Changelog¶

18.1.8 - 2023-10-05¶

WARNING - BREAKING CHANGE: Drop support for Python 3.7.

This breaking change should usually require a major and/or minor number update. Since berkeleydb traditional numbering is related to the higher Oracle Berkeley DB supported, I would usually wait until Oracle releases a new version to upgrade my own version and deprecate old Python support at the same time. Given that Oracle has not released a new Oracle Berkeley DB in almost five years, I must break this practice for now.

I am sorry if this update breaks your Python 3.7 environment. In that case, please pin your berkeleydb installation to version 18.1.6, the last Python 3.7 compatible release.

Send me constructive feedback if appropiate.

Progressing the implementation of PEP 489 – Multi-phase extension module initialization: https://peps.python.org/pep-0489/.

Types are now private per sub-interpreter, if your are compiling under Python >= 3.9.

Provide a per sub-interpreter capsule object.

Solve a tiny race condition when importing the module in multiple sub-interpreters at the same time.

Update the “api_version” value of the capsule object.

Solve a “deprecation warning” when using modern “setuptools”.

For testing, we require at least “setuptools” >= 62.1.0 installed on all supported Python versions.

Python 3.12 is officially supported.

18.1.7 - 2023-10-05¶

Yanked version.

18.1.6 - 2023-05-10¶

Initial implementation of PEP 489 – Multi-phase extension module initialization: https://peps.python.org/pep-0489/.

Update “setuptools” built-time dependency to version “>=65.5.0”. A “pip” modern enough will automatically take care of this.

We must be sure we are testing the correct library. Previously we could be testing the installed library instead of development code.

Python 3.12 added to the full test matrix.

Experimental Python 3.12 support. Tested under 3.12.0a7.

18.1.5 - 2022-01-21¶

WARNING - BREAKING CHANGE: Drop support for Python 3.6.

This breaking change should usually require a major and/or minor number update. Since berkeleydb traditional numbering is related to the higher Oracle Berkeley DB supported, I would usually wait until Oracle releases a new version to upgrade my own version and deprecate old Python support at the same time. Given that Oracle has not released a new Oracle Berkeley DB in almost four years, I must break this practice for now.

I am sorry if this update breaks your Python 3.6 environment. In that case, please pin your berkeleydb installation to version 18.1.4, the last Python 3.6 compatible release.

Send me constructive feedback if appropiate.

Python 3.10 support.

Testsuite works now in Python 3.11.0a4.

Python 3.11 added to the full test matrix.

Python 3.11 deprecates the ancient but undocumented method unittest.makeSuite() and it will be deleted in Python 3.13. We migrate the tests to unittest.TestLoader.loadTestsFromTestCase().

Experimental Python 3.11 support. Tested in 3.11.0a4.

18.1.4 - 2021-05-19¶

If your “pip” is modern enough, “setuptools” is automatically added as a built-time dependency.

If not, you MUST install “setuptools” package first.

18.1.3 - 2021-05-19¶

Docs in https://docs.jcea.es/berkeleydb/.

make publish build and publish the documentation online.

Python 3.10 deprecated distutils. setuptools is now an installation dependency.

make dist will generate the HTML documentation and will include it in the released package. You can unpack the package to read the docs.

Do not install tests anymore when doing pip install, although the tests are included in the package. You can unpack the package to study the tests, maybe in order to learn about how to use advanced Oracle Berkeley DB features.

This change had an unexpected ripple effect in all code. Hopefully for the better.

Python 3.10 couldn’t find build directory.

Python 3.10.0a2 test suite compatibility.

Python 3.10 added to the full test matrix.

After Python 3.7, threads are always available. Take them for granted, even in Python 3.6.

In the same direction, now some libraries are always available: pathlib, warnings, queue, gc.

Support DB.get_lk_exclusive() and DB.set_lk_exclusive() if you are linking against Oracle Berkeley DB 5.3 or newer.

WARNING - BREAKING CHANGE: The record number in the tuple returned by DB.consume() is now a number instead of a binary key.

WARNING - BREAKING CHANGE: The record number in the tuple returned by DB.consume_wait() is now a number instead of a binary key.

DB.consume() and DB.consume_wait() now can request partial records.

DB.get() and DB.pget() could misunderstand flags.

If you are using Oracle Berkeley DB 5.3 or newer, you have these new flags: DB_BACKUP_CLEAN, DB_BACKUP_FILES, DB_BACKUP_NO_LOGS, DB_BACKUP_SINGLE_DIR and DB_BACKUP_UPDATE, DB_BACKUP_WRITE_DIRECT, DB_BACKUP_READ_COUNT, DB_BACKUP_READ_SLEEP, DB_BACKUP_SIZE.

If you are using Oracle Berkeley DB 18.1 or newer, you have these new flags: DB_BACKUP_DEEP_COPY.

DBEnv.backup(), DBEnv.dbbackup() DB.get_backup_config() and DB.set_backup_config() available if you are using Oracle Berkeley DB 5.3 or newer. These methods allow you to do hot backups without needing to follow a careful procedure, and they can be incremental.

