leerho opened a new pull request, #526:
URL: https://github.com/apache/datasketches-cpp/pull/526
## What changed
A compact theta sketch holding exactly one entry in exact mode now sets bit
5 of the flags byte, the bit Java has always written as `SINGLEITEM_FLAG_MASK`.
The flags enum gains `IS_SINGLE_ITEM`, a private `is_single_item()` mirrors
Java's condition (`curCount == 1 && thetaLong == Long.MAX_VALUE`, here
`entries_.size() == 1 && !is_estimation_mode()`), and both version-3 writers OR
it in. The compressed form never carries a single item, so
`serialize_version_4` is untouched.
Readers do not change. All four deserializers test individual bits and have
always ignored this one, which is why C++ already reads Java's single-item
images; it now also reads its own.
## Why
This was the last theta image that differed between C++ and Java. With
`c11a5ea` (zero seed hash on an empty sketch, on master but not yet released)
every other theta fixture — empty, n=10 through 1,000,000, all six compressed
sizes, and non-empty-no-entries — is already byte-identical. The single-item
image differed only at byte 5: `0x1A` here, `0x3A` in Java.
Before, with the fixtures regenerated from master:
```
theta_n1_cpp.sk 01 03 03 00 00 1a cc 93 e5 6b 61 ee c8 80 44 20
theta_n1_java.sk 01 03 03 00 00 3a cc 93 e5 6b 61 ee c8 80 44 20
```
After this change the two files are identical and the Java
`ThetaSketchCrossLanguageTest.checkCpp` byte-for-byte comparison passes for all
eight sizes. The C++ `[serde_compat]` readers pass against the current Java
fixtures in the other direction.
## Tests
Three new cases pin the exact serialized images, which nothing asserted
before:
- the 8-byte empty image `{1, 3, 3, 0, 0, 0x1E, 0, 0}` from `serialize()`,
the stream form, and `serialize_compressed()`, using a non-default seed to show
the seed hash is not written;
- the 16-byte single-item image with flags `0x3A`, the seed hash present,
and the one hash, from all three writers, plus a round trip through
`deserialize()` both with and without the flag to show the reader does not
depend on it;
- one entry in estimation mode, which has a 3-long preamble and must not get
the flag.
Full theta and tuple suites pass.
🤖 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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
