Description:

pymsbuild 1.0.0

pymsbuild
This is a PEP 517 backend for building packages via MSBuild or dotnet build.
Configuration file
The file is named _msbuild.py, and is executed by running python -m pymsbuild.
The package definition specifies all the files that end up in the released packages.
from pymsbuild import *

METADATA = {
"Metadata-Version": "2.1",
"Name": "package",
"Version": "1.0.0",
"Author": "My Name",
"Author-email": "myemail@example.com",
"Description": File("README.md"),
"Description-Content-Type": "text/markdown",
"Classifier": [
"Development Status :: 3 - Alpha",
"Intended Audience :: Developers",
"Programming Language :: Python :: 3.9",
],
}

PACKAGE = Package(
"my_package",
PyFile(r"my_package\*.py"),
PydFile(
"_accelerator",
CSourceFile(r"win32\*.c"),
IncludeFile(r"win32\*.h"),
),
Package(
"subpackage",
PyFile(r"subpackage\*.py"),
),
)

Note that subpackages must be specified as a Package element, as the
nesting of Package elements determines the destination path. Otherwise you
will find all of your files flattened. Recursive wildcards are supported, however,
be aware that it is not always intuitive how the paths are going to be remapped.
Also note that without a source= named argument, all source paths are
relative to the configuration file.
pyproject.toml file
You will need this file in order for pip to build your sdist, but otherwise it's
generally easier and faster to use pymsbuild directly.
[build-system]
requires = ["pymsbuild"]
build-backend = "pymsbuild"

On Windows, a Visual Studio installation will be
required. It will be automatically detected, or the path to MSBuild can be
provided as the MSBUILD environment variable.
On other platforms, the .NET SDK will be
required. The dotnet command must be available on PATH or specified as the
MSBUILD environment variable.
If you have additional requirements for building either sdist or wheels, add
them as BuildSdistRequires or BuildWheelRequires values in METADATA. They
will be parsed after init_METADATA and/or init_PACKAGE have been called, so
may be modified by these functions.
[project] table support
There is no support for the
[project]
table at this time. All metadata that is written into the final distribution
files comes from your _msbuild.py file.
However, the pyproject.toml included in sdists is a direct copy of the one
from the root of your project. Other than the project table, sdists have no
predictable metadata for analysis tools to use, so if you want your project to
provide that metadata, feel free to list it in the pyproject.toml as well as
in your _msbuild.py (remembering to mark as
dynamic
anything that is updated by your build process).
A future release may automatically use _msbuild.py metadata to fill out
missing fields in a pyproject.toml project table, and pymsbuild init may use
the project table to initialise the configuration file. However, at this point,
both files are totally independent and the configuration file is the canonical
source of metadata.
Usage
Rebuild the current project in-place.
python -m pymsbuild

Interactively generate the _msbuild.py file with project spec.
(Or at least, it will, once implemented.)
python -m pymsbuild init

Build the project and output an sdist
python -m pymsbuild sdist

Output is put into dist by default, but can be overridden with --dist-dir
(-d).
Build the project and output a wheel
python -m pymsbuild wheel

Output is put into dist by default, but can be overridden with --dist-dir
(-d).
Clean any recent builds
python -m pymsbuild clean

Advanced Examples
Dynamic METADATA
Metadata may be dynamically generated, either on import or with the
init_METADATA function. This function is called and must either
return the metadata dict to use, or update METADATA directly.
However, if a PKG-INFO file is found adjacent to the configuration
file, it will be used verbatim. Sdist generation adds this file, so all
metadata is static from that point onward. init_METADATA is not
called in this case.
from pymsbuild import *

METADATA = {
"Metadata-Version": "2.1",
"Name": "package",
"Version": os.getenv("VERSION", "1.0.0"),
"Author": "My Name",
"Author-email": "myemail@example.com",
"Description": File("README.md"),
"Description-Content-Type": "text/markdown",
"Classifier": [
"Development Status :: 3 - Alpha",
"Intended Audience :: Developers",
"Programming Language :: Python :: 3.9",
],
}

def init_METADATA():
if os.getenv("BUILD_BUILDNUMBER"):
METADATA["Version"] = f"1.0.{os.getenv('BUILD_BUILDNUMBER', '')}"
# Updated METADATA directly, so no need to return anything

