[
https://issues.apache.org/jira/browse/CALCITE-3517?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17157102#comment-17157102
]
Chunwei Lei commented on CALCITE-3517:
--------------------------------------
Hi, [~laurent], do you think we can get it into 1.24?
> DiffRepository spends too much time writing XML, makes some tests 5x slower
> ---------------------------------------------------------------------------
>
> Key: CALCITE-3517
> URL: https://issues.apache.org/jira/browse/CALCITE-3517
> Project: Calcite
> Issue Type: Bug
> Reporter: Julian Hyde
> Priority: Major
> Fix For: 1.24.0
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Tests that use {{DiffRepository}} are spending far too much effort writing
> XML, even if the XML matches the reference file. For example, If I comment
> out [a call to set(tag,
> next)|https://github.com/apache/calcite/blob/ee83efd360793ef4201f4cdfc2af8d837b76ca69/core/src/test/java/org/apache/calcite/test/DiffRepository.java#L267],
> {{RelOptRulesTest}} improves from 32s to 6s; {{SqlToRelConverterTest}}
> improves from 24s to 4.7s; {{SqlPrettyWriterTest}} remains .8s.
> The {{DiffRepository.expand}} method is the cause of the inefficiency. It
> causes the entire XML document to be re-generated and written to disk. This
> is not just slow but quadratic - if a test has N cases, each test writes the
> XML document, an effort proportional to N.
> {{DiffRepository}} should remain conservative. If one of the tests fails, and
> a later test crashes, the output from the failed test should have been
> written out. It is acceptable if the test remains slow if there are test
> failures.
> {{DiffRepository}} is only used in tests; this bug does not affect production
> code.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
