Nuitka
1.0.0
This document is the recommended first read when you start usingNuitka. On this page, you will learn more about Nuitkafundamentals, such as license type, use cases, requirements, and credits.
Table of Contents
Requirements
Usage
Tutorial Setup and build on Windows
Use Cases
Tweaks
Typical Problems
Tips
Compilation Report
Performance
Unsupported functionality
Nuitka is the Python compiler. It is written in Python. It is a seamless replacement or extension to the Python interpreter and compilesevery construct that Python 2 (2.6, 2.7) and Python 3 (3.4 - 3.13) have, when itself run with that Python version.
It then executes uncompiled code and compiled code together in an extremely compatible manner.
You can use all Python library modules and all extension modules freely.
Nuitka translates the Python modules into a C level program that then
uses libpython and static C files of its own to execute in the same
way as CPython does.
All optimization is aimed at avoiding overhead, where it's unnecessary. None is aimed at removing compatibility, although slight improvements will occasionally be done, where not every bug of standard Python is emulated, e.g. more complete error messages are given, but there is a full compatibility mode to disable even that.
To ensure smooth operation of Nuitka, make sure to follow system requirements, that include the following components:
C Compiler
Python
Operating System
Architecture
You need a C compiler with support for C11 or alternatively a C++ compiler for C++03 [1].
Currently, this means, you need to use one of these compilers:
The MinGW64 C11 compiler, on Windows, must be based on gcc 11.2 or higher. It will be automatically downloaded if no usable C compiler is found, which is the recommended way of installing it, as Nuitka will also upgrade it for you.
Visual Studio 2022 or higher on Windows [2]. English language pack for best results (Nuitka filters away garbage outputs, but only for English language). It will be used by default if installed.
On all other platforms, the gcc compiler of at least version 5.1,
and below that the g++ compiler of at least version 4.4 as an
alternative.
The clang compiler on macOS X and most FreeBSD architectures.
On Windows, the clang-cl compiler on Windows can be used if
provided by the Visual Studio installer.
| [1] | Support for this C11 is given with gcc 5.x or higher or any clang version. The older MSVC compilers don't do it yet. But as a workaround, with Python 3.10 or older, the C++03 language standard is significantly overlapping with C11, it is then used instead. |
| [2] | Download for free from https://www.visualstudio.com/en-us/downloads/download-visual-studio-vs.aspx (the community editions work just fine). The latest version is recommended, but not required. On the other hand, there is no need to except to support pre-Windows 10 versions, and they might work for you, but support of these configurations is only available to commercial users. |
Python 2 (2.6, 2.7) and Python 3 (3.4 — 3.13) are supported. If at any moment, there is a stable Python release that is not in this list, rest assured it is being worked on and will be added.
Important
For Python 3.4 and only that version, we need other Python version as a compile time dependency.
Nuitka itself is fully compatible with all listed versions, but Scons as an internally used tool is not.
For these versions, you need a Python2 or Python 3.5 or higher installed as well, but only during the compile time. That is for use with Scons (which orchestrates the C compilation), which does not support the same Python versions as Nuitka.
In addition, on Windows, Python2 cannot be used because clcachedoes not work with it, there a Python 3.5 or higher needs to be
installed.
Nuitka finds these needed Python versions (e.g. on Windows via registry) and you shouldn't notice it as long as they are installed.
Increasingly, other functionality is available when another Python
has a certain package installed. For example, onefile compression
will work for a Python 2.x when another Python is found that has thezstandard package installed.
Moving binaries to other machines
The created binaries can be made executable independent of the Python
installation, with --standalone and --onefile options.
Binary filename suffix
The created binaries have an .exe suffix on Windows. On other
platforms they have no suffix for standalone mode, or .binsuffix, that you are free to remove or change, or specify with the-o option.
The suffix for acceleration mode is added just to be sure that the original script name and the binary name do not ever collide, so we can safely overwrite the binary without destroying the original source file.
It has to be CPython, Anaconda Python, or Homebrew
You need the standard Python implementation, called "CPython", to execute Nuitka because it is closely tied to implementation details of it.
It cannot be from the Windows app store
It is known that Windows app store Python definitely does not work, it's checked against.