Also see the earlier section regarding the pyproject.toml project table (and
the fact that it is not used by pymsbuild, but will be added to your sdist
without modification).
Separate packages
Packages are just Python objects, so they may be kept in variables and
used later. They also expose a members attribute, which is a list, so
that members can be added or inserted later, as well as find, findall and
insert methods to help (see the Dynamic Packages section below).
After the entire module is executed, the package in PACKAGE is the
only one used to generate output.
P1 = Package(
"submodule",
PyFile(r"src\submodule\__init__.py")
)

P2 = Package(
"submodule_2",
PyFile(r"src\submodule_2\__init__.py")
)

PACKAGE = Package("my_package", P1)
PACKAGE.members.append(P2)

Wildcard handling
Files can be added recursively using wildcard operators. These are
evaluated at generation time by pymsbuild and not by MSBuild/
dotnet build, as it allows greater control over target names.
PACKAGE = Package(
"my_package",
# All .py files, relative to the 'src' directory
PyFile(r"**\*.py"),
# All license files, if any, with path separators converted to '-'
File(r"**\license*", flatten="-", allow_none=True),
# All .bin files from all data directories, moved to the root
File(r"**\data\*.bin", flatten=True),
source="src"
)

flatten specifies the string sequence to replace path separators in
the name. Passing True indicates that only the file name should be
retained.
allow_none merely suppresses a build-time error when the wildcard
fails to match any files. This is usually an important problem, and
should be suppressed with care.
The flatten and allow_none properties are not written to the build
file. However, they are case-sensitive while MSBuild is not, so the
capitalised versions will be ignored for this processing and passed
through.
Final install location (also known as the element's name) are generated
from the default name (source file name or name argument and all
package names in the hierarchy) combined with the pattern according to
these rules:

if the pattern contains no wildcards, the default name is preserved
each segment from the first one containing a wildcard will be joined
to the parent of the default name
if the pattern filename contains no wildcards, it is preserved in the
final name. Otherwise, it is replaced by matched files

These rules ensure consistency across many forms of paths, making it
reliable to use calculated absolute paths with wildcards (for example,
a package extending the build system to add its own files). To create
a directory in the destination, use a new Package element:
# Installs as 'A/__init__.py'
PACKAGE = Package("A", PyFile("B/__init__.py"))
PACKAGE = Package("A", PyFile("B/source.py", name="__init__.py"))

# Installs as 'A/*.txt'
PACKAGE = Package("A", File("B/*.txt"))

# All of these install as 'A/B/*.txt'
PACKAGE = Package("A", Package("B", File("B/*.txt")))
PACKAGE = Package("A", Package("B", File("*.txt"), source="B"))
PACKAGE = Package("A", File("*/*.txt")) # assuming no other matches

Specifying the Name metadata (as opposed to name, which is a
keyword argument) will override the destination name of every matched
file. This is applied before flattening, and so will preserve the
relative path in whatever form is specified by flatten. To bypass
this additional processing and use the name as an MSBuild literal,
wrap it in a ConditionalValue with no condition:
PACKAGE = Package(
"my_package",
File("**/*.dat", Name=ConditionalValue("%(Filename)-1.dat")),
)

For more complex transforms on filename, we recommend using the
init_PACKAGE function described below.
Dynamic packages
After metadata processing, if an init_PACKAGE(tag=None) function
exists it will be called with the intended platform tag. It must modify
or return PACKAGE. This function is called for in-place, sdist and
wheel generation, however, for sdists (and any scenario that should not
generate binaries), tag will be None. Otherwise, it will be a
string like cp38-cp38-win32.
PACKAGE = Package(
"my_package",
PyFile(r"my_package\*.py"),
)

def init_PACKAGE(tag=None):
if tag and tag.endswith("-win_amd64"):
data_file = generate_data_amd64()
PACKAGE.members.append(File(data_file))

