Simple (and extensible) python3 externals for MaxMSP.
Cross-platform: currently builds 'natively' on macOS (x86_64
or arm64
) and Windows (with MSVC).
repo - https://github.com/shakfu/py-js
This project started out as an attempt (during a covid-19 lockdown) to develop a basic python3 external for Max/MSP. It then evolved into an umbrella project for exploring different ways of using python3 in Max/MSP.
Along the way, a number of externals have been developed for use in a live Max environment:
Python3 Externals:
name | sdk | lang | description |
---|---|---|---|
py | max-sdk | c | well-featured, many packaging options + cython api |
pyjs | max-sdk | c | js-friendly -- written as a Max javascript-extension |
mamba | max-sdk | c | single-header c library to nest a python3 interpreter in any external |
krait | max-sdk | c++ | single-header c++ library to nest a python3 interpreter in any external |
cobra | max-sdk | c | python3 external providing deferred and clocked function execution |
mxpy | max-sdk | c | a translation of pdpython into Max |
zedit | max-sdk | c | a web-based python editor using codemirror and the mongoose embedded webserver. |
pymx [2] | min-devkit | c++ | concise, modern, using pybind11 |
[1] pymx has been moved to its own github project because it uses the min-devkit sdk.
Alternative Python Implementation Externals:
name | sdk | lang | description |
---|---|---|---|
pktpy | max-sdk | c++ | uses the pocketpy single-header c++ library |
mpy [2] | max-sdk | c | a proof-of-concept embedding micropython |
[2] mpy
is not enabled by default since it is still in early stages and more of a proof-of-concept to embed micropython in an external. To build it use the -DBUILD_MICROPYTHON_EXTERNAL
option with cmake.
ZeroMQ-related Externals:
name | sdk | lang | description |
---|---|---|---|
jmx | max-sdk | c | explore how to embed a jupyter client or kernel in an external |
zpy | max-sdk | c | uses zeromq for 2way-comms with an external python process |
zthread | max-sdk | c | exploration of zeromq and Max threads |
Note: zeromq externals are not enabled by default since they require zeromq libraries to be installed. To build them use the -DBUILD_ZEROMQ_EXTERNALS
option with cmake.
The common objective in these externals is to help use and distribute python code and libraries in Max applications. Many can be considered experimental, with 80% of development time going to the first two externals (py
and pyjs
). Please see below for an overview and feature comparison.
At the time of this writing, and since the switch to max-sdk-base, the project has the following compatibility profile:
-
macOS: both x86_64 and Apple Silicon compatible. Note that the project intentionally only produces 'native' (
x86_64
xorarm64
) externals with no current plans for 'fat' or universal externals to serve both architectures. You can download codesigned, notarizedx86_64
-based andarm64
-based python3 externals from the releases section. -
Windows: windows support was provided relatively recently, with currently all Python3 externals and also the
pktpy
projects building without issues on Windows. The only caveat is that as of this writing python3 externals are dynamically linked to the local Python3.dll
and are therefore not relocatable. One idea to overcome this constraint is to include the external's dependencies in the 'support' folder. This will hopefully be addressed in future iterations. Thepktpy
external, however, is fully portable and self-contained.
This README will mostly cover the first two mature externals (py.mxo
and pyjs.mxo
) and their many build variations available via a custom python-based build system which was specifically developed to cater for different scenerios of packaging and deploying the externals in Max packages and standalones.
If you are interested in any of the other subprojects, please look into the respective folder in the py-js/source/projects
section.
The Quickstart section below covers general setup for all of the externals, and will get you up and running with the py
and pyjs
externals. The Building Experimental Externals using Cmake section provides additional info to build the other remaining externals, and the Building self-contained Python3 Externals for Packages and Standalones section covers more advanced building and deployment scenarios.
Please feel free to ask questions or make suggestions via the project's github issue tracker.
The following will give you a sense of the feature differences between the two core python3 max externals:
A general purpose Max external that embeds a python3 interpreter and is made up of three integrated parts which make it quite straightforward to extend:
-
The
py
Max external which is written in c using both the Max c-api and the Python3 c-api. -
A pure python module,
py_prelude.py
which is converted topy_prelude.h
and compiled withpy
and then pre-loaded into theglobals()
namespace of everypy
instance. -
A builtin
api
module which is derived from a cython-based wrapper of a subset of the Max c-api.
