:50,066 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/cmake -E capabilities
  2025-06-23 11:16:50,095 - scikit_build_core - INFO - CMake version: 3.31.6
  *** scikit-build-core 0.11.4 using CMake 3.31.6 (wheel)
  2025-06-23 11:16:50,099 - scikit_build_core - INFO - Implementation: cpython linux on aarch64
  2025-06-23 11:16:50,120 - scikit_build_core - INFO - Build directory: /data/data/com.termux/files/usr/tmp/tmpteae4sem/build
  *** Configuring CMake...
  2025-06-23 11:16:50,169 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/ninja --version
  2025-06-23 11:16:50,198 - scikit_build_core - INFO - Ninja version: 1.13.0
  2025-06-23 11:16:50,208 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/cmake -S. -B/data/data/com.termux/files/usr/tmp/tmpteae4sem/build -DCMAKE_BUILD_TYPE:STRING=Release -C/data/data/com.termux/files/usr/tmp/tmpteae4sem/build/CMakeInit.txt -DCMAKE_INSTALL_PREFIX=/data/data/com.termux/files/usr/tmp/tmpteae4sem/wheel/platlib -DCMAKE_MAKE_PROGRAM=/data/data/com.termux/files/usr/bin/ninja -DCMAKE_BUILD_PARALLEL_LEVEL=1 -D__BUILD_FOR_PYTHON:BOOL=ON -DOPENMP_RUNTIME=COMP -DOPENBLAS_INCLUDE_DIR=/data/data/com.termux/files/usr/include/openblas
  loading initial cache file /data/data/com.termux/files/usr/tmp/tmpteae4sem/build/CMakeInit.txt
  -- The C compiler identification is Clang 20.1.7
  -- The CXX compiler identification is Clang 20.1.7
  -- Detecting C compiler ABI info
  -- Detecting C compiler ABI info - done
  -- Check for working C compiler: /data/data/com.termux/files/usr/bin/clang - skipped
  -- Detecting C compile features
  -- Detecting C compile features - done
  -- Detecting CXX compiler ABI info
  -- Detecting CXX compiler ABI info - done
  -- Check for working CXX compiler: /data/data/com.termux/files/usr/bin/clang++ - skipped
  -- Detecting CXX compile features
  -- Detecting CXX compile features - done
  -- Found OpenMP_C: -fopenmp=libomp (found version "5.1")
  -- Found OpenMP_CXX: -fopenmp=libomp (found version "5.1")
  -- Found OpenMP: TRUE (found version "5.1")
  -- Performing Test MM_PREFETCH
  -- Performing Test MM_PREFETCH - Failed
  -- Performing Test MM_MALLOC
  -- Performing Test MM_MALLOC - Success
  -- Using _mm_malloc
  -- Configuring done (9.4s)
  -- Generating done (0.1s)

it is building fine with a minimal spike of the RAM at first, but:

• almost all the threads are at 100 percent, so it compiles in parallel still, despite -DCMAKE_BUILD_PARALLEL_LEVEL=1 there
• deja vu: it crashes Termux at exactly the same OpenMP wrapper file:

  *** Building project with Ninja...
  2025-06-23 11:16:59,848 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/cmake --build /data/data/com.termux/files/usr/tmp/tmpteae4sem/build --target _lightgbm
  [1/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/prediction_early_stop.cpp.o
  [2/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_prediction.cpp.o
  [3/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/sample_strategy.cpp.o
  [4/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/file_io.cpp.o
  [5/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config.cpp.o
  [6/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/boosting.cpp.o
  [7/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_model_text.cpp.o
  [8/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config_auto.cpp.o
  [9/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/json11.cpp.o
  [10/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt.cpp.o
  [11/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o
  [12/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_mpi.cpp.o
  [13/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/parser.cpp.o
  [14/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linker_topo.cpp.o
  [15/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/dcg_calculator.cpp.o
  [16/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset_loader.cpp.o
  [17/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset.cpp.o
  [18/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/network.cpp.o
  [19/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_socket.cpp.o
  [20/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/gpu_tree_learner.cpp.o
  [21/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/bin.cpp.o
  [22/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/metric.cpp.o
  [23/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/metadata.cpp.o
  [24/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/objective/objective_function.cpp.o
  [25/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/tree.cpp.o
  [26/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/utils/openmp_wrapper.cpp.o
Connection to 192.168.44.1 closed by remote host.
Connection to 192.168.44.1 closed.

