Done now -sorry about the noise.
There should be no warnings or output during initialization whether RIOT
initialization is via Jana, forced early by the application, or via the
ARQ test suite.
(The ARQ JMX support was not initializing in the right place - RIOT does
not need this to operate)
Andy
On 10/02/13 20:08, Andy Seaborne wrote:
Sorry - still not stable enough. I've removed it until I can be certain
it's working properly.
Andy
On 10/02/13 19:42, Andy Seaborne wrote:
On 10/02/13 16:24, Andy Seaborne wrote:
IOT - jena is not automatically initializing RIOT readers into
model.read currently; it only happens if ARQ initializes or RIOT.init is
explicitly called. May be doable for the release but I prefer to be
quite sure automatic initialization is stable for all ways to use Jena.
It turned out to be one localized cycle involving ModelCom rather than a
widespread problem.
ModelCom is the point where the existing reader/writer factories are
initialized; if these call RIOT.init themselves, then ModelCom has not
finished initializing. There are lots of constants initialized and it
was triggering a vocabulary to initialize ... which creates a Model
hence calls back into ModelCom, before it's complete.
Solution: moving the RIOT initialization to the end of ModelCom
initialization. It only needs to be before any .read call. It only
happens once.
Code committed to jena-core.
The test suite runs without RIOT.
Calling RIOT.init is very cheap once initialized.
You can tell if RIOT is initialized by either:
m.read("D.ttl", "TTL") ;
with no "file:" needed
or parse a Turtle file with an RDF 1.1 feature e.g.
------------
@prefix : <http://example/> .
:s :p :o%45 .
------------
Andy
