ByteBufBsonDocument and RawBsonDocument simplifications

[rozza]

Rationale

ByteBufBsonDocument#clone used to return a RawBsonDocument. The recent changes returned a normal BsonDocument, which is potentially expensive depending on its usage.

The ByteBufBsonDocument changes also added complex iterator logic, when RawBsonDocument deferred to BsonDocument iterators. As iteration is essentially a hydrating mechanism, there is opportunity for improvements for both implementations. By changing the RawBsonDocument iterators to be more efficient, ByteBufBsonDocument can now utilize these efficiency gains by proxy, relying on the cachedDocument iterators.

This change both reduces the complexity of ByteBufBsonDocument and relies on an improved RawBsonDocument implementation.

Summary of changes

• ByteBufBsonDocument:

  • Simplify by returning RawBsonDocument from toBsonDocument, avoiding full BSON deserialization. When there are no sequence fields, the body bytes are cloned directly. When sequence fields exist, BsonBinaryWriter.pipe() merges the body with sequence arrays efficiently.
  • Use toBsonDocument for iterators. This eliminates the need for custom iterators (IteratorMode, CombinedIterator, createBodyIterator, and sequence iterators) since entrySet/values/keySet now delegate to the cached RawBsonDocument.

• RawBsonDocument:

  • Renamed toBaseBsonDocument to override the default toBsonDocument implementation.
  • Implemented the iterators so that they don't need to fully convert the document to a BsonDocument.

• Tests:

  • Updated ByteBufBsonDocumentTest iteration tests.
  • Updated ByteBufBsonArrayTest#fromValues as entrySet now returns RawBsonDocument instances.

JAVA-6010

[strogiyotec]

I think I understand the login , the entrySet returns a SET though if I check previous implementation without a change it can still return duplicates
Here is the test case using latest main without this change

    @Test
    void testLatestPR() {
        BsonDocument doc = new BsonDocument()
                .append("a", new BsonInt32(1))
                .append("b", new BsonInt32(1))
                .append("c", new BsonInt32(1));
        RawBsonDocument rawDoc = new RawBsonDocument(doc, new BsonDocumentCodec());

        // BsonDocument preserves all 3 values
        assertEquals(3, doc.values().size());

        // RawBsonDocument should too — but LinkedHashSet deduplicates them to 1
        assertEquals(3, rawDoc.values().size());
    }

This test case is passing but fails in this branch because LinkedHashSet removes duplicate values, I believe we should preserve an already broken behavior and return duplicates

[strogiyotec]

the javadoc for RawBsonDocument say it's immutable, does it make sense to store entrySet/values in memory instead of calculating them all from scratch ? Something like what java.lang.String does with a hashcode, keep hashcode as class field but calculates it only once during the first hashCode call

Not mandatory

[strogiyotec]

this buffer is never closed, can we wrap it with try with resources

[strogiyotec]

I just checked , the close just sets buffer to null, so not mandatory but still good to have in case we later change the class and close() will be more complicated than that