[GitHub] [lucene] rmuir commented on pull request #172: LUCENE-9992: write empty vector fields when merging

[GitBox <gi...@apache.org>]

rmuir commented on pull request #172:
URL: https://github.com/apache/lucene/pull/172#issuecomment-856360112


   I hit the issue locally just running `./gradlew test`. A second time the tests pass.
   
   I think it isn't great that simple base-cases like empty are only exposed in random test failures (although: better than nothing!). So I am curious if there are any ideas on improving testing.
   
   I think there are a few challenges:
   * filenames/extensions used should be codec-private/impl, so we try to avoid hardcoding them
   * most codec tests are in base shared generic classes in test-framework: a good thing
   * new abstractions to the index (*Formats) are added rarely, so all the implicit/unwritten "rules" are forgotten by everyone.
   
   So today, nobody tests simple stuff like: should a filename exist or not? And it leaves us with too many unwritten assumptions about how Formats should behave in the corner cases.
   
   Maybe it should be as simple as various BaseXYZTestCase accepting a list of used file extensions as an abstract method or parameter? Then e.g. BaseDocValuesFormatTestCase could have some simple unit tests making asserts about e.g. what happens wrt filenames when there's no docvalues, or stuff gets merged away, or whatever the cases are? Maybe we should do it for every part of the index?
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@lucene.apache.org
For additional commands, e-mail: issues-help@lucene.apache.org