[
https://issues.apache.org/jira/browse/CALCITE-2471?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16634855#comment-16634855
]
Julian Hyde commented on CALCITE-2471:
--------------------------------------
I considered creating a class that implements CharSequence but is composed of a
list of CharSequences (any of which may be Strings). Its hashCode would be the
same as the equivalent String. However, I never did it because if average
string length is small then String will be a more efficient use of memory, and
I think we're close to break-even. I might be wrong.
(Remember, this bug causes digests to get way larger than that should. When it
is fixed, digest lengths will go back to under ~~100 chars on average.)
Having string digests - stored in the RelNode, not just computed on the fly -
has been incredibly useful. I would not abandon that, but CharSequence would
probably do just as well.
> RelNode description includes all tree when recomputed
> -----------------------------------------------------
>
> Key: CALCITE-2471
> URL: https://issues.apache.org/jira/browse/CALCITE-2471
> Project: Calcite
> Issue Type: Bug
> Components: core
> Reporter: Laurent Goujon
> Assignee: Julian Hyde
> Priority: Major
>
> The initial AbstractRelNode description is `{{typeName#id}}` but when
> recomputed, it includes all its inputs (and other attributes).
> When creating the final plan, HepPlanner visits each node bottom up, and
> recomputes the node digest. This causes node description to grow in size, up
> to the top node which contains basically a string copy of the whole tree. On
> large trees, this may consume a non-significant amount of memory.
> See email thread:
> https://lists.apache.org/thread.html/5e71c9f2b4e83865805841708f53471cfab1b5de673e6f162b68edc9@%3Cdev.calcite.apache.org%3E
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