It cannot be pyenv on macOS
It is known that macOS "pyenv" does not work. Use Homebrew instead for self compiled Python installations. But note that standalone mode will be worse on these platforms and not be as backward compatible with older macOS versions.
Supported Operating Systems: Linux, FreeBSD, NetBSD, macOS, and Windows (32 bits/64 bits/ARM).
Others will work as well. The portability is expected to be generally good, but the e.g. Nuitka's internal Scons usage may have to be adapted or need flags passed. Make sure to match Python and C compiler architecture, or else you will get cryptic error messages.
Supported Architectures are x86, x86_64 (amd64), and arm, likely many, many more.
Other architectures are expected to also work, out of the box, as Nuitka is generally not using any hardware specifics. These are just the ones tested and known to be good. Feedback is welcome. Generally, the architectures that Debian supports can be considered good and tested, too.
The recommended way of executing Nuitka is <the_right_python> -m
nuitka to be absolutely certain which Python interpreter you are
using, so it is easier to match with what Nuitka has.
The next best way of executing Nuitka bare that is from a source
checkout or archive, with no environment variable changes, most
noteworthy, you do not have to mess with PYTHONPATH at all for
Nuitka. You just execute the nuitka and nuitka-run scripts
directly without any changes to the environment. You may want to add thebin directory to your PATH for your convenience, but that step
is optional.
Moreover, if you want to execute with the right interpreter, in that
case, be sure to execute <the_right_python> bin/nuitka and be good.
Pick the right Interpreter
If you encounter a SyntaxError you absolutely most certainly have
picked the wrong interpreter for the program you are compiling.
Nuitka has a --help option to output what it can do:
nuitka --help
The nuitka-run command is the same as nuitka, but with a
different default. It tries to compile and directly execute a Python
script:
nuitka-run --help
This option that is different is --run, and passing on arguments
after the first non-option to the created binary, so it is somewhat more
similar to what plain python will do.
For most systems, there will be packages on the download page of Nuitka. But you can also
install it from source code as described above, but also like any other
Python program it can be installed via the normal python setup.py
install routine.
Notice for integration with GitHub workflows there is this Nuitka-Action that you should use that makes it really easy to integrate. You ought to start with a local compilation though, but this will be easiest for cross platform compilation with Nuitka.
Nuitka is licensed under the Apache License, Version 2.0; you may not use it except in compliance with the License.
You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
This is basic steps if you have nothing installed, of course if you have any of the parts, just skip it.
Download and install Python from https://www.python.org/downloads/windows
Select one of Windows x86-64 web-based installer (64 bits Python,
recommended) or x86 executable (32 bits Python) installer.
Verify it's working using command python --version.
python -m pip install nuitka
Verify using command python -m nuitka --version
mkdir HelloWorld
make a python file named hello.py
def talk(message):return "Talk " + messagedef main():print(talk("Hello World"))if __name__ == "__main__":main()Do as you normally would. Running Nuitka on code that works incorrectly is not easier to debug.
python hello.py
python -m nuitka hello.py
Note
This will prompt you to download a C caching tool (to speed up
repeated compilation of generated C code) and a MinGW64 based C
compiler, unless you have a suitable MSVC installed. Say yes to
both those questions.
Execute the hello.exe created near hello.py.
To distribute, build with --standalone option, which will not output
a single executable, but a whole folder. Copy the resultinghello.dist folder to the other machine and run it.
You may also try --onefile which does create a single file, but make
sure that the mere standalone is working, before turning to it, as it
will make the debugging only harder, e.g. in case of missing data files.
If you want to compile a whole program recursively, and not only the single file that is the main program, do it like this:
python -m nuitka --follow-imports program.py
Note
There are more fine-grained controls than --follow-importsavailable. Consider the output of nuitka --help. Including fewer
modules into the compilation, but instead using normal Python for it,
will make it faster to compile.
In case you have a source directory with dynamically loaded files, i.e.
one which cannot be found by recursing after normal import statements
via the PYTHONPATH (which would be the recommended way), you can
always require that a given directory shall also be included in the
executable:
python -m nuitka --follow-imports --include-plugin-directory=plugin_dir program.py
Note
If you don't do any dynamic imports, simply setting yourPYTHONPATH at compilation time is what you should do.
