wgtmac commented on code in PR #793:
URL: https://github.com/apache/iceberg-cpp/pull/793#discussion_r3510509098
##########
src/iceberg/update/update_schema.cc:
##########
@@ -282,6 +289,24 @@ std::vector<SchemaField> ApplyChangesVisitor::MoveFields(
return reordered;
}
+/// \brief Cast a default value literal to a target primitive type.
+///
+/// `Literal::CastTo` only returns a value for identical decimal types, but
schema
+/// evolution permits widening a decimal to a larger precision at the same
scale. Such a
+/// promotion leaves the unscaled value unchanged, so the literal is rebuilt
with the
+/// target type instead of going through `CastTo`. All other conversions
delegate to
+/// `CastTo`, which reports narrowing via AboveMax/BelowMin sentinels that
callers reject.
+Result<Literal> CastDefaultToType(const Literal& value,
+ const std::shared_ptr<PrimitiveType>&
target_type) {
+ if (value.type()->type_id() == TypeId::kDecimal &&
+ target_type->type_id() == TypeId::kDecimal) {
+ const auto& decimal_type = internal::checked_cast<const
DecimalType&>(*target_type);
+ return Literal::Decimal(std::get<Decimal>(value.value()).value(),
+ decimal_type.precision(), decimal_type.scale());
Review Comment:
Could we guard this decimal fast path with a same-scale check before
rebuilding the literal? `Decimal` stores only the unscaled value, so accepting
`decimal(9,3)` as a default for `decimal(18,2)` would reinterpret `1234` from
`1.234` to `12.34` instead of rejecting it. Java default parsing/serialization
requires the decimal default scale to match the column type, so only same-scale
precision widening should be accepted here.
--
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]
