rdblue commented on issue #207: Add external schema mappings for files written with name-based schemas #40 URL: https://github.com/apache/incubator-iceberg/pull/207#issuecomment-533836622 @rdsr, thanks for your work on this! I think there are some subtle problems with this because it changes assumptions about null values. For example, if an array-based map is not projected, the element struct will be an empty struct, not null. This assumes it will be null. The main problem is that I'm having trouble reasoning about the changes in this PR to decide whether they are correct. That makes me think that combining projection with the read schema building is risky because it makes the whole thing hard to maintain. What about instead of adding the name mapping and projection into the visitor building a read schema you instead added the mapping to PruneColumns? I think that would make sense: if a field doesn't have an ID, it is pruned; if it is mapped, then it may be pruned depending on the projected ID set. Then we wouldn't need to change the code for building a projection at all and both classes would be simpler. What do you think?
---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