The following provides a brief view of key attributes and methods:
globals
obj_count : number of active py objects
registry : global registry to lookup object names
patchers
subpatchers
py_repl : a basic single line repl for py
py_repl_plus : embeds a py object in a py_repl
py max external
attributes
name : unique object name
file : file to load into editor
autoload : load file at start
pythonpath : add path to python sys.path
debug : switch debug logging on/off
methods (messages)
core
import <module> : python import to object namespace
eval <expression> : python 'eval' semantics
exec <statement> : python 'exec' semantics
execfile <path> : python 'execfile' semantics
extra
assign <var> [arg] : max-friendly msg assignments to py object namespace
call <pyfunc> [arg] : max-friendly python function calling
pipe <arg> [pyfunc] : process py/max value(s) via a pipe of py funcs
fold <f> <n> [arg] : applies a two-arg function cumulatively to a sequence
code <expr|stmt> : alternative way to eval or exec py code
anything <expr|stmt> : anything version of the code method
time-based
sched <t> <fn> [arg] : defer a python function call by t millisecs
code editor
read <path> : read text file into editor
load <path> : combo of read <path> -> execfile <path>
run : run the current code in the editor
interobject
scan : scan patcher and store names of child objects
send <msg> : send an arbitrary message to a named object
meta
count : give a int count of current live py objects
inlets
single inlet : primary input (anything)
outlets
left outlet : primary output (anything)
middle outlet : bang on failure
right outlet : bang on success
General purpose python3 jsextension, which means that it's a c-based Max external which can only be accessed via the javascript js
interface.
pyjs max external (jsextension)
attributes
name : unique object name
file : file to load in object namespace
pythonpath : add path to python sys.path
debug : switch debug logging on/off
methods
core (messages)
import <module> : python import to object namespace
eval <expression> : python 'eval' semantics
exec <stmnt> : python 'exec' semantics
execfile <path> : python 'execfile' semantics
extra
code <expr|stmt> : eval/exec/import python code (see above)
in-code (non-message)
eval_to_json <expr> : python 'eval' returns json
This repo has a git submodule dependency with max-sdk-base. This is quite typical for Max externals.
This means you should git clone
as follows:
git clone https://github.com/shakfu/py-js.git
git submodule init
git submodule update
or more concisely:
git clone --recursive https://github.com/shakfu/py-js.git
Since Windows support still is relatively new, no releases have been made pending further testing.
Currently, the externals which are enabled by default in this project can be built with only a few requirements:
-
Install Visual Studio Community Edition or use the commercial versions as you like.
-
Install Python3 for Windows from python.org
-
(Optional) since Visual Studio has its captive cmake, you can use that, but it is preferable to install cmake independently.
After installation of the above you can build the externals inside your Documents/Max 8/Packages
folder as follows:
git clone --recursive https://github.com/shakfu/py-js
cd py-js
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build . --config Release
Open one of the .maxhelp
files or any of the files in the patchers
folders to see how things work.
As mentioned earlier, the py
and pyjs
objects are the most mature and best documented of the collection. Happily, there is also no need to compile them as they are available for download, fully codesigned and notarized, from the releases section.
If you'd rather build them or any of the other externals yourself then the process is straightforward:
-
You should have a modern
python3
cpython implementation installed on your Mac: preferably either from python.org or from Homebrew. Note that even system python3 provided by Apple will work in a number of cases. Python versions from 3.8 to 3.12 are tested and known to work. -
Make sure you also have Xcode installed.
-
Git clone the
py-js
repo as per the above method to a path without a space and without possible icloud syncing (i.e don't clone to$HOME/Documents/Max 8/Packages
) [?] and run the following in the cloned repo:make setup
The above will initialize and update the required git submodules and symlink the repo to
$HOME/Documents/Max 8/Packages/py-js
to install it as a Max Package and enable you to test the externals and run the patches.[?] It is possible to install
py-js
directly into$HOME/Documents/Max 8/Packages
, but it requires moving the place of compilation to a location in your filesystem that is not exposed to errors due to icloud syncing or spaces in the path. This split is possible, but it is not recommended for the purposes of this quickstart. -
Install cython via
pip3 install cython
, required for translating the cython-basedapi.pyx
, which wraps the the Max c-api, to c. -
To build only the
py
andpyjs
externals, type the following in the root directory of thepy-js
project (other installation options are detailed below):make
Note that typing make
here is the same as typing make default
or make all
. This will create two externals py.mxo
and pyjs.mxo
in your externals
folder. These are relatively small in size and are linked to your system python3 installation. This has the immediate benefit that you have access to your curated collection of existing python packages. The tradeoff is that these externals are dynamically linked with local dependencies and therefore not usable in standalones and relocatable Max packages.
No worries, if you need portable relocatable python3 externals for your package or standalone then make sure to read the Building self-contained Python3 Externals for Packages and Standalones section
Open up any of the patch files in the patchers
directory of the repo or the generated Max package, and also look at the .maxhelp
patchers to understand how the py
and the pyjs
objects work.
You can also use cmake
to build all externals using similar methods to the max-sdk
.
First make sure you have completed the Quickstart section above. Next you will install cmake
if necessary and a couple of additional dependencies for some of the subprojects. Of course, skip what is already installed:
brew install cmake zmq czmq
Now you can build all externals (including py
and pyjs
) in one shot using cmake:
make projects
After doing the above, the recommended iterative development workflow is to make changes to the source code in the respective project and then cd py-js/build
and cmake --build .
. This will cause cmake to only build modified projects efficiently.
Note that for some of the less developed externals and more experimental features please don't be surprised if Max seg-faults (especially if you start experimenting with the cython wrapped api
module which operates on the c-level of the Max SDK).