• tested twice.

So the "manual" method still is needed, unless you know of some recondite ninja args to forestall it.

Update: solved by using the same, but as an env variable instead: #6951 (comment)

Ver. 1.3

[Manamama]

For the record, it installs fine (as it should) on the same Droid in prooted Debian, as a ready wheel is used, I guess, so no compilation needed:

root@localhost:~# neofetch --off
root@localhost 
-------------- 
OS: Debian GNU/Linux 12 (bookworm) aarch64 
Host: realme RMX3085 
Kernel: 6.2.1-PRoot-Distro 
Uptime: 2 mins 
Packages: 1703 (dpkg) 
Shell: bash 5.2.15 
Terminal: proot 
CPU: MT6785V/CD (8) @ 2.000GHz 
Memory: 2802MiB / 5638MiB 

                         
                         


root@localhost:~# pip uninstall lightgbm
Found existing installation: lightgbm 4.6.0
Uninstalling lightgbm-4.6.0:
  Would remove:
    /usr/local/lib/python3.11/dist-packages/lightgbm-4.6.0.dist-info/*
    /usr/local/lib/python3.11/dist-packages/lightgbm/*
Proceed (Y/n)? y
  Successfully uninstalled lightgbm-4.6.0
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
root@localhost:~# pip install -v  lightgbm
Using pip 25.1.1 from /usr/local/lib/python3.11/dist-packages/pip (python 3.11)
Collecting lightgbm
  Obtaining dependency information for lightgbm from https://files.pythonhosted.org/packages/64/41/4fbde2c3d29e25ee7c41d87df2f2e5eda65b431ee154d4d462c31041846c/lightgbm-4.6.0-py3-none-manylinux2014_aarch64.whl.metadata
  Downloading lightgbm-4.6.0-py3-none-manylinux2014_aarch64.whl.metadata (17 kB)
Requirement already satisfied: numpy>=1.17.0 in /usr/local/lib/python3.11/dist-packages (from lightgbm) (2.2.6)
Requirement already satisfied: scipy in /usr/local/lib/python3.11/dist-packages (from lightgbm) (1.15.3)
Downloading lightgbm-4.6.0-py3-none-manylinux2014_aarch64.whl (3.5 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 3.5/3.5 MB 62.0 kB/s eta 0:00:00
Installing collected packages: lightgbm
Successfully installed lightgbm-4.6.0
WARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager, possibly rendering your system unusable. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv. Use the --root-user-action option if you know what you are doing and want to suppress this warning.
root@localhost:~# 

[Manamama]

Update: Grok AI, after much thinking and ferreting also in old GitHub tickets like this one, has finally found the elegant solution:

export CMAKE_BUILD_PARALLEL_LEVEL=1 indeed does enforce single threading during the build phase:

~/downloads $ ps aux | grep -E 'ninja|clang++'
u0_a278  17643  0.1  0.0 2126884 5204 pts/29   S<+   1970   0:00 /data/data/com.termux/files/usr/bin/ninja -j 1 _lightgbm
u0_a278  17999 78.3  2.2 2374764 131596 pts/29 R<    1970   0:02 /data/data/com.termux/files/usr/bin/clang++ -DEIGEN_DONT_PARALLELIZE -DEIGEN_MPL2_ONLY -DMM_MALLOC -DUSE_SOCKET -I/data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/external_libs/eigen -I/data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/external_libs/fast_double_parser/include -I/data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/external_libs/fmt/include -I/data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/include -Wno-error=implicit-function-declaration -fopenmp=libomp -pthread -Wextra -Wall -Wno-ignored-attributes -Wno-unknown-pragmas -Wno-return-type -O3 -fPIC -funroll-loops -O3 -DNDEBUG -std=gnu++11 -fopenmp=libomp -MD -MT CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o -MF CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o.d -o CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o -c /data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/src/io/train_share_states.cpp
u0_a278  18003  1.0  0.0 2125672 3296 pts/27   S<+   1970   0:00 grep -E ninja|clang++
~/downloads $ 

and the banal pip install -v lightgbm compiles fine now, also in Termux.

