uschindler commented on pull request #2052:
URL: https://github.com/apache/lucene-solr/pull/2052#issuecomment-748528098
Hi, I rewrote most of the ByteBuffer handling. The bugs came from the fact
that we used an additional outside bufferSize and bufferPos. All calcuations
were done based on this, and the fields inside the ByteBuffer got completely
out of sync, if alignment was wrong.
Instead of doing all this, I applied the standard ByteBuffer workflows as
documented ("how to use a ByteBuffer with flip, reset, remaining,...). The new
code uses ByteBuffer.hasRemaining() and remaining() to check if there's space.
The buffer size is buffer.limit() and the alignment in constructor is done with
overallocation. The code is now much cleaner, no strange assers and masks
anymore. The seek() method now aligns at real blockSize of OS device (using a
modulo operation).
Nevertheless, testing is a bit bad, as the default BaseDirectoryTestcase
methods often use the delegate, as the MERGE mode is done seldom. We should
maybe extend BaseDirectoryTestcase to use a "non-random" default IOContext for
all tests. I can do this in a later step.
In addition, there is no test using a real index. When using this directory
in reality it will bail out, as this method is not implemented!!!!
```java
@Override
public long getChecksum() {
throw new UnsupportedOperationException("this directory currently does
not work at all!");
}
```
We have to implement checksums!
----------------------------------------------------------------
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
