Marshall Schor wrote:
Thilo Goetz wrote:
Marshall Schor wrote:
Current design has 1 instance of FsClassRegistry per Cas per View.
With the redesign going on for JCas we no longer need one per View.
We don't even need one per Cas. CASes in a pool, sharing the same
type system, can share generators.
Anyone object if I change the impl to move the FSClassReg instance
into the typeSystemImpl instance, since these can be shared by CASes
(and I hope that they are when it's known the CASes are using the
same type system - such as a Cas Pool - if they're not, please let me
know).
The FsClassRegistry doesn't really belong in the type system impl. I
would prefer it if we had a new object, e.g., CASMetaData, that holds
both the type system and class registry. This might eventually also
hold the index meta data, if I finally do something about that. A CAS
could then be created from such a meta data object, and the same meta
data can be used for many CASes (like we're now doing with the type
system).
Could this could be phased in later - or do you think it would be
important to do now?
I don't see a strong need to add this other new class, myself, but that
could be because I haven't spent much time
thinking about this.
There are other things in the CAS impl that could be shared among
different CASes sharing a type system.
Some of them include the arrays: featureOffset, fsSpaceReq, and
creatable Type. Moving these into the typeSystemImpl
would compute these things once, when the type system was "locked". Do
you think this is a good idea?
Methods like initTypeVariables and initTypeCodeVars also seem to belong
to the typeSystemImpl - in that they could
be run once, automatically, when the type system is "locked". Does this
seem right?
Thanks. -Marshall
The type system holds the type system data (no surprise), the FS class
registry determines how Fss are created in the CAS, and the other things
you mention organize the data layout for the internal heap. The CAS needs
all of those, not the type system. The type system is locked when the
CAS needs a fixed type system to create its internal data structures.
So yes, it makes sense to create several CASes from the same meta data.
However, it doesn't make sense to call all that stuff the type system,
because it isn't.
--Thilo
