Feature/556 osgi uninstall

[PengZheng]

To fix #556, #557, part of #561, and #563, this PR introduces:

• Per bundle "lifecycle" read-write lock. To change bundle state, the write lock must be held.
• celix_framework_useActiveBundles, which holds the per bundle read lock when calling the provided use callback.
  The bundle state is guaranteed to be active during the callback. For other cases like celix_bundleContext_getBundleSymbolicName, such strong guarantee is not needed. Therefore, celix_framework_useBundles is still there.
  • Another interesting use case is observed in pubsub_topology_manager, a service/tracker owner's bundle object, which is tracked, is always safe to access. Calling celix_framework_useActiveBundles will lead to ABBA deadlock while calling celix_framework_useBundles will not. The test bundle (tst_activator.c) registers several subscribers synchronously, waiting the second registration to return (i.e. waiting for the event loop). The event loop is trying to add the second subscriber into topicReceivers, waiting for manager->topicReceivers.mutex. The owner of manager->topicReceivers.mutex is calling celix_framework_useActiveBundles, trying to hold test bundle's lifecycle read-lock. Unfortunately, the write lock is held before calling the test bundle's activator start.
    • Modifying the test bundle to do asynchronous registration works, but that will limit our users' freedom to perform synchronous registration.
• uninstall of semantics similar to OSGi specs. It removes the bundle from the bundle cache and returns the bundle id to the framework.
• unload operation, which is similar to uninstall but leaves the bundle in cache, so that next time reloaded from the cache, the bundle id does not change. To avoid any confusion, it is used only for testing purpose and is not exposed.
• A coarse install lock, which essentially serialize install/uninstall/unload operations. Thus bundle id namespace and the bundle cache are kept in sync.
• update implemented as unload plus install, avoid any bundle state inconsistency (Bundle state inconsistency in case of update failure #563) in case of update failure.
• Some build system cleanup, including:
  • Bring BUILD_RPATH back as a convenience feature.
  • A better workaround of the issue of private linking of transitive dependencies. (private linking of transitive dependencies conan-io/conan#13302)
  • Improve dependency specification in celix_bundle_libs and add_celix_bundle.

I plan to add thorough whitebox tests to the life cycle layer and the module layer after this merged and #522 implemented.

[codecov-commenter]

Codecov Report

Merging #569 (eec55b4) into master (ab2cc5e) will increase coverage by 0.20%.
The diff coverage is 80.84%.

❗ Current head eec55b4 differs from pull request most recent head 58a3c3b. Consider uploading reports for the commit 58a3c3b to get more accurate results

@@            Coverage Diff             @@
##           master     #569      +/-   ##
==========================================
+ Coverage   78.08%   78.28%   +0.20%     
==========================================
  Files         229      230       +1     
  Lines       35016    35013       -3     
==========================================
+ Hits        27341    27409      +68     
+ Misses       7675     7604      -71     

Impacted Files	Coverage Δ
...sub_topology_manager/src/pubsub_topology_manager.c	52.38% <0.00%> (-0.45%)	⬇️
libs/framework/src/bundle.c	72.01% <0.00%> (+2.35%)	⬆️
libs/framework/src/bundle_archive.c	100.00% <ø> (+3.37%)	⬆️
libs/framework/src/bundle_context.c	80.02% <25.00%> (-0.19%)	⬇️
libs/framework/src/framework.c	81.45% <82.35%> (+3.78%)	⬆️
...framework/src/framework_bundle_lifecycle_handler.c	86.27% <91.66%> (+30.60%)	⬆️
bundles/shell/shell/src/query_command.c	84.00% <100.00%> (ø)
bundles/shell/shell/src/std_commands.c	100.00% <100.00%> (ø)
bundles/shell/shell/src/unload_command.c	100.00% <100.00%> (ø)
libs/framework/src/bundle_revision.c	100.00% <100.00%> (+27.08%)	⬆️
... and 2 more

... and 6 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[PengZheng]

Could you help review this? @pnoltes
I intend to merge this before more work (and tests) on module/life cycle layer.

[pnoltes]

Could you help review this? @pnoltes I intend to merge this before more work (and tests) on module/life cycle layer.

Yes, happily. Please allow me a few days to find the time to thoroughly review this.

[pnoltes]

Nice changes and good to see the unload addition and that the update bundle is now possible.

I do have some remarks for the recursive mutex and that the read lock is only used with useActiveBundles

[pnoltes]

LGTM.

Maybe the useActiveBundle doc can mention that the framework ensures that the bundle state cannot change during a callback.