rustyconover opened a new pull request, #978:
URL: https://github.com/apache/arrow-nanoarrow/pull/978

   An offsets buffer holds `length + 1` offsets, so even an empty string, 
binary, list or map array has one: `offsets[0] = 0`. The array view of an empty 
array has a zero-length offsets buffer, though (`ArrowArrayViewSetArray()` 
deliberately doesn't require one), and `ArrowIpcEncoderCollectArray()` wrote 
that buffer as is. So every empty variable-size array nanoarrow writes has a 
0-byte offsets buffer.
   
   IPC permits that, and Arrow C++ reads it, but not every reader does. With 
nanoarrow 0.9.0 (Python) writing `struct<s: string, l: list<int32>>` with zero 
rows, arrow-go v18.8.0 reads the batch and then panics on 
`String.ValueOffsets()`:
   
   ```
   panic: runtime error: slice bounds out of range [:1] with capacity 0
   ```
   
   The same batch written by pyarrow 25.0.1 has a 4-byte offsets buffer (8 
bytes for `large_string`), and arrow-go reads it. Apache.Arrow for .NET had the 
same problem (apache/arrow-dotnet#462, now fixed there, but released versions 
still throw).
   
   The encoder now writes one zero offset (4 or 8 bytes) in place of a 
zero-length offsets buffer of an empty array, as Arrow C++ does. Non-empty 
arrays and other buffer types are unchanged.
   
   Tests:
   - `NanoarrowIpcEncoderWritesOffsetForEmptyArrays` encodes an empty 
`struct<s: string, b: binary, ls: large_string, l: list<int32>>` built the 
usual way. It decodes the message and checks that each child's offsets buffer 
holds one zero offset of the right width. On `main` the test fails: the buffers 
are 0 bytes.
   - `AssertArrayViewIdentical()`, used by the decoder round-trip tests, 
compared buffer sizes exactly. It now expects that one zero offset where the 
original empty array's view has a zero-length offsets buffer.
   
   `ctest` passes (Linux aarch64, Debug, 868 tests with 
`NANOARROW_BUILD_TESTS_WITH_ARROW`), and 393 tests under ASan/UBSan with clang. 
The changed files are formatted with clang-format 16.0.6, the pre-commit 
version.
   
   This PR's code and description were written by AI (Claude Code) and reviewed 
by a human maintainer.
   
   🤖 Generated with [Claude Code](https://claude.com/claude-code)
   


-- 
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