Also note that for py
and pyjs
externals the cmake
build method described does not yet create self-contained python externals which can be used in Max Packages and Standalones.
The following section addresses this requirement.
Currently, the py
and pyjs
externals can be additionally custom-built for the express purpose of packaging these for standalone and Max packages using a custom python build manager. This allows highly-customized build variants
to be produced.
The Makefile
in the project root provides a simplified interface to this builder. See the Current Status of Builders section for further information.
idx | command | type | format | py size | pyjs size |
---|---|---|---|---|---|
1 | make static-ext |
static | external | 9.0 | 8.8 |
2 | make static-tiny-ext |
static | external | 6.7 | 6.2 [2] |
3 | make shared-ext |
shared | external | 16.4 | 15.8 |
4 | make shared-tiny-ext |
shared | external | 6.7 | 6.2 [2] |
5 | make framework-pkg |
framework | package | 22.8 | 22.8 [3] |
[2] In this table, size figures are for python 3.10.x but for python 3.11.4 they increase to 8.5 MB and 8.1 respectively. Generally, external size increases with each new python version as features are added, but this is also somewhat mitigated by the removal of deprecated builtin packages and extensions. If you want to achieve the theoretical minimal size for the py
and pyjs
externals, use python 3.8.x and/or a tiny variant (with a more recent version). Another option, if you need circa 1 MB size for a self-contained external, look at the pktpy
subproject in this repo. Note the size of externals in Python 3.12.4 (although some of extra size is attributed improved ssl integration):
idx | command | type | format | py size | pyjs size |
---|---|---|---|---|---|
1 | make static-ext |
static | external | 15.0 | 13.3 |
2 | make static-tiny-ext |
static | external | 11.4 | 9.8 |
3 | make shared-ext |
shared | external | 20.4 | 18.7 |
4 | make shared-tiny-ext |
shared | external | 11.4 | 9.6 |
5 | make framework-ext |
shared | external | 22.5 | 20.8 |
[3] Size, in this case, is not the individual external but the uncompressed size of the package which includes patches, help files and both externals. This can also vary by python version used to compile the external.
This section assumes that you have completed the Quickstart above and have a recent python3 installation (python.org, homebrew or otherwise).
Again, if you'd rather not compile anything there are self-contained python3 externals which can be included in standalones in the releases section.
If you don't mind compiling (and have xcode installed) then pick one of the following options:
-
To build statically-compiled self-contained python3 externals:
make static-ext
You may also prefer the tiny variant:
make static-tiny-ext
-
To build self-contained python3 exernals which include a dynamically linked libpythonX.Y.dylib:
make shared-ext
or for the corresponding tiny variant:
make shared-tiny-ext
-
To build python3 externals in a package, linked to a python installation in its
support
foldermake framework-pkg
With all of the above options, a python3 source distribution (matching your own python3 version) is automatically downloaded from python.org with dependencies, and then compiled into a static or shared version of python3 which is then used to compile the externals.
At the end of this process you should find two externals in the py-js/externals
folder: py.mxo
and pyjs.mxo
.
Although the above options deliver somewhat different products (see below for details), with options (1) and (2) the external 'bundle' contains an embedded python3 interpreter with a zipped standard library in the Resources
folder and also has a site-packages
directory for your own code; with option (3), the externals are linked to, and have been compiled against, a relocatable python3 installation in the support
folder.
Depending on your choice above, the python interpreter in each external is either statically compiled or dynamically linked, and in all three cases we have a self-contained and relocatable structure (external or package) without any non-system dependencies. This makes it appropriate for use in Max Packages and Standalones.
There are other build variations which are discussed in more detail below. You can always see which ones are available via typing make help
in the py-js
project folder:
$ make help
>>> general
make projects : build all subprojects using standard cmake process
>>> pyjs targets
make : non-portable pyjs externals linked to your system
make homebrew-pkg : portable package w/ pyjs (requires homebrew python)
make homebrew-ext : portable pyjs externals (requires homebrew python)
make shared-pkg : portable package with pyjs externals (shared)
make shared-ext : portable pyjs externals (shared)
make shared-tiny-ext : tiny portable pyjs externals (shared)
make static-ext : portable pyjs externals (static)
make static-tiny-ext : tiny portable pyjs externals (static)
make framework-pkg : portable package with pyjs externals (framework)
make framework-ext : portable pyjs externals (framework)
make relocatable-pkg : portable package w/ more custom options (framework)
>>> python targets
make python-shared : minimal shared python build
make python-shared-ext : minimal shared python build for externals
make python-shared-pkg : minimal shared python build for packages
make python-static : minimal statically-linked python build
make python-framework : minimal framework python build
make python-framework-ext : minimal framework python build for externals
make python-framework-pkg : minimal framework python build for packages
make python-relocatable : custom relocatable python framework build
If you would like to see which build variations are compatible with your current setup, there's an automated test which attempts to compile all build variations in sequence and will log all results to a logs
directory:
make test
This can take a long time, but it is worth doing to understand which variants work on your particular setup.