Use --include-plugin-directory only if you make __import__()calls that Nuitka cannot predict, and that come from a directory, for
everything from your Python installation, use --include-module or--include-package.
Note
The resulting filename will be program.exe on Windows,program.bin on other platforms, but --output-filename allows
changing that.
Note
The resulting binary still depends on CPython and used C extension modules being installed.
If you want to be able to copy it to another machine, use--standalone and copy the created program.dist directory and
execute the program.exe (Windows) or program (other
platforms) put inside.
If you want to compile a single extension module, all you have to do is this:
python -m nuitka --module some_module.py
The resulting file some_module.so can then be used instead ofsome_module.py.
Important
The filename of the produced extension module must not be changed as
Python insists on a module name derived function as an entry point,
in this case PyInit_some_module and renaming the file will not
change that. Match the filename of the source code to what the binary
name should be.
Note
If both the extension module and the source code of it are in the same directory, the extension module is loaded. Changes to the source code only have effect once you recompile.
Note
The option --follow-import-to works as well, but the included
modules will only become importable after you imported thesome_module name. If these kinds of imports are invisible to
Nuitka, e.g. dynamically created, you can use --include-module or--include-package in that case, but for static imports it should
not be needed.
Note
An extension module can never include other extension modules. You will have to create a wheel for this to be doable.
Note
The resulting extension module can only be loaded into a CPython of the same version and doesn't include other extension modules.
If you need to compile a whole package and embed all modules, that is also feasible, use Nuitka like this:
python -m nuitka --module some_package --include-package=some_package
Note
The inclusion of the package contents needs to be provided manually;
otherwise, the package is mostly empty. You can be more specific if
you like, and only include part of it, or exclude part of it, e.g.
with --nofollow-import-to='*.tests' you would not include the
unused test part of your code.
Note
Data files located inside the package will not be embedded by this process, you need to copy them yourself with this approach. Alternatively, you can use the file embedding of Nuitka commercial.
For distribution to other systems, there is the standalone mode, which
produces a folder for which you can specify --standalone.
python -m nuitka --standalone program.py
Following all imports is default in this mode. You can selectively
exclude modules by specifically saying --nofollow-import-to, but
then an ImportError will be raised when import of it is attempted at
program run time. This may cause different behavior, but it may also
improve your compile time if done wisely.
For data files to be included, use the option--include-data-files=<source>=<target> where the source is a file
system path, but the target has to be specified relative. For the
standalone mode, you can also copy them manually, but this can do extra
checks, and for the onefile mode, there is no manual copying possible.
To copy some or all file in a directory, use the option--include-data-files=/etc/*.txt=etc/ where you get to specify shell
patterns for the files, and a subdirectory where to put them, indicated
by the trailing slash.
Important
Nuitka does not consider data files code, do not include DLLs, or Python files as data files, and expect them to work, they will not, unless you really know what you are doing.
In the following, non-code data files are all files, not matching on of these criterions.
| Suffix | Rationale | Solution |
|---|---|---|
.py | Nuitka trims even the stdlib modules to be included. If it doesn't see Python code, there is no dependencies analyzed, and as a result it will just not work. | Use --include-module on them instead |
.pyc | Same as .py. | Use --include-module on them from their source code instead. |
.pyo | Same as .pyc. | Use --include-module on them from their source code instead. |
.pyw | Same as .py. | For including multiple programs, use multiple --main arguments instead. |
.pyi | These are ignored, because they are code-like and not needed at run time. For thelazy package that actually would depend on them, we made a compile time solution
that removes the need. | Raise an issue if 3rd part software needs it. |
.pyx | These are ignored, because they are Cython source code not used at run time | |
.dll | These are ignored, since they usually are not data files. For the cases where 3rd
party packages do actually used them as data, e.g. .NET packages, we solve that in
package configuration for it. | Create Nuitka Package configuration for those, with dll section for the package that uses them.
