uima-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marshall Schor <...@schor.com>
Subject Re: ResourceManager lifecycle
Date Mon, 24 Oct 2016 14:50:45 GMT
I agree this is sensible to convey to users.

You make a clever point - about destroy() not being a part of existing RMs, so
no existing code would break.

I find your arguments convincing :-) I'll put in a Jira for this.

Thanks. -Marshall

On 10/24/2016 10:25 AM, Richard Eckart de Castilho wrote:
> If an RM is implicitly created, then I would consider the AE that implicitly
> created the RM the owner of it. If a user obtains that RM and passes it on,
> then that must be under the consciousness that the user is *not* the owner.
> If the user wants to be the owner of an RM, then the user should create it
> and pass it to the AE in the first place.
> Do you think this is not sensible/conveyable to users?
> Since so far, RMs did not support destroy at all, I believe it should also
> not break existing code if we implement it in this way.
> Best,
> -- Richard
>> On 24.10.2016, at 16:16, Marshall Schor <msa@schor.com> wrote:
>> Resource Managers are not always implicitly created by the framework. 
>> Sometimes, users create RMs exactly for the purpose of having some parts of them
>> shared among multiple pipelines / resources.
>> A possible use case:
>> * user code calls UIMA framework which creates RM as a side effect of user
>> calling produceResource (or one of its specializations, e.g. produceAnalysisEngine).
>> * user gets that RM, and then passes it to another produceResource call.
>> So we have mixed ownership: the first call the framework created the RM, the
>> 2nd, the RM was passed in by the user.
>> When the use "destroys" (for example) the Analysis Engine produced by the first
>> call a simple approach would be to destroy its RM, but that could break the
>> second use.
>> -Marshall

View raw message