A cython wrapping of the C++ Cap'n Proto library
More thorough docs are available at http://jparyani.github.io/pycapnp/.
pycapnp’s distribution has no requirements beyond a C++11 compatible compiler. GCC 4.8+ or Clang 3.3+ should work fine.
pycapnp has additional development dependencies, including cython and py.test. See requirements.txt for them all.
Building and installation
Install with pip install pycapnp. You can set the CC environment variable to control which compiler is used, ie CC=gcc-4.8 pip install pycapnp.
Or you can clone the repo like so:
git clone https://github.com/jparyani/pycapnp.git pip install --install-option '--force-cython' ./pycapnp
Note: for OSX, if using clang from Xcode 5, you may need to set CFLAGS like so:
CFLAGS='-stdlib=libc++' pip install pycapnp
If you wish to install using the latest upstream C++ Cap’n Proto:
pip install --install-option "--libcapnp-url" --install-option "https://github.com/sandstorm-io/capnproto/archive/master.tar.gz" --install-option "--force-bundled-libcapnp" .
Python 2.6/2.7 are supported as well as Python 3.2+. PyPy 2.1+ is also supported.
One oddity to note is that Text type fields will be treated as byte strings under Python 2, and unicode strings under Python 3. Data fields will always be treated as byte strings.
This project uses git-flow. Essentially, just make sure you do your changes in the develop branch. You can run the tests by installing pytest with pip install pytest, and then run py.test from the test directory.
Building a dumb binary distribution:
python setup.py bdist_dumb
Building a Python wheel distributiion:
python setup.py bdist_wheel
If it fails with an error like clang: error: no such file or directory: 'capnp/lib/capnp.cpp', then you need to cythonize fist. This can be done with:
python setup.py build --force-cython
There is some basic documentation here.
The examples directory has one example that shows off pycapnp quite nicely. Here it is, reproduced:
from __future__ import print_function import os import capnp import addressbook_capnp def writeAddressBook(file): addresses = addressbook_capnp.AddressBook.new_message() people = addresses.init('people', 2) alice = people alice.id = 123 alice.name = 'Alice' alice.email = 'firstname.lastname@example.org' alicePhones = alice.init('phones', 1) alicePhones.number = "555-1212" alicePhones.type = 'mobile' alice.employment.school = "MIT" bob = people bob.id = 456 bob.name = 'Bob' bob.email = 'email@example.com' bobPhones = bob.init('phones', 2) bobPhones.number = "555-4567" bobPhones.type = 'home' bobPhones.number = "555-7654" bobPhones.type = 'work' bob.employment.unemployed = None addresses.write(file) def printAddressBook(file): addresses = addressbook_capnp.AddressBook.read(file) for person in addresses.people: print(person.name, ':', person.email) for phone in person.phones: print(phone.type, ':', phone.number) which = person.employment.which() print(which) if which == 'unemployed': print('unemployed') elif which == 'employer': print('employer:', person.employment.employer) elif which == 'school': print('student at:', person.employment.school) elif which == 'selfEmployed': print('self employed') print() if __name__ == '__main__': f = open('example', 'w') writeAddressBook(f) f = open('example', 'r') printAddressBook(f)
Also, pycapnp has gained RPC features that include pipelining and a promise style API. Refer to the calculator example in the examples directory for a much better demonstration:
import capnp import socket import test_capability_capnp class Server(test_capability_capnp.TestInterface.Server): def __init__(self, val=1): self.val = val def foo(self, i, j, **kwargs): return str(i * 5 + self.val) def server(write_end): server = capnp.TwoPartyServer(write_end, bootstrap=Server(100)) def client(read_end): client = capnp.TwoPartyClient(read_end) cap = client.bootstrap() cap = cap.cast_as(test_capability_capnp.TestInterface) remote = cap.foo(i=5) response = remote.wait() assert response.x == '125' if __name__ == '__main__': read_end, write_end = socket.socketpair(socket.AF_UNIX) # This is a toy example using socketpair. # In real situations, you can use any socket. server(write_end) client(read_end)
- Fixed upload to PyPi (forgot to cythonize)
- Update bundled capnp version to v0.6.0 and fix related problems (thanks to @benmoran)
- Fix memleak with KjException (thanks to @tsh56)
- Bump bundled capnp version to v0.5.3.1
- Make enums hashable (thanks to @madeleine-empirical)
- Rework logic on when to build bundled libcapnp. Fixes cross-compilation (thanks to @benizl)
- Add traversal_limit_in_words and nesting_limit to RPC classes (thanks to @asilversempirical)
- Include class attributes in dir. This allows for code completion of class methods (thanks to @chaoflow )
- Allow setting lists with python tuples (thanks to @chaoflow)
- Fix traversal_limit_in_words and nesting_limit being ignored by from_bytes (thanks to @plesner)
- Fix bug that prevented event loop from actually being lazy initialized
- Fix possible recursive loop in KjException
- Add clear_write_flag method to builder classes
- Make the event loop be lazy initialized
- Add support for segment (de)serialization (thanks to @gcv). See to_segments/from_segments methods.