For rare cases, data-files section with special configuration might be the correct thing to do. |
.dylib | These are ignored, since they macOS extension modules or DLLs. | Need to add configuration with dll section or depends that are missing |
.so | These are ignored, since they Linux, BSD, etc. extension modules or DLLs. | Need to add configuration with dll section or depends that are missing |
.exe | The are binaries to Windows. | You can add Nuitka Package configuration to include those as DLLs and mark them as executable: yes |
.bin | The are binaries to non-Windows, otherwise same as .exe. |
Also folders are ignored, these are site-packages, dist-packagesand vendor-packages which would otherwise include a full virtualenv,
which is never a good thing to happen. And the __pycache__ folder is
also always ignored. On non-MacOS the file .DS_Store is ignored too,
and py.typed folders have only meaning to IDEs, and are ignored like.pyi files .
To copy a whole folder with all non-code files, you can use--include-data-dir=/path/to/images=images which will place those in
the destination, and if you want to use the --noinclude-data-filesoption to remove them. Code files are as detailed above DLLs,
executables, Python files, etc. and will be ignored. For those you can
use the --include-data-files=/binaries/*.exe=binary/ form to force
them, but that is not recommended and known to cause issues at run-time.
For package data, there is a better way, namely using--include-package-data, which detects all non-code data files of
packages automatically and copies them over. It even accepts patterns in
a shell style. It spares you the need to find the package directory
yourself and should be preferred whenever available. Functionally it's
very similar to --include-data-dir but it has the benefit to locate
the correct folder for you.
With data files, you are largely on your own. Nuitka keeps track of ones that are needed by popular packages, but it might be incomplete. Raise issues if you encounter something in these. Even better, raise PRs with enhancements to the Nuitka package configuration. We want 3rd party software to just work out of the box.
When that is working, you can use the onefile mode if you so desire.
python -m nuitka --onefile program.py
This will create a single binary, that extracts itself on the target, before running the program. But notice, that accessing files relative to your program is impacted, make sure to read the section Onefile: Finding files as well.
# Create a binary that unpacks into a temporary folderpython -m nuitka --onefile program.py
Note
There are more platform-specific options, e.g. related to icons,
splash screen, and version information, consider the --helpoutput for the details of these and check the section Tweaks.
For the unpacking, by default a unique user temporary path one is used,
and then deleted, however this default--onefile-tempdir-spec="{TEMP}/onefile_{PID}_{TIME}" can be
overridden with a path specification, then using a cached path, avoiding
repeated unpacking, e.g. with--onefile-tempdir-spec="{CACHE_DIR}/{COMPANY}/{PRODUCT}/{VERSION}"which uses version information, and user-specific cache directory.
Note
Using cached paths will be relevant, e.g. when Windows Firewall comes into play because otherwise, the binary will be a different one to it each time it is run.
Currently, these expanded tokens are available:
| Token | What this Expands to | Example |
|---|---|---|
| {TEMP} | User temporary file directory | C:Users...AppDataLocalsTemp |
| {PID} | Process ID | 2772 |
| {TIME} | Time in seconds since the epoch. | 1299852985 |
| {PROGRAM} | Full program run-time filename of executable. | C:SomeWhereYourOnefile.exe |
| {PROGRAM_BASE} | No-suffix of run-time filename of executable. | C:SomeWhereYourOnefile |
| {CACHE_DIR} | Cache directory for the user. | C:UsersSomeBodyAppDataLocal |
| {COMPANY} | Value given as --company-name | YourCompanyName |
| {PRODUCT} | Value given as --product-name | YourProductName |
| {VERSION} | Combination of --file-version & --product-version | 3.0.0.0-1.0.0.0 |
| {HOME} | Home directory for the user. | /home/somebody |
| {NONE} | When provided for file outputs, None is used | see notice below |
| {NULL} | When provided for file outputs, os.devnull is used | see notice below |
Important
It is your responsibility to make the path provided unique, on Windows a running program will be locked, and while using a fixed folder name is possible, it can cause locking issues in that case, where the program gets restarted.
Usually, you need to use {TIME} or at least {PID} to make a
path unique, and this is mainly intended for use cases, where e.g.
you want things to reside in a place you choose or abide your naming
conventions.