If you want to test or retest one individual variant, just prefix test-
to the name of variant as follows:
make test-shared-pkg
If you have downloaded any pre-built externals from releases or if you have built self-contained python externals as per the methods above, then you should be ready to use these in a standalone.
To release externals in a standalone they must be codesigned and notarized. To this end, there are scripts in py-js/source/projects/py/scripts
to make this a little easier.
If you included py.mxo
as an external in your standalone, then you should have no issue as Max will install it automatically during its build-as-standalone process.
You can test if it works without issues by building either of these two example patcher documents, included in py-js/patchers
, as a max standalone:
-
py_test_standalone_info_py.maxpat
-
py_test_standalone_only_py.maxpat
Open the resulting standalone and test that the py
object works as expected.
To demonstrate the above, a pre-built standalone that was built using exactly the same steps as above is in the releases section: py_test_standalone_demo.zip
.
If you opted to include pyjs.mxo
as an external in your standalone, then it may be a litte more involved:
You can first test if it works without issues by building 'a max standalone' from the test_standalone_pyjs.maxpat
patcher which is included in py-js/patchers/tests/test_standalone
.
Open the resulting standalone and test that the pyjs
object works as expected. If it doesn't then try the following workaround:
To fix a sometimes recurrent issue where the standalone build algorithm doesn't pick up pyjs.mxo
: if you look inside the built standalone bundle, py_test_standalone_only_pyjs.app/Contents/Resources/C74/externals
you may not find pyjs.mxo
. This is likely a bug in Max 8 but easily resolved. Fix it by manually copying the pyjs.mxo
external into this folder and then copy the javascript
and jsextensions
folders from the root of the py-js
project and place them into the pyjs_test_standalone.app/Contents/Resources/C74
folder. Now re-run the standalone app again and now the pyjs
external should work. A script is provided in py-js/source/projects/py/scripts/fix-pyjs-standalone.sh
to do the above in an automated way.
Please read on for further details about what the py-js externals can do.
Have fun!
This overview will cover the following two external implementations:
-
The
py
external provides a more featureful two-way interface between Max and python in a way that feels natural to both languages. -
The
pyjs
max external/jsextension provides aPyJS
class and a minimal subset of thepy
external's features which work well with the Maxjs
object and javascript code (like returning json directly from evaluations of python expressions).
Both externals have access to builtin python modules and the whole universe of 3rd party modules, and further have the option of importing a builtin api
module which uses cython to wrap selective portions of the max c-api. This allows regular python code to directly access the max-c-api and script Max objects.
There are 3 general deployment scenarios:
-
Linked to system python. Linking the externals to your system python (homebrew, built from source, etc.) This has the benefit of re-using your existing python modules and is the default option.
-
Embedded in package. Embedding the python interpreter in a Max package: in this variation, a dedicated python distribution (zipped or otherwise) is placed in the
support
folder of thepy/js
package (or any other package) and is linked to thepy
external orpyjs
extension (or both). This makes it size efficient and usable in standalones. -
Embedded in external. The external itself as a container for the python interpreter: a custom python distribution (zipped or otherwise) is stored inside the external bundle itself, which can make it portable and usable in standalones.
As of this writing all three deployment scenarios are availabe, however it is worth looking more closely into the tradeoffs in each case, and a number of build variations exist. This topic is treated in more detail below (see Build Variations)
Deployment Scenario | py |
pyjs |
---|---|---|
Linked to sys python | yes | yes |
Embeddded in package | yes | yes |
Embeddded in external | yes | yes |
The py
external has the following c-level methods:
category | method | param(s) | in/out | can change ns |
---|---|---|---|---|
core | import | module | in | yes |
core | eval | expression | out | no |
core | exec | statement | in | yes |
core | execfile | file | in | yes |
extra | assign | var, data | in | yes |
extra | call | var(s), data | out | no |
extra | code | expr or stmt | out? | yes |
extra | anything | expr or stmt | out? | yes |
extra | pipe | var, funcs | out | no |
extra | fold | f, n, args | out | no |
time | sched | ms, fun, args | out | no |
editor | read | file | n/a | no |
editor | load | file | n/a | no |
interobj | scan | n/a | no | |
interobj | send | name, msg, .. | n/a | no |
meta | count | n/a | no |
The pyjs
external implements the following c-level methods:
category | method | param(s) | in/out | can change ns |
---|---|---|---|---|
core | import | module | in | yes |
core | eval | expression | out | no |
core | exec | statement | in | yes |
core | execfile | file | in | yes |
extra | code | expr or stmt | out? | yes |
in-code | eval_to_json | expression | out | no |
In both cases, the code
method allows for import/exec/eval of python code, which can be said to make those 'fit-for-purpose' methods redundant. However, I have retained them since they are stricter in what they allow and further provide a helpful prefix in messages which indicates message intent.
py/js's core features have a one-to-one correspondance to python's very high layer as specified here. In the following, when we refer to object, we refer to instances of either the py
or pyjs
externals. A note of differences between the variations will be provided below.