- Fix response objects not referencing parents correctly
- Add test for large reads
- Fix build problem with Cython v0.24
- Include the changelog in the manifest (should fix install problems if pandoc is present)
- Include the traceback in exceptions
- Make sure to encode to utf-8, not the default encoding (thanks to @novas0x2a)
- Add –libcapnp-url option in installer to allow installing arbitrary libcapnp versions
- Support mmap objects for reading with from_bytes (thanks to @bpiwowar)
- Change read_multiple and read_multiple_packed to copy by default
- Fix mistakenly discarding the file parameter on reads
- Add reraise_kj_exception to the prettyPrint functions. (thanks to @kdienes)
- Fix KjException init (missing wrapper). (thanks to @E8-Storage)
- Add result_type to InterfaceMethodSchema
- Update bundled libcapnp to v0.5.2
- Add warnings for using old restorer methods. You should use bootstrap instead
- Fix warning from PyEventPort
- Handle AnyPointers better as arguments to RPC functions
- Add support for using keyword arguments with a named struct in an RPC
- Add bootstrap method to TwoPartyServer
- Add init method to lists
- Add support for unix sockets in RPC
- Fix a serious bug in TwoPartyServer that was preventing it from working when passed a string address.
- Fix bugs that were exposed by defining KJDEBUG (thanks @davidcarne for finding this)
- Update bundled C++ libcapnp to v0.5.1.2 security release
- Update bundled C++ libcapnp to v0.5.1.1 security release
- Add bootstrap RPC methods
- Fix possible segfault when importing multiple schemas
- Fix possible crash due to bad destructor ordering in MessageReader (by @JohnEmhoff)
- Default to no longer using cython
- Add read_multiple_bytes/read_multiple_bytes_packed methods
- Added Python 3.4 to the travis build matrix
- Bump version for bundled C++ libcapnp to v0.5.1
- Remove installation dependency on cython. We now have no dependencies since libcapnp will automatically build as well.
- Timer class capnp.getTimer()
- pycapnp is now thread-safe and allows an event loop to be run in each thread
- You must destroy and re-create the event loop to get this functionality (see test_threads.py)
- Inheritance now works correctly for interfaces (previously inherited methods were inaccessible from pycapnp)
- Add ability to import modules with dashes or spaces. Use underscores in place of them
- from_bytes with builder=True is no longer zero copy. It never worked correctly, and is much safer now
- Add num_first_segment_words argument wherever message creation can occur
- Allow restoring a null objectId by passing None to restore
- Support ordered dictionary in to_dict
- Add ListSchema class and schemas for native types under capnp.types which completes all the Schemas needed to be wrapped. See test_schema.py for examples using it
- Add automatic build of C++ libcapnp if it’s not detected on the system. Also add flags –force-bundled-libcapnp and –force-system-libcapnp respectively
- Fix build for new 0.21 release of Cython. 0.21 is now the minimum supported version of Cython.
- Fix to_dict not converting enums to strings
- Fix compilation problem with gcc 4.8
- Fix problem with uninitialized unions in _from_dict
- Add accesible version numbers for C++ libcapnp
- Remove onDrained since it was removed upstream
- Replace usage of strings as enum type with custom _DynamicEnum class.
- Also change Struct.which() method to be a property Struct.which and return an enum type (_DynamicEnumField, which behaves much like _DynamicEnum).