Important
For disabling output and stderr with --force-stdout-spec and--force-stderr-spec the values {NONE} and {NULL} achieve
it, but with different effect. With {NONE}, the corresponding
handle becomes None. As a result, e.g. sys.stdout will beNone, which is different from {NULL} where it will be backed
by a file pointing to os.devnull, i.e. you can write to it.
With {NONE}, you may e.g. get RuntimeError: lost sys.stdoutin case it does get used; with {NULL} that never happens.
However, some libraries handle this as input for their logging
mechanism, and on Windows this is how you are compatible withpythonw.exe which is behaving like {NONE}.
If you have a setup.py, setup.cfg or pyproject.toml driven
creation of wheels for your software in place, putting Nuitka to use is
extremely easy.
Let's start with the most common setuptools approach, you can,
having Nuitka installed of course, simply execute the targetbdist_nuitka rather than the bdist_wheel. It takes all the
options and allows you to specify some more, that are specific to
Nuitka.
# For setup.py if you don't use other build systems:setup( # Data files are to be handled by setuptools and not Nuitka
package_data={"some_package": ["some_file.txt"]},
..., # This is to pass Nuitka options.
command_options={ 'nuitka': { # boolean option, e.g. if you cared for C compilation commands '--show-scons': True, # options without value, e.g. enforce using Clang '--clang': None, # options with single values, e.g. enable a plugin of Nuitka '--enable-plugin': "pyside2", # options with several values, e.g. avoid including modules '--nofollow-import-to' : ["*.tests", "*.distutils"],
},
},
)# For setup.py with other build systems:# The tuple nature of the arguments is required by the dark nature of# "setuptools" and plugins to it, that insist on full compatibility,# e.g. "setuptools_rust"setup( # Data files are to be handled by setuptools and not Nuitka
package_data={"some_package": ["some_file.txt"]},
..., # This is to pass Nuitka options.
..., command_options={ 'nuitka': { # boolean option, e.g. if you cared for C compilation commands '--show-scons': ("setup.py", True), # options without value, e.g. enforce using Clang '--clang': ("setup.py", None), # options with single values, e.g. enable a plugin of Nuitka '--enable-plugin': ("setup.py", "pyside2"), # options with several values, e.g. avoid including modules '--nofollow-import-to' : ("setup.py", ["*.tests", "*.distutils"]),
}
},
)If for some reason, you cannot or do not want to change the target, you
can add this to your setup.py.
# For setup.pysetup( ..., build_with_nuitka=True)
Note
To temporarily disable the compilation, you could the remove above
line, or edit the value to False by or take its value from an
environment variable if you so choose, e.g.bool(os.getenv("USE_NUITKA", "True")). This is up to you.
Or you could put it in your setup.cfg
[metadata]build_with_nuitka = true
And last, but not least, Nuitka also supports the new build meta, so
when you have a pyproject.toml already, simple replace or add this
value:
[build-system]requires = ["setuptools>=42", "wheel", "nuitka", "toml"]build-backend = "nuitka.distutils.Build"# Data files are to be handled by setuptools and not Nuitka[tool.setuptools.package-data]some_package = ['data_file.txt'] [tool.nuitka]# These are not recommended, but they make it obvious to have effect.# boolean option, e.g. if you cared for C compilation commands, leading# dashes are omittedshow-scons = true# options with single values, e.g. enable a plugin of Nuitkaenable-plugin = "pyside2"# options with several values, e.g. avoid including modules, accepts# list argument.nofollow-import-to = ["*.tests", "*.distutils"]
Note
For the nuitka requirement above absolute paths likeC:Users...Nuitka will also work on Linux, use an absolute path
with two leading slashes, e.g. //home/.../Nuitka.
Note
Whatever approach you take, data files in these wheels are not handled by Nuitka at all, but by setuptools. You can, however, use the data file embedding of Nuitka commercial. In that case, you actually would embed the files inside the extension module itself, and not as a file in the wheel.
If you have multiple programs, that each should be executable, in the past you had to compile multiple times, and deploy all of these. With standalone mode, this, of course, meant that you were fairly wasteful, as sharing the folders could be done, but wasn't really supported by Nuitka.
Enter Multidist. There is an option --main that replaces or adds
to the positional argument given. And it can be given multiple times.
