JingsongLi commented on code in PR #1447:
URL: https://github.com/apache/paimon-rust/pull/1447#discussion_r4609779774


##########
crates/paimon/src/spec/schema.rs:
##########
@@ -672,8 +672,7 @@ impl TableSchema {
                             ),
                         });
                     }
-                    fields[idx] = DataField::new(old.id(), 
old.name().to_string(), target)
-                        .with_description(old.description().map(|s| 
s.to_string()));
+                    fields[idx] = old.clone().with_data_type(target);

Review Comment:
   Could we validate the existing default against `target` before publishing 
the type change? On an empty Format Table, narrowing `BIGINT DEFAULT 
2147483648` to `INT` succeeds here and keeps the out-of-range default. Every 
subsequent `new_write()` then fails because `FormatTableWriter::new` parses all 
defaults up front, even when the caller supplies a value for that column. A 
regression test should assert that this ALTER fails and leaves the schema 
version unchanged. [Java's type-change 
validation](https://github.com/apache/paimon/blob/2bf10c194b9213e406b8cf0732b9e947f835cebd/paimon-core/src/main/java/org/apache/paimon/schema/SchemaManagerUtils.java#L353-L365)



##########
crates/paimon/src/spec/schema.rs:
##########
@@ -733,6 +732,8 @@ impl TableSchema {
         new_schema.highest_field_id =
             
highest_field_id.max(Self::current_highest_field_id(&new_schema.fields));
 
+        super::generated_column::validate_evolution(&self.fields, 
&new_schema.fields)?;

Review Comment:
   Could we also enforce Java's generated-column constraints in 
`Schema::validate_final_schema`, which covers both CREATE and ALTER? Adding a 
`ROW` with a generated child to an ordinary table passes this top-level check. 
Setting `rowkind.field` to an existing generated `STRING` column also passes 
because the fields are unchanged. At creation time the builder likewise accepts 
a column with both `defaultValue` and `generationExpression`. Java rejects all 
three when it initializes the table store, so a successful Rust DDL can leave 
Java readers and writers unable to use the table. These checks only need the 
final fields and options. [Java schema 
validation](https://github.com/apache/paimon/blob/2bf10c194b9213e406b8cf0732b9e947f835cebd/paimon-core/src/main/java/org/apache/paimon/schema/SchemaValidation.java#L1503-L1537)
 ยท [Store 
initialization](https://github.com/apache/paimon/blob/2bf10c194b9213e406b8cf0732b9e947f835cebd/paimon-core/src/main/java/org/apache/paimon/table/AbstractFileSto
 reTable.java#L370-L379)



-- 
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