-
Per-object namespaces. Each object has a unique name (which is provided automatically or can be set by the user), and responds to an
import <module>
message which loads the specified python module in its namespace (essentially aglobals
dictionary). Notably, namespaces can be different for each instance. -
Eval Messages. Responds to an
eval <expression>
message in the left inlet which is evaluated in the context of the namespace.py
objects output results to the left outlet, send a bang from the right outlet upon success or a bang from the middle outlet upon failure.pyjs
objects just return anatomarray
of the results. -
Exec Messages. Responds to an
exec <statement>
message and anexecfile <filepath>
message which executes the statement or the file's code in the object's namespace. Forpy
objects, this produces no output from the left outlet, sends a bang from the right outlet upon success or a bang from the middle outlet upon failure. Forpyjs
objects no output is given.
The extra category of methods makes the py
or pyjs
object play nice with the max/msp ecosystem:
Implemented for py
objects at present:
-
Assign Messages. Responds to an
assign <varname> [x1, x2, ..., xN]
which is equivalent to<varname> = [x1, x2, ..., xN]
in the python namespace. This is a way of creating variables in the object's python namespace using max message syntax. This produces no output from the left outlet, a bang from the right outlet upon success, or a bang from the middle outlet upon failure. -
Call Messages. Responds to a
call <func> arg1 arg2 ... argN
kind of message wherefunc
is a python callable in the py object's namespace. This corresponds to the pythoncallable(*args)
syntax. This makes it easier to call python functions in a max-friendly way. If the callable does not have variable arguments, it will alternatively try to apply the arguments as a list i.e.call func(args)
. Future work will try makecall
correspond to a python generic function call:<callable> [arg1 arg2 ... arg_n] [key1=val1 key2=val2 ... keyN=valN]
. This outputs results to the left outlet, a bang from the right outlet upon success, or a bang from the middle outlet upon failure. -
Pipe message. Like a
call
in reverse, responds to apipe <arg> <f1> <f2> ... <fN>
message. In this sense, a value is piped through a chain of python functions in the objects namespace and returns the output to the left outlet, a bang from the right outlet upon success, or a bang from the middle outlet upon failure.
Implemented for both py
and pyjs
objects:
- Code or Anything Messages. Responds to a
code <expression || statement>
or (anything)<expression || statement>
message. Arbitrary python code (expression or statement) can be used here, because the whole message body is converted to a string, the complexity of the code is only limited by Max's parsing and excaping rules. (EXPERIMENTAL and evolving).
Implemented for pyjs
objects only:
- Evaluate to JSON. Can be used in javascript code only to automatically serialize the results of a python expression as a json string as follows:
evaluate_to_json <expression> -> JSON
.
Implemented for py
objects only:
-
Scan Message. Responds to a
scan
message with arguments. This scans the parent patcher of the object and stores scripting names in the global registry. -
Send Message. Responds to a
send <object-name> <msg> <msg-body>
message. Used to send typed messages to any named object. Evokes ascan
for the patcher's objects if aregistry
of names is empty.
Implemented for py
objects only.
-
Line REPL. The
py
object has two bpatcher linerepls
: one,py_repl_plux.maxpat
which embeds apy
object and another,py_repl.maxpat
which has an outlet to connect to one. The repls include a convenient menu with all of thepy
object's methods and also featurecoll
-based history via arrow-up/arrow-down recall of entries in a session. Acoll
can made to save all commands if required. -
Multiedit REPL. Another bpatcher,
py_multiedit.maxpat
, combines atextedit
object for writing multiliine python code to be executed in the respectivepy
external's namespace, and a simple line repl strictly for evaluating objects in the namespace. -
External Editor Filewatcher.
py_extedit.maxpat
is a bpatcher which wraps thefilewatcher
object and opens a watched file in an external editor. If the file is saved by the editor, it will be sent out as text via the outlet and can be received, for example, by thepy
object's inlet, to enable a kind of load-on-save workflow. -
Code Editor. Double-clicking on the
py
object opens a code-editor. This is populated by aread
message which reads a file into the editor and saves the filepath to the external's attribute. Aload
message alsoreads
the file followed byexecfile
. Saving the text in the editor uses the attribute filepath and execs the saved code to the object's namespace. -
Experimental Remote Console. A method (due to Iain Duncan) of sending code to the
py
node viaudp
has been implemented and allows for send-from-editor and send-from-interactive-console capabilities. The clients are still in their infancy, but this method looks promising since you get syntax highlighting, syntax checking, and other features. It assumes you want to treat yourpy
nodes as remotely accessibleserver/interpreters-in-max
.
zedit: [python interpreter / web server] <-> [web-editor / web-console]
zedit
: a python3 external with an embedded web server.zedit
is a python3-enabled external by virtue of using themamba
single-header library and also embeds the mongoose embedded webserver. On the frontend, it uses modern javascript, jquery-terminal and the widely-used code-mirror web text editor widget to create a web-editor / web-console which can be accessed from a browser and which communicates via the mongoose webserver with the underlying python interpreter.
