Commits

Show all
Author Commit Message Labels Comments Date
Jim Baker
Fixed typos in README.txt
Tags
v2.5.2b2
Jim Baker
Updated NEWS and README.txt in prep of 2.5.2b2
Oti Humbel
prepare for 2.5.2b2
Jim Baker
Backed out support for nulling out ThreadState support. We need a better solution that doesn't introduce synchronization issues and a backwards breaking API change. See the discussion in #1327.
Jim Baker
Files that were not included in r7115
Oti Humbel
default svn.revision to HEAD this allows rebuilding releases without remembering the exact revision number
Jim Baker
More work on #1327. ThreadStateMapping has a bad implementation of double-checked locking. The aspect of how it interacts with thread locals makes it subtle, so I'm going to make the whole lookup synchronized. Going forward, we should reconsider the use of ThreadLocal to manage the ThreadState mapping. Synchronizing this loses any performance from using ThreadLocal, but we have to deal with the other issues seen here. Also, I think we should expose the registration process more, in the case of usage with thread pools.
Jim Baker
Updated NEWS on bugs that were in fact fixed
Jim Baker
Fixed PyJavaType#setSlice so it properly works with Python objects that are subclasses of Java classes implementing List. Resolves #1397. Thanks Geoffrey French for the patch!
Oti Humbel
remove redundant cast
Jim Baker
Fixed #1503, Java constructors should take keyword arguments. Thanks Geoffrey French!
Jim Baker
To address #1327 - removed use of Guava's FinalizableReferenceQueue and now does a passive cleanup of PySystemState-related data, so that classloaders will hopefully GC under a range of setups. Also de-registers the shutdown hook.
Oti Humbel
fix the java tests by initializing the interpreter in setUp()
Jim Baker
Doc cleanup
Jim Baker
Use Guava's FinalizableReferenceQueue and a new PySystemStateCloser class to manage the closing of those resources that prevent classloader GC (because they have hard refs to classes) after a PySystemState itself is GCed. This refactoring also moved the closing of user-unclosed files associated with a PyFile to the holding PySystemState, instead of having to wait for JVM shutdown. Now the JVM shutdown hook does this closing of resources through the PySystemStateCloser.
javajuneau
Updating swing demos
javajuneau
Updating javaclasses demos
javajuneau
Updating applet demos
javajuneau
Updating applet demos
Jim Baker
Now test that defaultdict's __missing__/default_factory is computed atomically.
Jim Baker
Missing file (test of pep328 support) from r7102
Jim Baker
Fixes incomplete implementation of relative imports for PEP328, which could cause a NPE. Thanks Anselm Kruis!
Jim Baker
Module#addConstants incorrectly set the accessbility of the 'self' constant in the generated module bytecode to be final, but it can in fact be modified outside of the class initializer. Verified the regrtest passes with -Xcomp (except for test_jy_internals, but this looks like standard issues of working with GC in tests). Bumped bytecode magic. Thanks Ian Rogers for your analysis!
Jim Baker
Minor changes to PyLong, PyFloat that were missed by NetBean's accessor refactoring support. Added tests of how our subclasses of our types can now be abused to have treatable as the original type, but mutable. Use such dangerous functionality with care! See #1645 for further discussion of this issue, including rationale. Implemented to help support efforts like processing.py. Thanks Jonathan Feinberg for this patch set.
Jim Baker
Standard refactoring of some core Python types so that their implementation (including subclasses) do not access the underlying backing field, but uses an accessor. More fixes for #1645. In addition, there's support for defaultdict providing atomic construction of values for associated keys, using Guava MapMaker's makeComputingMap. This is a rather cool example of the additional abstraction provided by the refactoring (namely PyDictionary#getMap). Some tests are needed to demo how sweet this really is :) Note that we cannot, given our constructors calling super in a chain, readil…
Jim Baker
Expose the Python interpreter to sys, along with its associated JLine Reader (if available from JLineConsole). This enables a (so far) partial implemnentation of readline semantics suitable for completion support. See #1133.
Jim Baker
Refactored PyBoolean, PyFloat, PyLong so they consistently use getValue instead of accessing the field value directly. Fixes #1645, although we are awaiting more unit tests on how this be used by user types.
Jim Baker
Removed Tools directory, including jythonc. It was not being copied, and even though there's potentially useful code in there, we don't want what are effectively sandbox directories like this cluttering trunk and confusing users on what is being supported. Fixes #1643.
Jim Baker
Update for r7094
Jim Baker
Define the package name for a class dynamically loaded from sys.path. Fixes #1455. Thanks to "stan" for the patch!
  1. Prev
  2. Next