There are basicically two classes of cases:
1) Where the code is for obscure historical reasons or what I see as
misguided "optimization" split into a more libraries than necessary,
and these then are loaded at run-time. Instead, just use direct
linking.
2) Where dynamic loading is part of the functionality offered to some
upper (scripting etc) layer, or where some system-specific non-LO
library is loaded dynamically, as it is not necessarily present on
end-user machines. Can't have such in the DISABLE_DYNLOADING case.
Change-Id: I9eceac5fb635245def2f4f3320821447bb7cd8c0
This is such a fatal error that there is probably no point in trying to handle
it, so allow to simplify client code by removing the requirement to check for a
null return value.
Simplified some client code accordingly (modules configmgr and ure, and the code
generated by cppumaker and javamaker).
Change-Id: I51c0b270ec73409374f7439a47ee061407a46e31
removeRdbFiles suffered from a confusion that ImplementationInfo.uri denotes the
corresponding component (.so, .jar, etc.), but not the .rdb file. So removing
an .rdb file silently failed to remove the corresponding implementations, so re-
installing a similar enough .rdb (as typically happens during extension update)
would fail due to duplicate implementation names.
Change-Id: I25d4ff72656c99a3af509eef09e89c18cfd0aabe
Erasing from data_ member maps can destroy contained Implementations, which in
turn releases the UNO objects referenced from there, which in turn can cause
XComponents to dispose, which in turn can call arbitrary code, so must not be
done with rMutex locked. Witness the backtrace at
<https://bugs.freedesktop.org/attachment.cgi?id=65142> linked from fdo#52639
(where this fix appears otherwise unrelated to that issue's main topic).
Change-Id: If55a3841b761ec1d9a0ef61fe54784426c4ee442
...in loadImplementation (instead of using the context the ServiceManager itself
was created with). Otherwise, the handcrafted context containing a fake
theJavaVirtualMachine singleton in install_vm_singleton
(javaunohelper/source/vm.cxx) would not be honored, so that if a Java process
bootstraps native (binary) UNO and from there tries to obtain that singleton, it
would erroneously try to instantiate another JVM instead of using the existing
one. This was a regression introduced with the new ServiceManager and could be
witnessed by test-javanative in ure/source/uretest/Makefile failing.
Change-Id: I58cfbc8cdaea7ee4ab80fac728ea3e85676d69e1
...as some client code catches just the former and thus fails now. (This was a
regression introduced with the recent cppuhelper/source/defaultbootstrap.cxx.)
Change-Id: I8306797f8331d894ab4e7695478e3824e9f79197
This reverts commit b162aec625, which increased
code complexity for no benefit (the dubious scenario it was introduced for
concerned duplicate service rdbs rather than type rdbs, anyway).
...see comment in ServiceManager::createContentEnumeration for a rationale.
Splitting ImplementationInfo out of Implementation has become necessary to avoid
circular references.
Change-Id: I29aef81ce78b9ab71e18663f8c7e6ca913c6a650
Unfortunately this --enable-dbg-util only problem (caused by
_GLIBCXX_DEUBG) resurfaced, perhaps because of new std::string based
logging in sal; adapt all map files to export the unique symbol.
...that no longer uses XSimpleRegistry structures for the service data and thus
is potentially more performant.
* Registry-based functions from cppuhelper/bootstrap are deprecated now, client
code should always use defaultBootstrap_InitialComponentContext.
* References to the obsolete UNO_WRITERDB have been removed.
* Some of the functions in cppuhelper/source that are used from multiple .cxx
but had not been properly placed into .hxx have been cleaned up.
* css.lang.ServiceManager XSet insert/remove now support special
sequence<NamedValue> to improve live deployment/removal of XML-based extension
components data.
* 09524d410b "stoc: accelerate opening of multiple
XML .rdb files in a directory" and its follow-up
cb5c881a7f "avoid using the new rdb reading
logic for empty/non-existent directories" have been obsoleted by this change
and have been reverted again.
Instead of nesting these, we aggregate them into a single non-nested
registry, which saves lots of CPU at startup, sadly we can only do
that for the new-style XML registries, so we have to sniff files,
nevertheless this is still far faster. The merged xml files also
break the XSimpleRegistry::getURL() method - but it appears not
to get called.