When given multiple times, Nuitka will create a binary that contains the
code of all the programs given, but sharing modules used in them. They
therefore do not have to be distributed multiple times.
Let's call the basename of the main path, and entry point. The names of
these must, of course, be different. Then the created binary can execute
either entry point, and will react to what sys.argv[0] appears to
it. So if executed in the right way (with something like subprocessor OS API you can control this name), or by renaming or copying the
binary, or symlinking to it, you can then achieve the miracle.
This allows to combine very different programs into one.
Note
This feature is still experimental. Use with care and report your findings should you encounter anything that is undesirable behavior
This mode works with standalone, onefile, and mere acceleration. It does not work with module mode.
For integration with GitHub workflows there is this Nuitka-Action that you should use that makes it really easy to integrate. You ought to start with a local compilation though, but this will be easiest for cross platform compilation with Nuitka.
This is an example workflow that builds on all 3 OSes
jobs:build: strategy: matrix: os: [macos-latest, ubuntu-latest, windows-latest]
runs-on: ${{ matrix.os }}
steps:
- name: Check-out repository uses: actions/checkout@v4 - name: Setup Python uses: actions/setup-python@v5 with: python-version: '3.10' cache: 'pip' cache-dependency-path: | **/requirements*.txt - name: Install your Dependencies run: | pip install -r requirements.txt -r requirements-dev.txt - name: Build Executable with Nuitka uses: Nuitka/Nuitka-Action@main with: nuitka-version: main script-name: your_main_program.py # many more Nuitka options available, see action doc, but it's best # to use nuitka-project: options in your code, so e.g. you can make # a difference for macOS and create an app bundle there. onefile: true - name: Upload Artifacts uses: actions/upload-artifact@v3 with: name: ${{ runner.os }} Build path: | # match what's created for the 3 OSes build/*.exe build/*.bin build/*.app/**/*If you app is a GUI, e.g. your_main_program.py should contain these
comments as explained in Nuitka Options in the code since on macOS
this should then be a bundle.
# Compilation mode, standalone everywhere, except on macOS there app bundle# nuitka-project-if: {OS} in ("Windows", "Linux", "FreeBSD"):# nuitka-project: --onefile# nuitka-project-if: {OS} == "Darwin":# nuitka-project: --standalone# nuitka-project: --macos-create-app-bundle#For good looks, you may specify icons. On Windows, you can provide an icon file, a template executable, or a PNG file. All of these will work and may even be combined:
# These create binaries with icons on Windowspython -m nuitka --onefile --windows-icon-from-ico=your-icon.png program.py python -m nuitka --onefile --windows-icon-from-ico=your-icon.ico program.py python -m nuitka --onefile --windows-icon-template-exe=your-icon.ico program.py# These create application bundles with icons on macOSpython -m nuitka --macos-create-app-bundle --macos-app-icon=your-icon.png program.py python -m nuitka --macos-create-app-bundle --macos-app-icon=your-icon.icns program.py
Note
With Nuitka, you do not have to create platform-specific icons, but instead it will convert e.g. PNG, but also other formats on the fly during the build.
Entitlements for an macOS application bundle can be added with the
option, --macos-app-protected-resource, all values are listed on
this page from Apple
An example value would be--macos-app-protected-resource=NSMicrophoneUsageDescription:Microphone
access for requesting access to a Microphone. After the colon, the
descriptive text is to be given.
Note
Beware that in the likely case of using spaces in the description part, you need to quote it for your shell to get through to Nuitka and not be interpreted as Nuitka arguments.
On Windows, the console is not opened by programs unless you say so.
Nuitka defaults to not show it, you can force it by using--console=force though, then the program will open a new terminal
Window when its executed.
Splash screens are useful when program startup is slow. Onefile startup itself is not slow, but your program may be, and you cannot really know how fast the computer used will be, so it might be a good idea to have them. Luckily, with Nuitka, they are easy to add for Windows.