- TwoPartyServer.run_forever() now will handle more than 1 simulataneous connection.
- Change exception wrapper to detect and raise AttributeError for field lookup exceptions (Fixes problem in Python3.x __dir__)
- Allow setting of fields with python dicts.
- Remove python 3.2 from travis tests. Python 3.2 still should work fine, but it’s more trouble than it’s worth to write unicode tests that work in both it and Python2.
- Fix problems with null characters in Text/Data fields. Fixes #19
- Initial working version of RPC
- Add get_root_as_any to _MessageReader
- Add capnp.pxd for public declarations of cython classes
- Fix problems compiling with gcc4.7
- Change naming of ReaderOption parameters to be pep8 compliant
- Add ReaderOptions to read/read_packed/from_bytes
- Add defaults flag to capnp-json. Also remove ‘which’ field
- Add capnp-json serializer script. Also fix bugs in from_dict
- Fix build for clang/python3. Also remove -fpermissive
- Add as_builder method to Struct Reader
- Add warning when writing the same message more than once
- First working version of capability interfaces
- Wrap InterfaceSchema
- Fix setting string fields to support all types of strings
- Fix changed API for DynamicObject/ObjectPointer
- Add not having installed the C++ libcapnp library to ‘Common Problems’
- Add shortstr function for use in capnp_test_pycapnp.py
- Add test script for testing with https://github.com/kaos/capnp_test
- Add handling of DynamicObject
- Fix lists of lists or dicts for from_dict
- Fix problem with to_dict
- Add DynamicStructBuilder.tobytes() and .from_bytes()
- Change == on StructSchema to return cbool
- Add Builder and Reader ABCs for each struct type
- Fix handling of empty path ‘’ in load_module
- Add from_dict
- Fix bug in exception handling for which(). Also standardize exceptions.
- Change import hook to require modules to end in ‘_capnp’
- Add import monkey patch function.
- Change naming for functions to conform to PEP 8. Also deprecate old read/write API
- Update preferred method for reading/writing messages from files
- Forgot to change project name in setup.py
- Change all references to old project name (change from capnpc-python-cpp to pycapnp)
- Change DynamicValue.Reader lists to be returned as _DynamicListReader
- Unify setters for DynamicList and DynamicStruct
- Add shortcuts for reading from / writing to files. In Python, it doesn’t make much sense to force people to muck around with MessageReaders and MessageBuilders since everything is landing on the heap anyway. Instead, let’s make it easy: MyType.read[Packed]From(file) reads a file and returns a MyType reader. MyType.newMessage() returns a MyType builder representing the root of a new message. You can call this builder’s write[Packed]To(file) method to write it to a file.
- Store Builders by value rather than allocate them separately on the heap (matches treatment of Readers). v0.3 fixes the bug that made this not work.
- Wrap MessageBuilder::setRoot().
- Add tests based on TestAllTypes from the C++ test.capnp. Fix problems uncovered in capnp.pyx.
- Implement str and repr for struct and list builders. str uses prettyPrint while repr shows the type name and the low-whitespace stringification. Also implement repr for StructSchema, just because why not?
- Change load to use a global SchemaParser. Make structs settable as field
- Add docstrings for new functions and _DynamicResizableListBuilder
- Add initial tests
- Add _capnp for original Cython module. Meant for testing.
- Lowercase schema so it conforms to member naming conventions
- Expose _StructSchema’s raw node
- Add some useful _StructSchema, reader, and builder methods
- Add full orphan functionality. Also, allow special orphan lists
- Finish up adding docstrings to all public classes/methods
- Add a ton of docstrings and add to official docs
- Add DynamicOrphan
- Add intersphinx for linking to python docs
- Add C++ library version check
- Add handling of constants in schemas
- Fix new error with DynamicValue.Builder no longer being copyable
- Fix Void namespace change
- Updated capnp schema to conform with new union rules
- Fix for the removal of DynamicUnion from the C++ API
- Add MANIFEST.in to include README
- Update docs with lines about upgrading setuptools
- Initial commit of docs
- Add querying unnamed enums to structs
- Fix enum interface change for benchmark
- Random formatting cleanup
- Allow import paths in the schema loader
- Add travis CI
- Initial working version