voonhous commented on code in PR #18961:
URL: https://github.com/apache/hudi/pull/18961#discussion_r3449398313
##########
hudi-spark-datasource/hudi-spark4-common/src/main/java/org/apache/hudi/variant/Spark4VariantShreddingProvider.java:
##########
@@ -398,4 +437,240 @@ public boolean allowNumericScaleChanges() {
return true;
}
}
+
+ /**
+ * Base {@link ShreddingUtils.ShreddedRow} with all accessors throwing;
concrete rows override
+ * only the accessors valid for their nesting context. This is the read-path
mirror of the
+ * write-path {@link AvroShreddedResult}: it reads Avro records to feed
Spark's reconstruction
+ * ({@link ShreddingUtils#rebuild}).
+ */
+ abstract static class BaseAvroShreddedRow implements
ShreddingUtils.ShreddedRow {
+ @Override
+ public boolean isNullAt(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public boolean getBoolean(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public byte getByte(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public short getShort(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public int getInt(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public long getLong(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public float getFloat(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public double getDouble(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public BigDecimal getDecimal(int ordinal, int precision, int scale) {
+ throw unsupported();
+ }
+
+ @Override
+ public String getString(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public byte[] getBinary(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public UUID getUuid(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public ShreddingUtils.ShreddedRow getStruct(int ordinal, int numFields) {
+ throw unsupported();
+ }
+
+ @Override
+ public ShreddingUtils.ShreddedRow getArray(int ordinal) {
+ throw unsupported();
+ }
+
+ @Override
+ public int numElements() {
+ throw unsupported();
+ }
+
+ private static UnsupportedOperationException unsupported() {
+ return new UnsupportedOperationException("Accessor not valid for this
shredded row context");
+ }
+ }
+
+ /**
+ * A shredded variant struct {@code {value, [metadata], typed_value}}. Maps
the Spark
+ * {@link VariantSchema} ordinals (variantIdx / topLevelMetadataIdx /
typedIdx) back to the named
+ * Avro fields, and reads {@code typed_value} for scalar/object/array
reconstruction.
+ */
+ static final class AvroVariantRow extends BaseAvroShreddedRow {
+ private final GenericRecord record;
+ private final VariantSchema schema;
+
+ AvroVariantRow(GenericRecord record, VariantSchema schema) {
+ this.record = record;
+ this.schema = schema;
+ }
+
+ private String fieldNameFor(int ordinal) {
+ if (ordinal == schema.typedIdx) {
+ return HoodieSchema.Variant.VARIANT_TYPED_VALUE_FIELD;
+ }
+ if (ordinal == schema.variantIdx) {
+ return HoodieSchema.Variant.VARIANT_VALUE_FIELD;
+ }
+ if (ordinal == schema.topLevelMetadataIdx) {
+ return HoodieSchema.Variant.VARIANT_METADATA_FIELD;
+ }
+ throw new IllegalArgumentException("Unexpected shredded ordinal: " +
ordinal);
+ }
+
+ @Override public boolean isNullAt(int ordinal) {
+ return record.get(fieldNameFor(ordinal)) == null;
+ }
+
+ @Override public byte[] getBinary(int ordinal) {
+ return toByteArray((ByteBuffer) record.get(fieldNameFor(ordinal)));
+ }
+
+ // The scalar getters below read typed_value directly: Spark only invokes
them for the scalar
Review Comment:
Good call. Went with a small `typedValue()` helper instead of per-getter
comments: the scalar/struct/array getters now read `typedValue()` rather than
`record.get(VARIANT_TYPED_VALUE_FIELD)`, so the ignored `ordinal` is obvious at
every call site and the rationale lives in one comment on the helper.
--
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]
