JGIT SSH commit signing / NetBinox removal / JNA as OSGI bundle#9483
JGIT SSH commit signing / NetBinox removal / JNA as OSGI bundle#9483matthiasblaesing wants to merge 6 commits into
Conversation
matthiasblaesing
added
DO NOT squash
git
|
This is an alternative approach to #9472 |
Netbinox is a patched version of Equinox (the Eclipse OSGI runtime conainer). The patch is bases on a Equinox version from 2014. The patch was evolved to allow "Import-Package" declaration for JDK packages if they are declared optional. This works as long as the imports are declared optional. This is not the case for a growing number of libraries. NetBeans already carries an up-to-date OSGI container (Apache Felix), so use that and drop the custom dependency. In the wake of this the following modules were removed too: - o.eclipse.core.contenttype - o.eclipse.core.jobs - o.eclipse.core.net - o.eclipse.core.runtime - o.eclipse.core.runtime.compatibility.auth - o.eclipse.equinox.app - o.eclipse.equinox.common - o.eclipse.equinox.preferences - o.eclipse.equinox.registry - o.eclipse.equinox.security It is assumed that equinox was added to be able to run eclise modules, which in turn were required for Mylin (Bugtracking integration). The latter was dropped, so the dependencies should be removed to reduce the dependency footprint.
…m -> o.n.libs.jna.platform Netigso (the NetBeans OSGI Integration layer) makes NetBeans modules available to the OSGI environment. For JNA this works as long as the dependent modules use the "Import-Package" header. It fails if the JNA bundle is imported via "Require-Bundle". The reason for this is the mismatch between the upstream bundle name and the NetBeans module name. The alias/shim modules reexport the JNA modules with the upstream bundle name.
matthiasblaesing
force-pushed
the
jgit-signing-alt
branch
from
fa52cad to
b4757fc
Compare
lkishalmi
left a comment
lkishalmi
left a comment
There was a problem hiding this comment.
I read these through. For me it makes sense.
Thank you!
matthiasblaesing
force-pushed
the
jgit-signing-alt
branch
from
b4757fc to
fbe1179
Compare
|
Updated in place: The jgit signing commit was missing the |
matthiasblaesing
force-pushed
the
jgit-signing-alt
branch
2 times, most recently
from
fd7914e to
0e1df63
Compare
|
Updated in place:
Mina explicitly states (https://github.com/apache/mina-sshd/blob/master/docs/dependencies.md#ed25519-java, hightlights added):
|
This adds the required dependencies and switches the jgit ssh integration from JSch to Apache Mina. This adds four new dependencies: - Apache MINA SSH Client implementation (org.apache.sshd:sshd-osgi and org.apache.sshd:sshd-sftp) - JGIT ssh client implementation using Apache MINA (inkl. the ssh-agent integration) Adjustments to the implementation: - The jgit integration for mina does not query the credential provider for a passphrase when opening the identity file, but instead uses a password callback (existing Credentialsprovider was adjusted to tollerate this) - The NetBeans ssh connection settings allow to choose between username/password and key authentication. There is no fallback between the two. The concept of a custom SshSessionFactory was transfered from the jsch implementation and reimplemented on top of the mina integration.
The patching was in place to enable loading commons-lang3 into the NetBinox container (patched equinox). With NetBinox dropped in favor of Apache Felix, this can be now removed.
matthiasblaesing
force-pushed
the
jgit-signing-alt
branch
from
0e1df63 to
88d2b1f
Compare
matthiasblaesing
added
the
do not merge
|
The removal of NetBinox causes problem is feature-on-demand. I'll have to investigate. With the next release approaching I don't plan to merge this for 31. |
|
Please separate the Netbinox removal effort into its own PR.
I don't understand why the Netbinox removal is part of such a Jumbo PR? A major change like that deserves its own PR, I believe! I am not interested in commenting JGit SSH commit signing change, but I'd like to add some notes about Netbinox - which I still consider the major success of NetBeans in the 2010-2015 "Oracle" era.
Is there any other problem with Netbinox except it is essentially unmaintained (just like a majority of NetBeans Platform code)?
It is not just "one of two". It is the faster one. It is also the one necessary for running Mylyn the "OSGi way". It is the base for integration with JDeveloper. Maybe such "cut offs" are necessary given the manpower left, but after enumerating them, I shall probably ask: What would I have to do to create a perception of Netbinox being maintained? |
|
@jtulach for me NetBinox is unmaintained because:
For the "why this giantic PR": Because this is the usecase that was the final straw. The removal makes only sense when you see the context. The equinox is a mylyn requirement does not hold anymore as mylyn on NetBeans is history. Even if it would be, I'd like to see tests on current OSGI runtimes with up-to-date mylyn dependencies that verify that mylyn is equinox only. |
Feature on Demand scans the feature list and uses the URL for the feature to identify the correct feature. URL uses the registered Handler to implement #equals and #hashCode. When run with NetBinox both the layers and the FeatureInfo yield URLs using the JarClassLoader. When run with Felix as OSGI runtime the layers report URLHandlersStreamHandlerProxy. Instead of changing the URL logic, use the URI to find a valid FeatureInfo.
3 participants
The primary motivation of this PR is moving the ssh support for JGIT from JSch to the Apache Mina based implementation. According to the JGIT project support for JSch is deprecated and there are no plans to move from the unmaintained version
com.jcraft:jschtocom.github.mwiede:jsch.The move to the Mina based client is also required to enable SSH signing of commits.
As preparation NetBinox, the one of the two OSGI implementations NetBeans currently offers, is dropped as it is essentially unmaintained.
A second necessary preparation is to make JNA available as an OSGI bundle of the correct name. JNA was available to the OSGI environment via package imports, but the agent integration code of jgit declares a bundle requirement and thus needs the correct name. This is accomplished by providing an alias/shim module that reexports the original packages with the correct name in addition to the NetBeans module name.
Closes: #9471
Closes: #7572