This provides the FFI backend for the couchbase module. This project is mainly of use to those who wish to use the couchbase module with pypy. Normal CPython extensions do not work very well on pypy, and thus the recommended way of interfacing via pypy in a C library is to use the cffi module.

See the _1.x_ branch for the version of this module that uses the 1.x client and API (though ffi module in that branch is no longer maintained)

You will need the cffi Python module (the newer, the better), the libcouchbase development files, and the actual couchbase module (see its instructions for how to build it).

Each version of the CFFI library has a strict dependency on an exact version of the couchbase Python client, due to the tight coupling of internals.

The current version depends on:

• libcouchbase > 2.4.7
• couchbase = 2.0.0-beta2 (or 2.0.0b2)

Once you've built all that, you must load this module first before any others; thus:

import couchbase_ffi
from couchbase.bucket import Bucket
# ...

This module will inject itself as the couchbase._libcouchbase module, allowing you to use anything that depends on the C extension normally.

This module does not monkey-patch the couchbase Bucket class, but rather the underlying C internals; therefore the ability of a drop-in increases.

The first time you use the module, the couchbase_ffi._cinit module will attempt to generate the appropriate stubs for the library. Regeneration may also take place if you have upgraded your libcouchbase installation.

Because regenerating the header might involve some hacks in dynamically patching the header file to make it acceptable to pycparser, it may fail at times.

Ennvironment variables affecting how the headers are built:

• PYCBC_CFFI_REGENERATE forces the header to be rebuilt
• PYCBC_CFFI_PREFIX sets the installation prefix for libcouchbase

Everything the normal C extension supports. This has been tested using the normal couchbase.tests module and the twisted tests.

I may have missed something here and there, but I can't think of anything that isn't supported.

Better tests.

Here are some examples using the bench.py script in the main Couchbase Python distribution. These were all executed against a locally running memcached like so

Example:

./examples/txbench.py -U memcached://localhost -t 1 -C 1
#....

Here are the results:

For twisted:

The T and C values show how many total clients were spawned, and how many independent logical sequences (threads) of operations were being performed.