Note that all files to be included in an sdist must be referenced when
tag is None. Conditional compilation is best performed using conditions
in the package elements, rather than using init_PACKAGE. However, if you
are going to use init_PACKAGE, you should remove elements rather than
adding them if they should be included in your sdist.
Files added as part of a wildcard can be removed by adding a RemoveFile
element. These may be added dynamically during init_PACKAGE, and must
appear after the element that included the files.
PACKAGE = Package(
"my_package",
PyFile(r"my_package\*.py"),
RemoveFile(PyFile, r"my_package\_internal.py"),
)

To exclude files from a wildcard in the first place, chain the .excluding
method on the original element. The pattern will be evaluated in exactly the
same way as the inclusion pattern, and any paths that match will be omitted.
PACKAGE = Package(
"my_package",
PyFile(r"my_package\*.py").excluding("my_package\internal*.py"),
)

Alternatively, a condition may be added to the file pattern to only include
files matching MSBuild style conditions. Because these will be applied to
item groups, the %() metadata syntax should be used to access information
for the element being added. Either the .if_ method or the
ConditionalValue wrapper may be used.
PACKAGE = Package(
"my_package",
PyFile(r"my_package\*.py").if_("%(Filename) != '_internal'"),
File(ConditionalValue("*.txt", condition="%(Filename.StartsWith(`internal`))")),
)

Package members can be located during the dynamic stage using the
find and findall functions. These take a path of member identifiers
(typically their name property) and will return those that match.
'**' segments are supported for recursive searches.
PACKAGE = Package(
"my_package",
Package("sub1", File("license.txt")),
Package("sub2", File("license.txt")),
)

def init_PACKAGE(tag=None):
for e in PACKAGE.findall("sub*/license.txt"):
e.name = "LICENSE"

When inserting members, the insert function combines a find with
the insert, and supports offset and range options. In general, only
subclassed element types should insert additional elements, and only
into themselves at construction.
class MyPydFile(PydFile):
def __init__(self, name, *members, **options):
super().__init__(name, *members, **options)
self.insert(
# Member path to insert before - this one is inherited from PydFile
self.CommonToolsetImports.name,
# Member to insert (in this case, an iterable)
[Property(PROP1, VALUE1), Property(PROP2, VALUE2)],
# Offset it by 1, so inserts after the found element (default 0)
offset = 1,
# Iterate over the insertion value; otherwise insert it as-is
range = True
)

Source offsets
If you keep your source in a src folder (recommended), provide the
source= argument to Package in order to properly offset filenames.
Because it is a named argument, it must be provided last.
This is important for sdist generation and in-place builds, which need
to match package layout with source layout. Simply prefixing filename
patterns with the additional directory is not always sufficient.
Note that this will also offset subpackages, and that subpackages may
include additional source arguments. However, it only affects
sources, while the package name (the first argument) determines where
in the output the package will be located. In-place builds will create
new folders in your source tree if it does not match the final
structure.
PACKAGE = Package(
"my_package",
PyFile(r"my_package\__init__.py"),
source="src",
)

Project file override
Both Package and PydFile types generate MSBuild project files and
execute them as part of build, including sdists. For highly customised
builds, this generation may be overridden completely by specifying the
project_file named argument. All members are then ignored.
By doing this, you take full responsibility for a valid build,
including providing a number of undocumented and unsupported targets.
Recommendations:

lock your pymsbuild dependency to a specific version in pyproject.toml
generate project files first and modify, rather than writing by hand
(pass --temp-dir to specify the path where they will be generated)
read the pymsbuild source code, especially the targets folder
consider contributing/requesting your feature or developing an extension
(see pymsbuild-winui and
pymsbuild-rust for examples)

PACKAGE = Package(
"my_package",
PydFile("_accelerator", project_file=r"src\accelerator.vcxproj")
)

Compiler/linker arguments
Rather than overriding the entire project file, there are a number of
ways to inject arbitrary values into a project. These require
familiarity with MSBuild files and the toolsets you are building with.
The Property element inserts a <PropertyGroup> with the value you
specifiy at the position in the project the element appears.
Note that project files also interpret (most) named arguments as
properties, so the two properties shown here are equivalent.
PYD = PydFile(
"module",
Property("WindowsSdkVersion", "10.0.18363.0"),
...
# Alternative to Property(), but named arguments must be
# specified last, so using Property() may be more readable
WindowsSdkVersion="10.0.18363.0",
)