For the splash screen, you need to specify it as a PNG file, and then make sure to disable the splash screen when your program is ready, e.g. has completed the imports, prepared the window, connected to the database, and wants the splash screen to go away. Here we are using the project syntax to combine the code with the creation, compile this:
# nuitka-project: --onefile# nuitka-project: --onefile-windows-splash-screen-image={MAIN_DIRECTORY}/Splash-Screen.png# Whatever this is, obviouslyprint("Delaying startup by 10s...")import time, tempfile, ostime.sleep(10)# Use this code to signal the splash screen removal.if "NUITKA_ONEFILE_PARENT" in os.environ: splash_filename = os.path.join( tempfile.gettempdir(), "onefile_%d_splash_feedback.tmp" % int(os.environ["NUITKA_ONEFILE_PARENT"]),
) if os.path.exists(splash_filename): os.unlink(splash_filename)print("Done... splash should be gone.")
...# Rest of your program goes here.For analysis of your program and Nuitka packaging, there is the Compilation Report available. You can also make custom reports by providing your template, with a few of them built-in to Nuitka. These reports carry all the detail information, e.g. when a module was attempted to be imported, but not found, you can see where that happens. For bug reporting, it is very much recommended to provide the report.
You can attach copyright and trademark information, company name, product name, and so on to your compilation. This is then used in version information for the created binary on Windows, or application bundle on macOS. If you find something that is lacking, please let us know.
By default, Nuitka compiles without --deployment which leaves a set
of safe guards and helpers on, that are aimed at debugging wrong uses of
Nuitka.
This is a new feature, and implements a bunch of protections and helpers, that are documented here.
So after compilation, sys.executable is the compiled binary. In case
of packages like multiprocessing, joblib, or loky what these
typically do is to expect to run from a full python withsys.executable and then to be able to use its options like -c
command or -m module_name and then be able to launch other code
temporarily or permanently as a service daemon.
With Nuitka however, this executes your program again, and puts these
arguments, in sys.argv where you maybe ignore them, and then you
fork yourself again to launch the helper daemons. Sometimes this ends up
spawning CPU count processes that spawn CPU count processes that... this
is called a fork bomb, and with almost all systems, that freezes them
easily to death.
That is why e.g. this happens with default Nuitka:
./hello.dist/hello.bin -l fooL -m fooM -n fooN -o fooO -p Error, the program tried to call itself with '-m' argument. Disable with '--no-deployment-flag=self-execution'.
Your program may well have its own command line parsing, and not use an
unsupported package that does attempt to re-execute. In this case, you
need at compile time to use --no-deployment-flag=self-executionwhich disables this specific guard.
Some packages output what they think is helpful information about what
the reason of a failed import might mean. With compiled programs there
are very often just plain wrong. We try and repair those in
non-deployment mode. Here is an example, where we change a message that
asks to pip install (which is not the issue) to point the user to the
include command that makes an imageio plugin work.
- module-name: 'imageio.core.imopen'
anti-bloat:
- replacements_plain:'`pip install imageio[{config.install_name}]` to install it': '`--include-module={config.module_name}` with Nuitka to include it''err_type = ImportError': 'err_type = RuntimeError' when: 'not deployment'The deployment mode is relatively new and has constantly more features
added, e.g. something for FileNotFoundError should be coming soon.
All these helpers can of course be disabled at once with--deployment but keep in mind that for debugging, you may want to
re-enable it. You might want to use Nuitka Project options and an
environment variable to make this conditional.
Should you disable them all?
We believe, disabling should only happen selectively, but with PyPI upgrades, your code changes, all of these issues can sneak back in. The space saving of deployment mode is currently negligible, so attempt to not do it, but review what exists, and if you know that it cannot affect you, or if it does, you will not need it. Some of the future ones, will clearly be geared at beginner level usage.
Binaries compiled on Windows with default settings of Nuitka and no further actions taken might be recognized by some AV vendors as malware. This is avoidable, but only in Nuitka commercial there is actual support and instructions for how to do it, seeing this as a typical commercial only need. https://nuitka.net/doc/commercial.html
For Linux standalone it is pretty difficult to build a binary that works on other Linux versions. This is mainly because on Linux, much software is built specifically targeted to concrete DLLs. Things like glibc used, are then encoded into the binary built, and it will not run with an older glibc, just to give one critical example.
The solution is to build on the oldest OS that you want to see supported. Picking that and setting it up can be tedious, so can be login, and keeping it