For pyjs
objects, code editing is already provided by the js Max object.
A subset of the Max c-api is wrapped by the cython-based api
module (api.pyx
). Prior to compilation it is converted to c and then compiled into the external. This exposes a Python builtin module called api
to all python code running on py
objects.
The api
module includes functions and cython extension classes which make it relatively easy to call Max c-api methods from python. This is without doubt the most powerful feature of the py
external.
As of this writing the following extension classes which wrap their corresponding Max datastructures are included in the api
module: Atom
, AtomArray
, Table
, Buffer
, Dictionary
, Database
, Linklist
, Binbuf
, Hashtab
and Patcher
.
In addition, a cython extension class, PyExternal
, gives python code access to the c-based py
external's data and methods.
To give a sense of the level of integration which is possible as a result of this module, the following example demonstrates how numpy
and scipy.signal
can be used to read and write to and from a live Max buffer~
object using the api
module's Buffer
extension class:
import api
import numpy as np
from scipy import signal
def get_buffer_samples(name: str, sample_file: str) -> np.array:
buf = api.create_buffer(name, sample_file)
xs = np.array(buf.get_samples())
assert len(xs) == buf.n_samples
api.post(f"get {n_samples} samples from buffer {name}")
return xs
def set_buffer_samples(name: str, duration_ms: int):
buf = api.create_empty_buffer(name, duration_ms)
t = np.linspace(0, 1, buf.n_samples, endpoint=False, dtype=np.float64)
xs = signal.sawtooth(2 * np.pi * 5 * t)
buf.set_samples(xs)
api.post(f"set {buf.n_samples} samples to buffer {name}")
See the examples/tests
folder and the patchers/tests
folder for more examples.
As mentioned previously, the py-js builder
subproject can be used to build fit-for-purpose python variants for python3 externals. In addition, it can also package, sign, notarize and deploy the same externals for distribution.
These features are implemented in py-js/source/project/py/builder/packaging.py
and are exposed via two interfaces:
$ python3 -m builder package --help
usage: builder package [-h] [-v VARIANT] [-d] [-k KEYCHAIN_PROFILE]
[-i DEV_ID]
...
options:
-h, --help show this help message and exit
-v VARIANT, --variant VARIANT
build variant name
-d, --dry-run run without actual changes.
-k KEYCHAIN_PROFILE, --keychain-profile KEYCHAIN_PROFILE
Keychain Profile
-i DEV_ID, --dev-id DEV_ID
Developer ID
package subcommands:
package, sign and release external
additional help
collect_dmg collect dmg
dist create project distribution folder
dmg package distribution folder as .dmg
notarize_dmg notarize dmg
sign sign all required folders recursively
sign_dmg sign dmg
staple_dmg staple dmg
Since the Makefile
frontend basically just calls the builder
interface in a simplified way, we will use it to explain the basic sequential packaging steps.
-
Recursively sign all externals in the
external folder
and/or binaries in thesupport
foldermake sign
-
Gather all project resources into a distribution folder and then convert it into a
.dmg
make dmg
-
Sign the DMG
make sign-dmg
-
Notarize the DMG (send it to Apple for validation and notarization)
make notarize-dmg
-
Staple a valid notarization ticket to the DMG
make staple-dmg
-
Zip the DMG and collect into in the
$HOME/Downloads/PY-JS
foldermake collect-dmg
To do all of the above in one step:
make release
Note that it is important to sign externals (this is done by Xcode automatically) if you want to to distribute to others (or in the case of Apple Silicon, even use yourself). If the externals are signed, then you can proceed to the notarization step if you have an Apple Developer License (100 USD/year) or, alternatively, you can ask users to remove the product's quarantine state or let Max do this automatically on opening the external.
To complete the notarization process, an Apple Developer Account and an app-specific password are required.
-
Create local credentials based on your apple developer id and app-specific password
xcrun notarytool store-credentials "<keychain-profile-name>" --apple-id "<apple-id>" --team-id <developer-team-id> --password "<app-specific-password>"
-
Export
DEV_ID
andKEYCHAIN_PROFILE
environment variables:export DEV_ID="<first> <lastname>" export KEYCHAIN_PROFILE="<name-of-credentials>"
-
Run the whole process (i.e. steps 1-6) with one command:
make release
There are a number of Github actions in the project which basically automate the testing, packaging, and possibly the notariztion steps described above.
-
The externals in this project have been mostly developed on MacOS and have not yet been extensively tested on Windows.
-
Despite their relative maturity, the
py
andpyjs
objects are still only v0.2.x and still need further unit/functional/integration/field testing! -
As of this writing, the
api
module, does not (like apparently all 3rd party python c-extensions) unload properly between patches and requires a restart of Max to work after you close the first patch which uses it. Unfortunately, this is a known bug in python which is being worked on and may be fixed in future versions (python 3.13 perhaps?). -
Numpy
, the popular python numerical analysis package, falls in the above category. As of python 3.9.x, it thankfully doesn't crash but gives the following error:
[py __main__] import numpy: SystemError('Objects/structseq.c:401: bad argument to internal function')
This just means that the user opened a patch with a py-js
external that imports numpy
, then closed the patch and (in the same Max session) re-opened it, or created a new patch importing numpy
again.