The ItemDefinition element inserts an <ItemDefinitionGroup> with
the type and metadata you specify at the position in the project the
element appears. These generally apply metadata to all subsequent items
of that type.
PYD = PydFile(
"module",
ItemDefinition("ClCompile", PreprocessorDefinitions="Py_LIMITED_API"),
...
)

The ConditionalValue item may wrap any element value to add
conditions or concatenate the value. This may also be used on source
arguments for file elements.
...
Property("Arch", ConditionalValue("x86", condition="$(Platform) == 'Win32'")),
Property("Arch", ConditionalValue("x64", if_empty=True)),
...
ItemDefinition(
"ClCompile",
AdditionalIncludeDirectories=
ConditionalValue(INCLUDES + ";", prepend=True),
ProprocessorDefinitions=
ConditionalValue(";Py_LIMITED_API", append=True),
),
...

The Prepend shortcut may be used to reduce the amount of text for
property values. Remember to include the appropriate separator. It is
usually a semicolon.
...
ItemDefinition(
"ClCompile",
AdditionalIncludeDirectories=Prepend(INCLUDES + ";"),
ProprocessorDefinitions=Prepend("Py_LIMITED_API;"),
),
...

ConditionalValue may also be used to dynamically update values in the
init_PACKAGE function, allowing you to keep the structure mostly
static but insert values from the current METADATA (which is fully
evaluated by the time init_PACKAGE is called). This saves having to
access internal members of other types in order to replace literal
values.
Assign a ConditionalValue to a variable without specifying any
condition, then use the variable in a later Property element and
modify its value attribute in init_PACKAGE.
VER = ConditionalValue("1.0.0")

PYD = PydFile(
"module",
Property("Version", VER),
CSourceFile(r"src\*.c"),
IncludeFile(r"src\*.h"),
)

def init_PACKAGE(tag):
VER.value = METADATA["Version"]

As a last resort, the LiteralXml element inserts plain text directly
into the generated file. It will be inserted as a child of the
top-level Project element.
...
LiteralXml("<Import Project='my_props.props' />"),
...

Version info for DLLs/PYDs
Platform: Windows
To embed version info into a compiled extension module, add a VersionInfo
element into the PydFile. All the fields from
https://learn.microsoft.com/en-us/windows/win32/menurc/versioninfo-resource
are available, using the names as shown in the tables (e.g.
FILEVERSION for the '1,0,0,0' fields and FileVersion for the string
table entry).
The recommended usage is to add a default instance into your project and then
use init_METADATA to find it again and update based on the final metadata.
PACKAGE = Package(
"package",
PydFile("mod1", VersionInfo()),
PydFile("mod2", VersionInfo()),
)

def init_METADATA():
# Update our metadata
METADATA["Version"] = calculate_current_version()
# Ensure built modules reflect these updates
for vi in PACKAGE.findall("*/VersionInfo"):
vi.from_metadata(METADATA)

from_metadata will fill in any empty fields from the set of metadata that is
passed in.
Alternate config file
To use a configuration file other than _msbuild.py, specify the
--config (-c) argument or the PYMSBUILD_CONFIG environment
variable.
python -m pymsbuild --config build-spec.py sdist
python -m pymsbuild --config build-spec.py wheel

