You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Michael Munday (Jira)" <ji...@apache.org> on 2020/07/29 14:06:00 UTC
[jira] [Created] (SPARK-32485) RecordBinaryComparatorSuite test
failures on big-endian systems
Michael Munday created SPARK-32485:
--------------------------------------
Summary: RecordBinaryComparatorSuite test failures on big-endian systems
Key: SPARK-32485
URL: https://issues.apache.org/jira/browse/SPARK-32485
Project: Spark
Issue Type: Bug
Components: Tests
Affects Versions: 3.0.0
Reporter: Michael Munday
The fix for SPARK-29918 broke two tests on big-endian systems:
* testBinaryComparatorWhenSubtractionIsDivisibleByMaxIntValue
* testBinaryComparatorWhenSubtractionCanOverflowLongValue
These tests date from a time where subtraction was being used to do multi-byte comparisons. They try to trigger old bugs by feeding specific values into the comparison. However the fix for SPARK-29918 modified the order in which bytes are compared when comparing 8 bytes at a time on little-endian systems (to match the normal byte-by-byte comparison). This fix did not affect big-endian systems. However the expected output of the tests was modified for all systems regardless of endianness. So the tests broke on big-endian systems.
It is also not clear that the values compared in the tests match the original intent of the tests now that the bytes in those values are compared in order (equivalent to the bytes in the values being reversed).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org