- Aug 12, 2022
-
-
Thomas Watson authored
The fix for double resolution for optional resources introduced a bug that would allow non-payload fragments to resolve even if a hook disabled them from resolution. The fix is to check if the fragment is disabled while resolving non-payload fragments
-
- Aug 11, 2022
-
-
Thomas Watson authored
If the triggers are not mandatory then just resolve all unresolved resources. If triggers are mandatory then do not attempt to resolve all unresolved resources (except for attaching non-payload fragments). The way the ModuleResolver would resolve mandatory triggers followed by trying to resolve optional resources caused issues with the timeout for the second resolution of the optional resources. Instead of fixing the timeout for the second resolution this fix stops doing two resolutions altogether. The main difference is if the triggers are mandatory then the resolver will only attempt to resolve the triggers and any dependencies the trigger tries to pull in during resolution. Before the ModuleResolver would also attempt to resolve all unresolved resources as optional. In the OSGi framework this only impacts APIs (mainly on Bundle) that require the Bundle to be resolved when the API is called. For example, Bundle.start, Bundle.findEntries, Bundle.loadClass.
-
Ed Merks authored
-
- Aug 01, 2022
-
-
Christoph Läubrich authored
RegistryFactory.getRegistry can return null, but the static methods to not check for null. FindSupport uses System.getProperty(PROP_NL) Activator.getContext().getProperty(PROP_NL) both can return null
-
Ed Merks authored
of javax.servlet.jsp than 2.3. The Orbit 2.2.0 version of javax.servlet.jsp has been removed from the target platform by the following PR and the newer version 2.3.3 version is now available, so the org.eclipse.equinox.jsp.jasper should accommodate that new version. https://github.com/eclipse-platform/eclipse.platform.releng.aggregator/pull/412
-
- Jul 26, 2022
-
-
Hannes Wellmann authored
-
Hannes Wellmann authored
-
Hannes Wellmann authored
-
BJ Hargrave authored
We move the concurrent map to be a instance field of ServiceRegistry instead of a static. We want to avoid static mutable state to support multiple framework instances in the same JVM. We also include the DEADLOCK type in the exception. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Hannes Wellmann authored
Reproducer test for: https://github.com/eclipse-equinox/equinox.framework/issues/55 Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Hannes Wellmann authored
Prevent one interrupted check on each call to lock(). From the start of lock() to useLock.tryLock(... everything should be as performant as possible. Afterwards we are usually waiting so it doesn't really matter since we cannot wait faster. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Hannes Wellmann authored
Remove unnecessary elements and dead-lock detection logic into dedicated isDeadLocked() method to make the code more readable. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
BJ Hargrave authored
This is close to behavior of intrinsic locking which does not alter the thread interrupt status. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
BJ Hargrave authored
When throwing the deadlock exception we also need to remove the lock from the awaiting map. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
BJ Hargrave authored
Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Hannes Wellmann authored
The ServiceUse.lock() method returns an AutoCloseable so the caller can use try-with-resources to obtain and then close the lock. This is much closer to intrinsic locking in coding practice. Instead of using a long-ish timeout for deadlock detection, which could be wrong on slow systems, we now use a map to hold the thread to lock parings and examine the lock chain to see if the current thread is in the chain. Signed-off-by: Hannes Wellmann <wellmann.hannes1@gmx.net> Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
BJ Hargrave authored
Numerous comments and asserts needed updating. All uses under lock are run through withLock which has cleaned up the exception handling to properly manage interrupt state. NLS is used for exception messages. Lock subclass toString is cleaned up to avoid string concatenation when we are already using a StringBuilder. Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Alain Picard authored
Signed-off-by: Alain Picard <picard@castortech.com> Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Alain Picard authored
Using a reentrant lock on Service use instead of simple synchronized blocks. Also added quite a bit of logging to help tracing. Signed-off-by: Alain Picard <picard@castortech.com> Signed-off-by: BJ Hargrave <hargrave@us.ibm.com>
-
Andrey Loskutov authored
This reverts change on ServletRegistration from commit a1a9eaf5 because it moved dangerous initialization of org.eclipse.equinox.http.servlet.internal.multipart.MultipartSupportFactoryImpl during iterator.next() call outside of the try/catch block. MultipartSupportFactoryImpl depends on optionally available org.apache.commons.fileupload package, and its init is *expected* to fail with ServiceConfigurationError if the package is not available. Fixes https://github.com/eclipse-equinox/equinox/issues/88
-
- Jul 22, 2022
-
-
Hannes Wellmann authored
Signed-off-by: Hannes Wellmann <wellmann.hannes1@gmx.net>
-
- Jul 20, 2022
-
-
Hannes Wellmann authored
... without "Only if loop variable used" option being activated. And manually remove now unused imports, which the clean-up does not yet detect (a JDT issue is already created). Don't apply clean-up to embedded OSGi and Felix sources.
-
Hannes Wellmann authored
-
Hannes Wellmann authored
Remove '<skipAPIAnalysis>true</skipAPIAnalysis>' properties that are now obsolete since the platform-build detects automatically if the API-check should run or not. For o.e.osgi.util the additional compiler-args are not necessary anymore because this plug-in does not contain any sources(in that process it was forgotten to remove them). Their removal makes the pom trivial so it can be removed to use Tycho-pomless instead. Part of https://github.com/eclipse-equinox/equinox/issues/9
-
Hannes Wellmann authored
-
Hannes Wellmann authored
Use the enhancements of https://github.com/eclipse-platform/eclipse.platform/pull/54 Use addMinimalBundleSet(), addThisBundle() and addBundle(Class) and leverage the fact that o.e.core.tests.harness now pulls all requirements into the test-runtime. Use the same pattern for the additionally required osgi bundles, which happens automatically if addBundle(Class) is used. This also fixes https://github.com/eclipse-equinox/equinox/issues/81
-
- Jul 19, 2022
-
-
Hannes Wellmann authored
-
Hannes Wellmann authored
-
Mickael Istria authored
Change Bundle-SymbolicName
-
- Jul 13, 2022
-
-
Christoph Läubrich authored
-
- Jul 10, 2022
-
-
Hannes Wellmann authored
Manually remove now unused Exception type imports. The issue has been reported to JDT.
-
Hannes Wellmann authored
-
Hannes Wellmann authored
to Equinox-sources (ignore embedded felix and osgi sources)
-
Hannes Wellmann authored
-
- Jul 09, 2022
-
-
Mickael Istria authored
-
- Jul 01, 2022
-
-
Mickael Istria authored
-
- Jun 30, 2022
- Jun 29, 2022
-
-
Александър Куртаков authored
That should avoid having to touch the feature everytime Jetty version is changed. Minor version bumped for easily tracking this change.
-
- Jun 28, 2022
-
-
Александър Куртаков authored
-
- Jun 27, 2022
-
-
Thomas Watson authored
Hack in exports, provides and requires so that bnd will properly produce module-info exports, provides and uses. There may be a way to get BND just to ues our manifest and not try to produce its own to make a module-info. But I've not found out how yet.
-