# Alternatively
$env:PYMSBUILD_CONFIG = "build-spec.py"
python -m pymsbuild sdist wheel

Generated sdists will rename the configuration file back to
_msbuild.py in the package to ensure that builds work correctly.
There is no need to override the configuration file path when
building from sdists.
Note that this is different from the PYMSBUILD_CONFIGURATION
variable, which is used to select debug/release settings for compiled
modules.
Cross-compiling wheels
Cross compilation may be used by overriding the wheel tag, ABI tag,
or build platform, as well as the source for Python's includes and
libraries. These all use environment variables, to ensure that the
same setting can flow through a package installer's own process.
It is also possible to permanently override the wheel tag by adding
a 'WheelTag' metadata value, or the ABI tag by adding an 'AbiTag'
metadata value.
The wheel tag is used for the generated wheel file, and to fill in a
missing ABI tag and platform.
The ABI tag is used for any native extension modules, and to fill in
a missing platform.
Any * elements in the wheel tag are filled in from other locations.
For example, specifying *-none-any will infer the interpreter field
from the current runtime, whil py3-none-* will infer the platform
from the currnet system (or a specific ABI tag).
The platform is used to determine the MSBuild target platform. It
cannot yet automatically select the correct Python libraries, and so
you will need to set PYTHON_INCLUDES and PYTHON_LIBS (or with a
PYMSBULID_ prefix) environment variables as well to locate the
correct files.
You can override the platform toolset with the 'PlatformToolset'
metadata value, for scenarios where this information ought to be
included in an sdist.
The set of valid platforms for auto-generated .pyd project files are
hard-coded into pymsbuild and are currently Win32, x64, ARM and
ARM64. Custom project files may use whatever they like. These
platforms should behave properly cross-platform, though in general only
x64 and ARM64 are supported.
# Directly specify the resulting wheel tag
# This is used for the wheel filename/metadata
$env:PYMSBUILD_WHEEL_TAG = "py38-cp38-win_arm64"

# Directly set the ABI tag (or else taken from wheel tag)
# This is used for extension module filenames
$env:PYMSBUILD_ABI_TAG = "cp38-win_arm64"

# Specify the Python platform (or else taken from ABI tag)
# This is used for MSBuild options
$env:PYMSBUILD_PLATFORM = "win_arm64"

# Specify the paths to ARM64 headers and libs
$env:PYTHON_INCLUDES = "$pyarm64\Include"
$env:PYTHON_LIBS = "$pyarm64\libs"

# If necessary, specify an alternate C++ toolset
$env:PLATFORMTOOLSET = "Intel C++ Compiler 19.1"

Cython
Cython support is available from the pymsbuild.cython module.
from pymsbuild import PydFile, ItemDefinition
from pymsbuild.cython import CythonIncludeFile, CythonPydFile, PyxFile

PACKAGE = CythonPydFile(
"cython_module",
ItemDefinition("PyxCompile", IncludeDirs=PYD_INCLUDES),
CythonIncludeFile("mod.pxd"),
PyxFile("mod.pyx"),
)

The CythonPydFile type derives from the regular PydFile and also
generates a C++ project, so all options that would be available there may
also be used.
The PyxCompile.IncludeDirs metadata specifies search paths for Cython
headers (*.pxd). You may also need to specify
ClCompile.AdditionalIncludeDirectories for any C/C++ headers.
Two-Step Builds
By default, the sdist and wheel commands will perform the entire
process in a single invocation. However, sometimes there are build steps
that must be manually performed between compilation and packaging.
To run the build in two stages, invoke as normal, but add the
--layout-dir argument followed by a directory. The package will be
laid out in this directory so that you can perform any extra processing.
Later, use the pack command and specify the --layout-dir again. If
you have added new files into the layout directory, specify each with an
--add option (filenames starting with @ are treated as
newline-separated, UTF-8 encoded text files listing each new file). These
paths may be absolute or relative to the layout directory, but only files
located within the layout directory will be included.
All other options are retained from the original invocation.
python -m pymsbuild sdist --layout-dir tmp

# Generate additional metadata in tmp/EXTRA.txt

python -m pymsbuild pack --layout-dir tmp --add tmp/EXTRA.txt

# List many additional files in build/TO_ADD.txt

python -m pymsbuild pack --layout-dir tmp --add @build/TO_ADD.txt

Experimental Features
DLL Packing
Experimental. (POSIX is very experimental)
DLL Packing is a way to compile a complete Python package (.py source
and resource files) into an extension module. It is basically equivalent
to packing in a ZIP file, except that additional native code may also be
included (though not an entire native module), and the whole file may be
cryptographically signed and validated by the operating system.
DllPackage is a drop-in substitute for the Package type. It will
generate a native extension module the same as the PydFile type,
but only includes Python source and resource files.
from pymsbuild import *
from pymsbuild.dllpack import *