Long live Grok, I am copying its finds for what they be worth:

Key Findings from Documentation

1. CMake Documentation: CMAKE_BUILD_PARALLEL_LEVEL
   The official CMake documentation provides strong evidence that CMAKE_BUILD_PARALLEL_LEVEL is a universal method for controlling build parallelism across platforms and build tools, including Ninja.
   Source: CMake 4.0.3 Documentation - CMAKE_BUILD_PARALLEL_LEVEL

Quote:
If set, specifies the maximum number of concurrent processes to use when building. If not set or empty, the build tool’s default number of processes is used. For the Ninja generator, this is equivalent to the -j command-line option.

Details:
Introduced in CMake 3.12, CMAKE_BUILD_PARALLEL_LEVEL is an environment variable that controls the number of jobs for cmake --build.

It applies to all CMake generators, including Ninja, Make, and Visual Studio, making it cross-platform.

For Ninja, setting CMAKE_BUILD_PARALLEL_LEVEL=1 directly translates to ninja -j1, ensuring single-threaded execution.

The variable is checked during the build phase (cmake --build), not configuration, ensuring it affects the actual compilation process.

Indeed:

  -- Build files have been written to: /data/data/com.termux/files/usr/tmp/tmpxvbs9b7y/build
  *** Building project with Ninja...
  2025-06-23 11:47:21,166 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/cmake --build /data/data/com.termux/files/usr/tmp/tmpxvbs9b7y/build --target _lightgbm
  [1/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/boosting.cpp.o
  [2/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_model_text.cpp.o
  [3/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt_prediction.cpp.o
  [4/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/gbdt.cpp.o
  [5/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/prediction_early_stop.cpp.o
  [6/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/boosting/sample_strategy.cpp.o
  [7/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/bin.cpp.o
  [8/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config_auto.cpp.o
  [9/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/config.cpp.o
  [10/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset_loader.cpp.o
  [11/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/dataset.cpp.o
  [12/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/file_io.cpp.o
  [13/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/json11.cpp.o
  [14/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/metadata.cpp.o
  [15/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/parser.cpp.o
  [16/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/train_share_states.cpp.o
  [17/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/io/tree.cpp.o
  [18/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/dcg_calculator.cpp.o
  [19/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/metric/metric.cpp.o
  [20/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linker_topo.cpp.o
  [21/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_mpi.cpp.o
  [22/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/linkers_socket.cpp.o
  [23/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/network/network.cpp.o
  [24/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/objective/objective_function.cpp.o
  [25/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/data_parallel_tree_learner.cpp.o
  [26/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/feature_histogram.cpp.o
  [27/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/feature_parallel_tree_learner.cpp.o
  [28/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/gpu_tree_learner.cpp.o
  [29/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/gradient_discretizer.cpp.o
  [30/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/linear_tree_learner.cpp.o
  [31/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/serial_tree_learner.cpp.o
  [32/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/tree_learner.cpp.o
  [33/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/treelearner/voting_parallel_tree_learner.cpp.o
  [34/36] Building CXX object CMakeFiles/lightgbm_objs.dir/src/utils/openmp_wrapper.cpp.o
  [35/36] Building CXX object CMakeFiles/lightgbm_capi_objs.dir/src/c_api.cpp.o
  [36/36] Linking CXX shared library /data/data/com.termux/files/usr/tmp/pip-install-64fxmon6/lightgbm_bd97325f71be476eb9bf7e226d27edd1/lib_lightgbm.so
  *** Installing project into wheel...
  2025-06-23 12:08:15,775 - scikit_build_core - INFO - RUN: /data/data/com.termux/files/usr/bin/cmake --install /data/data/com.termux/files/usr/tmp/tmpxvbs9b7y/build --prefix /data/data/com.termux/files/usr/tmp/tmpxvbs9b7y/wheel/platlib
  -- Install configuration: "Release"
  -- Installing: /data/data/com.termux/files/usr/tmp/tmpxvbs9b7y/wheel/platlib/lightgbm/lib/lib_lightgbm.so
  *** Making wheel...
  2025-06-23 12:08:15,900 - scikit_build_core - INFO - Discovered Python package at lightgbm
  *** Created lightgbm-4.6.0-py3-none-linux_aarch64.whl
  Building wheel for lightgbm (pyproject.toml) ... done
  Created wheel for lightgbm: filename=lightgbm-4.6.0-py3-none-linux_aarch64.whl size=2244415 sha256=f7e211adc213897cafbe99c8c527529b9067d7090d2d5cc38fc5d004eedaff1d
  Stored in directory: /data/data/com.termux/files/home/.cache/pip/wheels/cb/83/4f/255634f94c01f16556323cbf2c4f92981396f889463f14acd6
Successfully built lightgbm
Installing collected packages: lightgbm
Successfully installed lightgbm-4.6.0

