Passive Component Registration
Motivation
UNO components in the various language bindings (dynamic libraries for binary/C++ UNO, jars for Java UNO, etc.) have mechanisms to actively report information about the services and singletons they offer (exported symbol component_writeInfo
for dynamic libraries, public static __writeRegistryServiceInfo
method for jars, etc.). This active component registration has various drawbacks:
- Calls to actively register the components (
regcomp
) need to be made in carefully crafted environments, ensuring that the relevant UNO components can actually be loaded and their code executed. Especially for Java and Python components, this is nontrivial.- At OOo build time, when a
services.rdb
is assembled ininstsetoo_native
. - At OOo installation time, when bundled extensions that contain UNO components are registered.
- At OOo run time, when extensions that contain UNO components are registered or revoked.
- At OOo build time, when a
- Loading the UNO components to actively execute their code can be time consuming (especially if the components' functionality is not otherwise needed, like when assembling a
services.rdb
at build time).
Therefore, it is sought to replace active component registration with a passive approach, where a UNO component is accompanied by a representation of the relevant information that can be obtained without running the component's code.
An XML Format
The canonic choice for representing such information these days is XML. The necessary information is encoded as follows:
An XML document encoding information about multiple UNO components (i.e., corresponding to a services.rdb
) has a root element with local name components
and namespace URI http://openoffice.org/2010/uno-components
.
The components
root element contains zero or more elements with local name component
and namespace URI http://openoffice.org/2010/uno-components
, each with an attribute with unprefixed name uri
, the value of which is a URI (in OOo-internal form) denoting the relevant UNO component (and which may be a vnd.sun.star.expand
URL), and an attribute with unprefixed name loader
, the value of which is the name of the relevant UNO service with which to load the component (e.g., com.sun.star.loader.SharedLibrary
or com.sun.star.loader.Java2
). Each component
element in turn contains zero or more service
elements, followed by zero or more singleton
elements.
Each element with local name service
or singleton
and namespace URI http://openoffice.org/2010/uno-components
has an attribute with unprefixed name name
, the value of which is the name of the respective service or singleton.
(The information provided by active component registration contains redundancies that are dropped from this XML format. For one, UNO components are listed in an IMPLEMENTATIONS
section and each service and singleton is again listed in a SERVICES
resp. SINGLETONS
section. For another, each singleton specifies the service implementing it.)
There was already a somewhat similar textual format for regcomplazy
in use, which however was based on ini-file syntax (but with irregularities, containing lines that do not match the key=value
format) instead of XML. This has been superseded by the approach described here, and regcomplazy
has been removed.
As we will see later, within the OOo build system there are also XML files describing single UNO components, and which have a component
element instead of a components
element as their root element, but otherwise use the same structure as the above XML format.
XML at Run Time
There was a choice whether to either take the XML files and inject them into registry-format services.rdb
files (i.e., to still call regcomp
resp. unopkg
at build resp. installation time), or to use the XML files directly at run time. I decided for the latter.
The configmgr
re-write showed that it is acceptable performance-wise to read a handful of XML files during OOo start up, using an XML reader specifically written for that task. So it should be acceptable to replace the existing registry-format services.rdb
files (for a start, the URE layer services.rdb
and the basis layer services.rdb
and legacy_binfilters.rdb
) with XML files.
The advantages are easier and potentially faster creation (no need for regcomp
like code any more), and easier manipulation (due to the textual format). Also, the registry format has a known design defect, in that it cannot handle arbitrary combinations of key names.
Work in Progress
Work is done on CWS sb129 , currently as a stack of MQ patches:
simpleregistry-cleanup
cleans up the stoc/source/simpleregistry
code so that subsequent patches can better modify it.
xmlreader
extracts the XmlReader
code from configmgr
to a new xmlreader
URE module, so that it can be reused from stoc
. An open problem is how to version its C++ ABI (currently done via symbol visibility, which does not allow for versioning).
textualservices
extends the implementation of com.sun.star.registry.SimpleRegistry
in stoc/source/simpleregistry
, so that it can internally also read XML files in addition to registry files, and present them to its clients as if they had the structure of registry-format services.rdb
files. This appeared to be the simplest approach for now, given how deep the knowledge about registry-format files is rooted in the code base (see, for example, the functionality in cppuhelper/bootstrap.hxx
that bootstraps a UNO environment based on rdb
files); it is still subject to change, however.
passive
replaces the active registration of all of OOo's UNO components with passive registration. The resulting XML files are still called services.rdb
etc. (instead of, say, services.xml
), mainly because [ure/source/README
] declares the URE services.rdb
as part of the published URE interface (acknowledging only its existence, not its internal format, and especially warning against registering anything into it).