To fix it, just restart Max and use it normally in your patch. Treat each patch as a session and restart Max after each session. It's a pain, but unfortunately a limitation of current python c-extensions.
-
core
features relying on pure python code are supposed to be the most stable, and should not crash under most circumstances,extra
features are less stable since they are more experimental, etc.. -
The
api
module is the most experimental and evolving part of this project, and is completely optional. If you don't want to use it, don't import it or don't use an external which provides it.
As mentioned earlier, as of this writing this project uses a combination of a Makefile
in the project root, a basic cmake
build option and a custom python build system, builder
, which resides in the py-js/source/py/builder
package. The Makefile
is a kind of 'frontend' to the more complex python build system. The latter can be used directly of course. A view into its many options can be obtained by typing the following:
cd py-js/source/py
python3 -m builder --help
builder
was developed to handle the more complex case of downloading the source code of python (from python.org) and also its dependencies from their respective sites and then building custom python binaries with which to reliably compile python3 externals which are portable, relocatable, self-contained, small-in-size, and usable in Max Packages and Standalones.
One of the objectives of this project is to cater to a number of build variations. As of this writing, the following table gives an overview of the different builds and their differences:
There is generally tradeoff of size vs. portability:
build command | format | size_mb | deploy_as | pip | portable | numpy |
---|---|---|---|---|---|---|
make | framework | 0.3 | external | yes [1] | no | yes |
make homebrew-ext | hybrid [3] | 13.6 | external | no | yes | yes |
make homebrew-pkg | hybrid [3] | 13.9 | package | yes | yes | yes |
make static-ext | static | 9.0 | external | no | yes | no [2] |
make shared-ext | shared | 15.7 | external | no | yes | yes |
make shared-pkg | shared | 18.7 | package | yes | no [4] | yes |
make framework-ext | framework | 16.8 | external | no | yes | yes |
make framework-pkg | framework | 16.8 | package | yes | yes | yes |
[1] has automatic access to your system python's site-packages
[2] current static external implementation does not work with numpy due to symbol access issues.
[3] hybrid means that the source system was a framework
and the destination system is shared
.
[4] the shared-pkg variant does not build a compliant 'Framework-type' bundle and hence cannot be notarized.
-
pip: the build allows or provides for pip installation
-
portable: the externals can be deployed as portable packages or standalones
-
numpy: numpy compatibility
The Max package format is a great way to move a bunch of related patches and externals around. This format also makes a lot of sense for py-js
, giving a number of advantages over other alternatives:
-
Portable: Relocatable, you can move it around and it still works.
-
Extendable: Can include a full fit-for-purpse python3 installation in the
support
directory with its own site-packages. Packages can bepip
installed and all of thesite-packages
is automatically made available to the thin 'client' python3 externals in the package'sexternals
folder. -
Size-efficient, since you don't need to duplicate functionality in each external
-
Standalone installable: Recent changes in Max have allowed for this to work in standalones. Just create your standalone application from a patcher which which includes the
py
andpyjs
objects. Once it is built into a<STANDALONE>
then copy the whole aforementionedpy
package to<STANDALONE>/Contents/Resources/C74/packages
and delete the redundantpy.mxo
in<STANDALONE>/Contents/Resources/C74/externals
since it already exists in the just-copied package. -
Better for codesigning / notarizing scenarios since Packages are not sealed bundles like externals.
On the other hand, sometimes you just want an external which embeds a python distribution and custom extensions and code:
-
Portable: Relocatable, you can move it around and it still works.
-
Extendable: Can include new pure python code and be provided with new additionas to
sys.path
-
Size-efficient and fit-for-purpose
-
Standalone installable. Easiest to install in standalones
-
Can be codesigned and notarized relatively easily. [1]
[1] If you want to codesign and notarize it for use in your standalone or package, the codesigning / notarization script and related entitlements file can be found in the source/py/scripts folder.
relocatable-python is Greg Neagle's excellent tool for building standalone relocatable Python.framework bundles.
It works so well, that its been included in the builder
application as an external (embedded dependency).
It can be seen in the relocatable-pkg
make option which will download a nice default Python.framework
to the support
directory used for compiled both py
and pyjs
externals:
make relocatable-pkg
More options are available if you use the builder
package directly:
$ python3 -m builder pyjs relocatable_pkg --help
usage: __main__.py pyjs relocatable_pkg [-h] [--destination DESTINATION]
[--baseurl BASEURL]
[--os-version OS_VERSION]
[--python-version PYTHON_VERSION]
[--pip-requirements PIP_REQUIREMENTS]
[--pip-modules PIP_MODULES]
[--no-unsign] [--upgrade-pip]
[--without-pip] [--release] [-b] [-i]
[--dump]
optional arguments:
-h, --help show this help message and exit
--destination DESTINATION
Directory destination for the Python.framework
--baseurl BASEURL Override the base URL used to download the framework.