Changelog moved to Sphinx documentation.

18.1.2 - 2020-12-07¶

Releases 18.1.0 and 18.1.1 were incomplete. Thanks to Mihai.i for reporting.

Export exception DBMetaChksumFail (from error DB_META_CHKSUM_FAIL) if running Oracle Berkeley DB version 6.2 or newer.

Support Heap access method if you are linking against Oracle Berkeley DB 5.3 or newer.

DB.put() can add new records or overwrite old ones in Heap access method.

DB.append() was extended to support Heap access method.

DB.cursor() was extended to support Heap access method.

Implement, test and document DB.get_heapsize(), DB.set_heapsize(), DB.get_heap_regionsize() and DB.set_heap_regionsize().

Export exception DBHeapFull (from error DB_HEAP_FULL).

DB.stats() provides stats for Heap access method.

WARNING - BREAKING CHANGE: Add dbtype member in DBObject object in the C API. Increase C API version. This change has ripple effect in the code.

WARNING - BREAKING CHANGE: primaryDBType member in DBObject object in the C API is now type DBTYPE. Increase C API version. This change has ripple effect in the code.

Now DB.get_type() can be called anytime and it doesn’t raise an exception if called before the database is open. If the database type is not known, DB_UNKNOWN is returned. This is a deviation from the Oracle Berkeley DB C API.

WARNING - BREAKING CHANGE: DB.type() method is dropped. It was never documented. Use DB.get_type().

DB.stats() returns new keys in the dictionary:

Hash, Btree and Recno access methods: Added metaflags (always) and ext_files (if linked against Oracle Berkeley DB 6.2 or newer).

Queue access method: Added metaflags (always).

18.1.1 - 2020-12-01¶

If you try to install this library in an unsupported Python environment, instruct the user about how to install legacy bsddb3 library.

Expose DBSite object in the C API. Increase C API version.

WARNING - BREAKING CHANGE: Ancient release 4.2.8 added weakref support to all bsddb.db objects, but from now on this feature requires at least Python 3.9 because I have migrated from static types to heap types. Let me know if this is a problem for you. I could, for example, keep the old types in Python < 3.9, if needed.

Details:

Py_tp_dictoffset / Py_tp_finalize are unsettable in stable API https://bugs.python.org/issue38140

bpo-38140: Make dict and weakref offsets opaque for C heap types (#16076) https://github.com/python/cpython/commit/3368f3c6ae4140a0883e19350e672fd09c9db616

_iter_mixin and _DBWithCursor classes have been rewritten to avoid the need of getting a weak reference to DBCursor objects, since now it is problematic if Python < 3.9.

Wai Keen Woon and Nik Adam sent some weeks ago a patch to solve a problem with DB.verify() always succeeding. Refactoring in that area in 18.1.0 made that patch unneeded, but I added the test case provided to the test suite.

DBEnv.cdsgroup_begin() implemented.

DBTxn.set_priority() and DBTxn.get_priority() implemented. You need to link this library against Oracle Berkeley DB >= 5.3.

DBEnv.set_lk_max() was deprecated and deleted long time ago. Time to delete it from documentation too.

WARNING - BREAKING CHANGE: DB.compact() used to return a number, but now it returns a dictionary. If you need access to the old return value, you can do DB.compact()['pages_truncated'].

DB.compact() has been supported txn parameter for a long time, but it was not documented.

The dictionary returned by DB.compact() has an end entry marking the database key/page number where the compaction stopped. You could use it to do partial/incremental database compaction.

Add an optional parameter to DBEnv.log_flush().

You can override the directory where the tests are run with TMPDIR environment variable. If that environment variable is not defined, test will run in /tmp/ram/ if exists and in /tmp if /tmp/ram/ doesn’t exists or it is not a directory. The idea is that /tmp/ram/ is a ramdisk and the test will run faster.

18.1.0 - 2020-11-12¶

bsddb name is reserved in PYPI, so we rename the project to berkeleydb. This has been a long trip: http://mailman.jcea.es/pipermail/pybsddb/2008-March/000019.html

18.1.0-pre¶

Support Oracle Berkeley DB 18.1.x.

Drop support for Oracle Berkeley DB 4.7, 5.1 and 6.1.

Drop support for Python 2.6, 2.7, 3.3, 3.4 and 3.5.

The library name is migrated from bsddb3 to bsddb. Reasons:

In the old days, bsddb module was integrated with Python < 3 . The release rate of new Python interpreters was slow, so bsddb was also distributed as an external package for faster deployment of improvements and support of new Oracle Berkeley DB releases. In order to be able to install a new version of this package without conflicting with the internal python bsddb, a new package name was required. At the time, the chosen name was bsddb3 because it was the major release version of the supported Oracle Berkeley DB library.

After Oracle released Berkeley DB major versions 4, 5, 6 and 18, bsddb3 name was retained for compatibility, although it didn’t make sense anymore.

bsddb3 seems to refer to the Python 3 version of bsddb. This was never the case, and that was confusing. Even more now that legacy bsddb3 is the Python 2/3 codebase and the new bsddb is Python 3 only.

