henriquejsza opened a new pull request, #4916:
URL: https://github.com/apache/bookkeeper/pull/4916

   Fix #4736
   
   ### Motivation
   
   `FileInfo.moveToNewLocation` copies the ledger index file to a new location 
with
   `FileChannel.transferTo`:
   
   ```java
   long written = 0;
   while (written < size) {
       long count = fc.transferTo(written, size, newFc);
       ...
       written += count;
   }
   if (written <= 0 && size > 0) {
       throw new IOException("Copying to new location " + rlocFile + " failed");
   }
   ```
   
   `transferTo` may transfer fewer bytes than requested. Because the count 
passed to
   every call is `size` (not the number of bytes still to copy), a call after a
   partial transfer reads from position `written` for up to `size` more bytes. 
When
   the source file holds more than `size` bytes, that copies past the requested
   range, so the new location can end up with more than `size` bytes. This 
matters
   in practice: `moveToNewLocation` is called with `getSizeSinceLastWrite()`, 
which
   can be smaller than the actual file size.
   
   The post-loop `written <= 0 && size > 0` check reported in the issue is
   unreachable: when `size > 0` the loop runs at least once and either throws or
   increments `written` to a positive value.
   
   ### Changes
   
   - Pass the remaining byte count (`size - written`) to `transferTo` on each
     iteration so exactly `size` bytes are copied.
   - Extract the copy loop into a small package-private `transferFully` helper 
and
     remove the unreachable post-loop check.
   - Add `FileInfoTest` covering a partial transfer, asserting that exactly 
`size`
     bytes are copied.
   


-- 
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.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to