[
https://issues.apache.org/jira/browse/MYFACES-3236?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Leonardo Uribe resolved MYFACES-3236.
-------------------------------------
Resolution: Fixed
Fix Version/s: 2.1.2
2.0.8
> UIData performance improvements
> -------------------------------
>
> Key: MYFACES-3236
> URL: https://issues.apache.org/jira/browse/MYFACES-3236
> Project: MyFaces Core
> Issue Type: Improvement
> Components: JSR-314
> Affects Versions: 2.0.7, 2.1.1
> Reporter: Leonardo Uribe
> Assignee: Leonardo Uribe
> Fix For: 2.0.8, 2.1.2
>
>
> Some days ago there was a discussion on dev list titled:
> [core] performance: performance hints
> With the following intention (proposed by Martin Koci):
> MK>> is it possible to introduce performance hints in myfaces-core? Hints
> MK>> similar to javax.faces.component.visit.VisitHint but related to
> MK>> performance improvements. Example:
>
> MK>> For dataTable like:
> MK>> <a:dataTable
> MK>> <a:column>
> MK>> #{aExpression}
>
> MK>> it's completely unnecessary to save per-row state. Currently there
> is no
> MK>> elegant way how to do read-only table (state per-row is always
> MK>> maintained). If user wants (fast) readOnly table, he/she must extend
> MK>> UIData and re-implemenent setRowIndex method. But hint say
> MK>> "org.apache.myfaces.core.UIData.saveRowState"=false can solve it
> MK>> elegantly - if present (in component.getAttributes()) UIData skips
> MK>> row-state-saving and restoring methods entirely.
>
> MK>> Lifespan of those hints can be request (faceContext.attributes) or
> view
> MK>> (component.attributes)
>
> The discussion there was to create or not a hint but a review of the default
> algorithm was never done. In theory, UIData.setRowIndex do the following
> tasks:
> 1. If the component is in row -1 (no row), take a "snapshot" of the
> components that implements EditableValueHolder, to restore them later when
> rows are traversed.
> 2. If the component is in a row, save the current state of
> EditableValueHolder components.
> 3. Move to the selected row.
> 4. If no state saved found, restore EditableValueHolder components from saved
> initial state.
> 5. If state saved found, restore EditableValueHolder components from stored
> saved state.
> After an in-deep analysis, the conclusion was it is not really necessary to
> create the hint. Instead, we can create a better algorithm that take
> advantage of
> the fact that add a non transient EditableValueHolder component inside a
> dataTable row on render response time leads to a illegal state.
> Since UIData.setRowIndex is a code that is called multiple times, specially
> for large datatable sets. Doing some performance tests, it was notice the
> current
> algorithm uses a lot of memory resources. The proposal is do the following:
> 1. Iterate using index instead iterator for children. This can duplicate the
> lines of code
> 2. Do not create Object[]{null,null} instances, use a private static final
> variable.
> 3. Cache DataListener[].
> 4. If no EditableValueHolder instances found, skip state saving code, but
> reset all ids.
> 5. Prevent unnecessary calls to getContainerClientId().
> 6. Provide initial size for created ArrayList instances to reduce memory
> usage.
> 7. Iterate over row state using index, instead iterator, and take advantage
> instances are ArrayList.
> 8. Do not instantiate ArrayList, unless it is necessary.
>
> Tests done shows a really big improvement, specially when non
> EditableValueHolder component instances are inside the datatable, which is a
> very common use case.
> Based on this patch we can do other things like:
> 1. Port this code to UIRepeat.
> 2. Port this code to Tomahawk.
> 3. Reuse already created state instances, which can improve performance on
> postback requests.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