Since from now on this library is Python 3 only, I would hate that Python 2 users upgrading their Berkeley DB libraries would render their installation unable to run. In order to avoid that, a new name for the package is a good idea.

I decided to go back to bsddb, since Python 2.7 is/should be dead.

If your are running Python 3, please update your code to use bsddb instead of bsddb3.

The old practice was to do:

import bsddb3 as bsddb

Now you can change that to:

import bsddb

This library was usually know as bsddb, bsddb3 or pybsddb. From now on, it is bsddb everywhere.

Testsuite driver migrated to Python 3.

Since Oracle Berkeley DB 4.7 is not supported anymore, ancient method DBEnv.set_rpc_server() is not available anymore.

If you try to install this package on Python 2, an appropriate error is raised and directions are provided.

Remove dead code for unsupported Python releases.

Remove dead code for unsupported Oracle Berkeley DB releases.

WARNING: Now ALL keys and values must be bytes (or ints when appropriate). Previous releases did mostly transparent encoding. This is not the case anymore. All needed encoding must be explicit in your code, both when reading and when writing to the database.

In previous releases, database cursors were iterable under Python 3, but not under Python 2. For this release, database cursors are not iterable anymore. This will be improved in a future release.

In previous releases, log cursors were iterable under Python 3, but not under Python 2. For this release, log cursors are not iterable anymore. This will be improved in a future release.

Support for DB_REPMGR_CONF_DISABLE_SSL flag in DB_ENV.rep_set_config().

WARNING: In Oracle Berkeley DB 18.1 and up, Replication Manager uses SSL by default.

This configuration is currently unsupported.

If you use Oracle Berkeley DB 18.1 and up and Replication Manager, you MUST configure the DB environment to not use SSL. You must do

DB_ENV.rep_set_config(db.DB_REPMGR_CONF_DISABLE_SSL, 1)

in your code.

This limitation will be overcomed in a future release of this project.

open() methods allow path-like objects.

DBEnv.open() accepts keyword arguments.

DBEnv.open() allows no homedir and a homedir of None.

DB.set_re_source() uses local filename encoding.

DB.set_re_source() accepts path-like objects if using Python 3.6 or up.

DB.verify() was doing nothing at all. Now actually do the job.

DB.verify() accepts path-like objects for filename and outfile if using Python 3.6 or up.

DB.upgrade() accepts path-like objects if using Python 3.6 or up.

DB.remove() accepts path-like objects if using Python 3.6 or up.

DB.remove() could leak objects.

DB.rename() accepts path-like objects if using Python 3.6 or up.

DB.rename() correctly invalidates the DB handle.

DB.get_re_source() returns unicode objects with the local filename encoding.

DB_ENV.fileid_reset() accepts path-like objects if using Python 3.6 or up.

DB_ENV.log_file() correctly encode the filename according to the system FS encoding.

DB_ENV.log_archive() correctly encode the filenames according to the system FS encoding.

DB_ENV.lsn_reset() accepts path-like objects if using Python 3.6 or up.

DB_ENV.remove() accepts path-like objects if using Python 3.6 or up.

DB_ENV.remove() used to leave the DBENV handle in an unstable state.

DB_ENV.dbrename() accepts path-like objects for filename and newname if using Python 3.6 or up.

DB_ENV.dbremove() accepts path-like objects if using Python 3.6 or up.

DB_ENV.set_lg_dir() uses local filename encoding.

DB_ENV.set_lg_dir() accepts path-like objects if using Python 3.6 or up.

DB_ENV.get_lg_dir() returns unicode objects with the local filename encoding.

DB_ENV.set_tmp_dir() uses local filename encoding.

DB_ENV.set_tmp_dir() accepts path-like objects if using Python 3.6 or up.

DB_ENV.get_tmp_dir() returns unicode objects with the local filename encoding.

DB_ENV.set_data_dir() uses local filename encoding.

DB_ENV.set_data_dir() accepts path-like objects if using Python 3.6 or up.

DB_ENV.get_data_dirs() returns a tuple of unicode objects encoded with the local filename encoding.

DB_ENV.log_prinf() requires a bytes object not containing ‘0’.

The DB_ENV.lock_get() name can not be None.

DB_ENV.set_re_pad() param must be bytes or integer.

DB_ENV.get_re_pad() returns bytes.

DB_ENV.set_re_delim() param must be bytes or integer.

DB_ENV.get_re_delim() returns bytes.

In the C code we don’t need statichere neither staticforward workarounds anymore.

db.DB* objects are created via the native classes, not via factories anymore.

Drop support for dbtables. If you need it back, let me know.

In Python 3.9, find_unused_port has been moved to test.support.socket_helper. Reported by Michał Górny.

If we use set_get_returns_none() in the environment, the value could not be correctly inherited by the child databases. Reported by Patrick Laimbock and modern GCC warnings.

Do not leak test files and directories.