in Termux. Do notice the interesting changed sequence of these compiled objects, too.

-> I am about to add a version of this CMAKE_BUILD_PARALLEL_LEVEL to my bag of tricks then.

More Grok AI's musings about it all, for what they may be worth:

Explanation: Why the Sequence of Files Compiled Differs
The difference in the compilation order of files between Build A (parallel, without export CMAKE_BUILD_PARALLEL_LEVEL=1, crashed due to OOM) and Build B (single-threaded, with CMAKE_BUILD_PARALLEL_LEVEL=1) is due to how Ninja schedules tasks based on its dependency graph and parallelism settings. Below, I’ll use the Ninja documentation to explain this, focusing on the mechanisms that govern task execution order, and confirm that CMAKE_BUILD_PARALLEL_LEVEL=1 is a robust method to enforce single-threaded compilation, as evidenced by your successful build.

1. Ninja’s Dependency-Based Scheduling
   Source: Ninja Manual, Conceptual overview:
   "Ninja evaluates a graph of dependencies between files, and runs whichever commands are necessary to make your build target up to date as determined by file modification times. [...] Build statements describe the dependency graph of your project, while rule statements describe how to generate the files along a given edge of the graph."

Explanation:
Ninja uses a dependency graph generated by CMake (from LightGBM’s CMakeLists.txt) to determine which files (e.g., boosting.cpp, prediction_early_stop.cpp) need to be compiled into object files (.o). Each file is a node, and edges represent dependencies (e.g., a .cpp file depending on a header).

When multiple tasks are “ready” (their dependencies are satisfied), Ninja can execute them in any order, as the documentation states that build statements define the graph but don’t enforce a strict execution sequence for independent tasks.

In parallel mode (Build A, likely -j8 on your 8-core CPU), Ninja schedules multiple ready tasks concurrently, and the log order reflects when tasks start or complete, which depends on system resources (e.g., CPU scheduling, I/O). This leads to variability in the logged order.

In single-threaded mode (Build B, -j1 via CMAKE_BUILD_PARALLEL_LEVEL=1), Ninja executes one task at a time, following a sequential traversal of the graph. The choice of the next task is deterministic within a single run but may differ from parallel runs due to different scheduling constraints.
...

Build A started with prediction_early_stop.cpp, followed by gbdt_prediction.cpp, etc., reflecting the order in which parallel tasks were logged.

Build B started with boosting.cpp, followed by gbdt_model_text.cpp, etc., reflecting a sequential traversal starting from a different independent task. The documentation’s lack of a strict order for ready tasks explains this difference.

...
No Strict Order for Independent Tasks
Source: Ninja Manual, Build statements:
"Build statements declare a relationship between input and output files. [...] Such a declaration says that all of the output files are derived from the input files. When the output files are missing or when the inputs change, Ninja will run the rule to regenerate the outputs."

Explanation:
LightGBM’s source files (e.g., boosting.cpp, prediction_early_stop.cpp) are independent tasks if they don’t depend on each other’s outputs (e.g., no shared object files or headers requiring sequential compilation). Ninja can choose any ready task to execute next.

In parallel mode, Ninja picks multiple tasks, and the log order depends on runtime factors (e.g., which clang++ process finishes first). In single-threaded mode, Ninja picks one task at a time, potentially starting with a different file due to internal prioritization (e.g., alphabetical or graph node order).

Your logs show this: Build A’s order (prediction_early_stop.cpp, gbdt_prediction.cpp, ...) differs from Build B’s (boosting.cpp, gbdt_model_text.cpp, ...), reflecting different scheduling paths through the same graph.

Ver. 1.4