[
https://issues.apache.org/jira/browse/IGNITE-14748?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Ivan Bessonov reassigned IGNITE-14748:
--------------------------------------
Assignee: Ivan Bessonov
> Ordered @NamedConfigValue
> -------------------------
>
> Key: IGNITE-14748
> URL: https://issues.apache.org/jira/browse/IGNITE-14748
> Project: Ignite
> Issue Type: Sub-task
> Reporter: Alexander Belyak
> Assignee: Ivan Bessonov
> Priority: Major
> Labels: iep-55
>
> Implement some order for @NamedConfigValue fields.
> Imagine that we have some
>
> {code:java}
> @Config
> public class PKIndexConfigurationSchema {
> @Value
> String type;
> @NamedConfigValue
> IndexColumnConfigurationSchema columns.
>
> {code}
> and
>
> {code:java}
> @Config
> public class IndexColumnConfigurationSchema {
> @Value
> String name;
> @Value
> boolean asc;
> @Value
> boolean affinityCol;
> }
> {code}
>
> For now we have to use indexes to store such config like:
>
> {noformat}
> "PK":
> "type":"PrimaryKey",
> "columns": {
> "0": {
> "name":"REGION",
> "asc":true,
> "affinity":true
> },
> "1": {
> "name":"COMPANY",
> "asc":true,
> "affinity":false
> }
> }
> {noformat}
>
> because we have to keep it's order.
> But if configuration keep order for @NamedConfigValue it can look like:
>
> {noformat}
> "PK":
> "type":"PrimaryKey",
> "columns": {
> "REGION": {
> "asc":true,
> "affinity":true
> },
> "COMPANY": {
> "asc":true,
> "affinity":false
> }
> }
> {noformat}
> And to allow insert value in the middle it will be nice to have some methods
> like:
> * listChange.create(idx, key, consumer(elementChange))
> or
> * listChange.createAfter(prevKeyOrNull, key, consumer(elementChange))
> in addition to existing:
> * listChange.create(key, consumer(elementChange))
> * listChange.update(key, consumer(elementChange))
> * listChange.delete(key)
> BTW, lets remove listChange.update method.
> h3. Implementation notes
> It would make sense to store order number inside of named list entry. It
> would look like implicit configuration parameter {{<idx>}}, for example. This
> value will be recalculated on every update.
> Index will be stored in named list itself, entries will not contain it.
> Reason is simple - named list entries can be used as regular "inner" nodes
> and we can't distinguish one from the another. That's why index is implicit.
> h3. API notes
> I don't get why we need to remove update method. It would be helpful to
> update their semantics, like "create" would throw "AlreadyExistsException" or
> something, update would do similar thing with "KeyNotFound"...
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
