[
https://issues.apache.org/jira/browse/UIMA-2498?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13616705#comment-13616705
]
Marshall Schor commented on UIMA-2498:
--------------------------------------
Deserializing into a CAS whose type system is a superset of the CAS being read
is supported. The missing feature slots are initialized upon deserialization
with "0" which is 0 for numbers, and null for strings and FeatureStructure
references.
Re: CasCompleteSerializer - that's an APIThis stuff has its own interfaces;
suggestions welcome for how to best integrate it (if at all) into existing
APIs. The one place it's already somewhat integrated is if you use the CASImpl
"reinit(input-stream)" and the input-stream you feed it was produced by this
new version of binary compressed serialization, then it will deserialize it
(assuming identical type systems, and no support for delta cas). Different
APIs are needed for those cases.
Delta Cas formats are supported too - these are intended to be used when
serializing to a remote service, which then sends back the compressed "delta"
cas of only the changes.
> add lenient version for binary compressed serialization/deserialization
> -----------------------------------------------------------------------
>
> Key: UIMA-2498
> URL: https://issues.apache.org/jira/browse/UIMA-2498
> Project: UIMA
> Issue Type: Improvement
> Components: Core Java Framework
> Affects Versions: 2.4.0SDK
> Reporter: Marshall Schor
> Assignee: Marshall Schor
> Priority: Minor
> Fix For: 2.4.1SDK
>
>
> Extend the binary compressed serialization to support cases where the type
> systems are not exactly the same.
> There are 2 use cases.
> First: the source is a previously saved file. The goal is to deserialize it
> into e.g. a tool, where the type system in the tool may be somewhat different
> than the type system used to create the file. (For instance, it may be at a
> different version level).
> Second: the source is a client for a UIMA-AS service. In this case, the
> client has read the service's type system, and has merged it with its own.
> Difference in the type systems could be:
> Type exists in one, not in the other;
> Type exists in both, but with different features (including those from super
> types). Features could be added/subtracted. Features could have different
> ranges (incompatible ranges should cause error messages).
> A suggested impl approach: create a mapper that maps typecodes and feature
> codes; set it up by comparing two type systems. For the first use case,
> implement a version of deserialization that takes an extra input of the
> source type system, and creates the converter, and then does deserialization
> with the conversions. For the 2nd use case, during initialization time,
> after the service's type system has been read (for merging into the client's
> type system definition), use this to create the same mappper between type
> codes / feature codes; when sending a CAS via binary serialization, send it
> via the mapping converter for type codes and feature codes.
> Try to arrange things so that the creation of the mapper can be done once per
> "set" of CASes, rather than once per CAS.
> Note that managing out-of-type-system data is superseded by support for
> delta-CAS formats.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