PACKAGE = DllPackage(
"packed_package",
PyFile("__init__.py"),
File("data.txt"),
...
)

An entire existing library, such as cryptography could be packed
like this:
from pymsbuild import *
from pymsbuild.dllpack import *

MODULE_TO_PACK = "cryptography"

from importlib.util import find_spec
spec = find_spec(MODULE_TO_PACK)
if not spec:
raise RuntimeError(f"{MODULE_TO_PACK} must be installed")

PACKAGE = DllPackage(
MODULE_TO_PACK,
PyFile("**/*.py"),
PydRedirect("**/*.pyd"),
source = spec.submodule_search_locations[0],
)

See the azure-pack sample in our source repository for a more
complete example.
DllPackage is a subclass of PydFile, and so all logic or elements
by that type are also available. ClCompile elements will be compiled
and linked into the output and functions may be exposed in the root of
the package using the Function element.
// extra.c

PyObject *my_func(PyObject *, PyObject *args, PyObject **kwargs) {
...
}

PACKAGE = DllPackage(
"packed_package",
PyFile("__init__.py"),
CSourceFile("extra.c"),
CFunction("my_func"),
...
)

Nested extension modules
To allow referencing other extension modules that would normally be
nested within the module, add a PydRedirect element and reference the
extension module. The filename does not have to match the original
name, or even need to be a normally importable name, as it will be
passed directly to the module loader. The file will be included in your
wheel in the expected location (alongside the packed DLL). Wildcards
are supported.
PACKAGE = DllPackage(
"packed",
PydRedirect(source="packed/nested.pyd", name="packed-nested.pyd"),
...
)

Other PydFile modules may be nested inside the DllPackage, which
will automatically add a redirect, as well as building the module.
The nested module will be built using the name specified and sit
adjacent to the packed module, but should be imported via the
packed module.
The ImportName metadata may be specified on either a PydRedirect or
a PydFile to specify the name that must be used to import the module.
Redirected extension modules do not need to have an importable name
when ImportName is specified. You might include an invalid character
in the filename to ensure the module is not importable directly.
When specifying ImportName, the name of the packed DLL must be used
as the first part.
PACKAGE = DllPackage(
"packed",
PydRedirect("module/nested.pyd", ImportName="packed.nested"),
...
)

Encryption
To encrypt your content using symmetric AES encryption, provide the
name of the environment variable holding your key as the
EncryptionKeyVariable option. The key will need to be a valid size
(usually 16, 24 or 32 bytes) when encoded to UTF-8 or decoded from
base 64. Base 64 keys should start with base64:.
The same variable will need to be set when importing the module. It is
your responsibility to protect the key! The benefit of this encryption
is best realised when you avoid storing the key to disk. That way, an
attacker who steals a copy of your module is unlikely to have access to
the key. An attacker with access to a running copy of your module will
be able to easily extract the key.
PACKAGE = DllPackage(
"package",
...,
EncryptionKeyVariable="MY_KEY_VARIABLE"
)

> $env:MY_KEY_VARIABLE="base64:MDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDAwMDA="
> python -m pymsbuild
> del env:\MY_KEY_VARIABLE
> python -c "import package"
ImportError: Module cannot be decrypted

Redirected or nested extension modules are not encrypted.
Cross-platform builds
Experimental.
With the .NET SDK installed,
pymsbuild is able to run builds on platforms other than Windows.
The dotnet command must be available on PATH or specified as the
MSBUILD environment variable.
In general, no platform-specific modifications to a build script are
required. Cython and pyd builds are transparently mapped to the target
system. To run build-time actions for specific platforms, add them to
init_PACKAGE and check the tag argument to determine the target
platform.
When building native components on POSIX, a python3-config script is
needed to determine compilation options. By default, only the location
adjacent to the running interpreter is checked. This may be overridden
by setting the PYTHON_CONFIG variable to the preferred command.

License

For personal and professional use. You cannot resell or redistribute these repositories in their original state.

Share

• Share on Facebook
• Share on Twitter