--os-version OS_VERSION
Override the macOS version of the downloaded pkg.
Current supported versions are "10.6", "10.9", and
"11". Not all Python version and macOS version
combinations are valid.
--python-version PYTHON_VERSION
Override the version of the Python framework to be
downloaded. See available versions at
https://www.python.org/downloads/mac-osx/
--pip-requirements PIP_REQUIREMENTS
Path to a pip freeze requirements.txt file that
describes extra Python modules to be installed. If not
provided, no modules will be installed.
--pip-modules PIP_MODULES
list of extra Python modules to be installed.
--no-unsign Do not unsign binaries and libraries after they are
relocatablized.
--upgrade-pip Upgrade pip prior to installing extra python modules.
--without-pip Do not install pip.
--release set configuration to release
-b, --build build python
-i, --install install python to build/lib
--dump dump project and product vars
If you are developing the package in $HOME/Documents/Max 8/Packages/py
and you have your iCloud drive on for Documents, you will find that make
or xcodebuild
will reliably fail with 1 error during development, a codesigning error that is due to icloud sync creating detritus in the dev folder. This can be mostly ignored (unless your only focus is codesigning the external).
The solution is to move the external project folder to folder that's not synced-with-icloud (such as $HOME/Downloads
for example) and then run xattr -cr .
in the project directory to remove the detritus (which ironically Apple's system is itself creating) and then it should succeed (provided you have your Info.plist
and bundle id
correctly specified). Then just symlink the folder to $HOME/Documents/Max 8/Packages/
to prevent this from recurring.
I've tried this several times and and it works (for "sign to run locally" case and for the "Development" case).
The coding style for this project can be applied automatically during the build process with clang-format
. On OS X, you can easily install this using brew:
brew install clang-format
The style used in this project is specified in the .clang-format
file.
-
relocatable-python: A tool for building standalone relocatable Python.framework bundles. (used in this project)
-
python-build-standalone: Produce redistributable builds of Python. (Interesting but not used in this project)
-
Python Apple Support: A meta-package for building a version of Python that can be embedded into a macOS, iOS, tvOS or watchOS project. (directly inspired static linking approach)
-
python-cmake-buildsystem: A cmake buildsystem for compiling Python. Not currently used in this project, but may be used in the future.
-
py2max : using python3 with Max in an offline capacity to generate max patches.
-
maxutils : scripts and utilities to help with codesigning and notarization of Max standalones and externals.
-
pocketpy: C++17 header-only Python interpreter for game engines.
-
micropython: a lean and efficient Python implementation for microcontrollers and constrained systems
I was motivated to start this project because I found myself recurrently wanting to use some python libraries or functions in Max.
Looking around for a python max external I found the following:
-
Thomas Grill's py/pyext โ Python scripting objects for Pure Data and Max is the most mature Max/Python implementation and when I was starting this project, it seemed very promising but then I read that the 'available Max port is not actively maintained.' I also noted that it was written in C++ and that it needed an additional c++ flext layer to compile. I was further dissuaded from diving in as it supported, at the time, only python 2 which seemed difficult to swallow considering Python2 is basically not developed anymore. Ironically, this project has become more active recently, and I finally was persuaded to go back and try to compile it and finally got it running. I found it to be extremely technically impressive work, but it had probably suffered from the burden of having to maintain several moving dependencies (puredata, max, python, flext, c++). The complexity probably put off some possible contributors which would have made the maintenance of the project easier for Thomas. In any case, it's an awesome project and it would be great if this project could somehow help py/ext in some way or the other.
-
max-py -- Embedding Python 2 / 3 in MaxMSP with
pybind11
. This looks like a reasonable effort, but only 9 commits and no further commits for 2 years as of this writing. -
nt.python_for_max -- Basic implementation of python in max using a fork of Graham Wakefield's old c++ interface. Hasn't really been touched in 3 years.
-
net.loadbang.jython -- A jython implementation for Max which uses the MXJ java interface. It's looks like a solid effort using Jython 2.7 but the last commit was in 2015.
Around the time of the beginning of my first covid-19 lockdown, I stumbled upon Iain Duncan's Scheme for Max project, and I was quite inspired by his efforts and approach to embed a scheme implementation into a Max external.
So it was decided, during a period with less distractions than usual, to try to make a minimal python3 external, learn the max sdk, the python c-api, and also how to write more than a few lines of c that didn't crash.
It's been an education and I have come to understand precisely a quote I remember somewhere about the c language: that it's "like a scalpel". I now understand this to mean that in skilled hands it can do wonders, otherwise you almost always end up killing the patient.
Thanks to Luigi Castelli for his help with Max/MSP questions, to Stefan Behnel for his help with Cython questions, and to Iain Duncan for providing the initial inspiration and for saving me time with some great implementation ideas.
Thanks to Greg Neagle for zeroing in on the relocatability problem and sharing his elegant solution for Python frameworks via his relocatable-python project on Github.