sainimal1ba-hue commented on issue #13217:
URL: https://github.com/apache/seatunnel/issues/13217#issuecomment-6895560362

   > Thanks for separating the problem into source reading, the SeaTunnel 
model, and target DDL rendering. This is a real inconsistency, but it is a 
cross-module data-contract design, not yet an unclaimed implementation PR.
   > 
   > I rechecked current `dev` at `b3828eb141d3ca669181084dbd73bfb534064e08`:
   > 
   > * `Column` still carries `defaultValue` as an untyped `Object`.
   > * `MySqlCatalog.buildColumn` reads only `COLUMN_DEFAULT`.
   > * `MysqlCreateTableSqlBuilder.buildColumnIdentifySql` renders the name, 
type, nullability and comment, but no column `DEFAULT`.
   > * `JdbcDialect.applySchemaChange` deliberately emits a default only when 
source and target dialect names match; its TODO explicitly leaves 
cross-database default DDL unsupported, and the special-value hook only 
recognizes MySQL defaults.
   > 
   > I also checked open and closed PRs for this issue number and 
`DefaultValueConverter`; none is an exact implementation lane. The open 
JSON-format default-value PRs 
[#11633](https://github.com/apache/seatunnel/pull/11633) and 
[#11634](https://github.com/apache/seatunnel/pull/11634) address missing/null 
record values and must remain separate from catalog and DDL semantics.
   > 
   > Before any code PR, please drive this as an STIP/design discussion and 
link it here. The design needs explicit maintainer agreement on the following 
boundaries:
   > 
   > 1. **Compatibility.** Define how an additive representation coexists with 
the existing public `Column.defaultValue` constructors and with old 
serialized/catalog metadata that contains only that field. No reader, copy 
path, schema-change event, or connector may reinterpret an existing value 
merely because the new model exists.
   > 2. **Supported semantic vocabulary.** A closed, canonical expression 
catalog is preferable to free text, but it must state the v1 values exactly. 
Typed literals, `NULL`, current-time expressions, sequences, generated 
expressions, and arbitrary source functions have different portability and 
safety properties. An unrecognized expression must never be quoted as a literal 
or emitted as target SQL by guesswork.
   > 3. **Unsupported-target policy.** Please decide per path whether an 
unmappable default blocks the schema operation, is explicitly skipped under a 
user-selected policy, or has another observable outcome. A silent `WARN` plus a 
changed target schema is not sufficient because it changes data semantics.
   > 4. **Scope and proof.** Pick one initial source/target pair and one path 
first (auto-create or CDC schema evolution), then specify an executable matrix 
for literal strings/escapes, numeric and decimal values, `NULL`, temporal 
expressions, add/modify events where applicable, target-side DDL inspection, 
and schema-event replay/restore. Cross-database expression mapping can be a 
later, separately approved phase.
   > 
   > [@sainimal1ba-hue](https://github.com/sainimal1ba-hue), thank you for 
offering to help. Please do not open an implementation PR yet. A focused 
STIP/design that resolves the four boundaries above is the useful next 
contribution; after consensus we can define a small first implementation slice 
without mixing incompatible source and target semantics.
   
   @DanielLeens makes sense, no implementation PR from me. I'll draft the STIP 
instead: dev@ post first, then a [STIP-N] design issue linked back here so the 
thread stays in one place.
   
   Quick positions so the draft isn't starting from zero, happy to be argued 
down on any of them: the typed model sits next to the existing defaultValue 
Object and nothing reinterprets old metadata; closed catalog for v1 (typed 
literals, NULL, current-time family), anything else stays unmapped/verbatim and 
never gets guessed into SQL; unmappable default fails the schema op by default 
with an opt-in to drop and continue, no silent WARN; scope is MySQL source, 
auto-create path, one JDBC target with the test matrix, CDC evolution and 
cross-db mapping later.
   
   Also want the STIP to settle up front whether the converter owns the DDL 
text or the actual target behaviour, timestamps are where those diverge. 
@corgy-w  your milestone list will be the skeleton.


-- 
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