[
https://issues.apache.org/jira/browse/UIMA-5536?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16129397#comment-16129397
]
Marshall Schor commented on UIMA-5536:
--------------------------------------
Here's a slight change to the implementation, to reduce the "surprise"
(mismatch between user's expectations and what the system does).
For the case where no typePriorities (either because none were specified, or
because the "select" API was used without turning on typePriority):
* For the move-to-leftmost one: have the move-to-left-most-one include the
types in the compare. This guarantees that if the index contains instances of
that type, the iterator will find all of them that might be equal. And if one
exists, it will be the first one returned.
* For the skip when the bound is equal to the FS, have the equal include the
type, if using v2 style bounded skipping. This matches the v2 implementation.
> uv3 remove positionUsesType from select api
> -------------------------------------------
>
> Key: UIMA-5536
> URL: https://issues.apache.org/jira/browse/UIMA-5536
> Project: UIMA
> Issue Type: Bug
> Affects Versions: 3.0.0SDK-alpha02
> Reporter: Marshall Schor
> Assignee: Marshall Schor
> Fix For: 3.0.0SDK-beta
>
>
> The select API attempts to combine some of uimaFIT's capability for
> iterating. The design included a configuration choice "positionUsesType".
> This was implemented incorrectly for some cases, and it seems it would be
> hard to fix. This Jira suggests removing it altogether.
> What it does: It affects two bits of iterator operation:
> # moveTo(fs) - move to the leftmost item that "matches" fs
> # for bounded iterators: skip the item which is "equal" to the bound.
> Details: The "matches" in 1) is based on the comparator for the index. For
> Annotation indexes, this includes the typePriority. However, the select API
> framework by default doesn't use type priorities (can be configured though).
> This means, if the underlying iterator is using type priorities, then doing a
> moveTo on it will move to a spot which might not be the "left-most" when type
> priorities are not used. The correction for this has two flavors: depending
> on the setting of positionUsesType. If not specified, the correction worked,
> and moved to the left-most same-begin-end-ignoring-type. If specified, the
> correction was ignoring this, and a fix would need to scan in both directions
> to find the right type. Although this could be done, it's complex. And I
> think this is designing beyond a real need.
> Use 2 is for skipping the bounding annotation when iterating (if it is in the
> index). The normal mode for skipping is to skip only the item if it is ==
> identical to the bounding annotation. But (for UIMA v2 bounded iterator
> compatibility) a mode where the compare is done using the Annotation
> Comparator is provided. The positionUsesType would narrow this equal test by
> also requiring that the types match. I think this is unnecessarily complex,
> since I think most uses will be using the == style.
> So I propose to remove positionUsesType from the implementation /
> documentation.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
