[PIG-3373] XMLLoader returns non-matching nodes when a tag name spans through the block boundary - ASF JIRA

XML

Word

Printable

JSON

Patch Info:

Patch Available
Release Note:

Hide
I added a new patch that fixes this bug. It turned out that this bug happens only when the input file is .bz2 compressed and the non-matching tag spans two file splits in the compressed file. Since it's almost impossible to tailor an example that has this bug since the compression is virtually non-deterministic, I included a random generator that generates this test case.
I don't like the idea of discovering a bug using this randomly generated file since, by definition, it's non-deterministic, I attached the test file for reference.
The fix is still the same as the previous patch, but this time, the test fails without this fix.

Show
I added a new patch that fixes this bug. It turned out that this bug happens only when the input file is .bz2 compressed and the non-matching tag spans two file splits in the compressed file. Since it's almost impossible to tailor an example that has this bug since the compression is virtually non-deterministic, I included a random generator that generates this test case. I don't like the idea of discovering a bug using this randomly generated file since, by definition, it's non-deterministic, I attached the test file for reference. The fix is still the same as the previous patch, but this time, the test fails without this fix.

When node start tag spans two blocks this tag is returned even if it is not of the type.
Example: For the following input file

XMLoader with tag type 'event' should return only the first one but it actually returns both of them

is part of

PIG-3865 Remodel the XMLLoader to work to be faster and more maintainable