FWD » Core Development » Database

Symptom¶

A schema ASSIGN trigger that declares an OLD value but no NEW value, and that also references the table buffer, converts to an assign(...) method whose parameter list contains the old-value initialization statement. The generated class does not compile.

Seen in testcases (data/trg/order-order-num-a.p, data/trg/orderline-order-num-a.p):

  [javac] src/com/goldencode/testcases/data/trg/OrderOrderNumA.java:39: error: <identifier> expected
  [javac]    public void assign(OrderOrderNumA.this.oldVal = TypeFactory.initInput((integer) _oldVal), final Order_.Buf _order_, final BaseDataType _oldVal)

Reproduction¶

  trigger procedure for assign of order.orderNum old oldVal like order.orderNum.

  if support.db.DatabaseTriggers:fireAssign('order', 'orderNum', string(order.orderNum), string(oldVal))
  then
      return no-apply.

A trigger that declares both NEW and OLD and does not read the buffer converts correctly (e.g. data/trg/order-customer-num-a.p):

  public void assign(final Undoable _newVal, final BaseDataType _oldVal)
  {
     OrderCustomerNumA.this.oldVal = TypeFactory.initInput((integer) _oldVal);
     OrderCustomerNumA.this.newVal = TypeFactory.initInputOutput((integer) _newVal);
     ...

Root cause¶

rules/convert/variable_definitions.rules attaches the promoted OLD/NEW value initialization to the method block, and finds that block as the first child of the method_def:

  <action>meth_block_id = getAst(lastid).getAncestor(-1, java.method_def).getChildAt(0).id</action>

When the trigger references its table, a buffer parameter (final Order_.Buf _order_) is emitted as the first child of the method_def, ahead of the block, so the initialization is grafted into the parameter list. The line was introduced with the trigger-parameter rework of 2024-12-18 (rev 15623.1.5) and is present in trunk.

Fix¶

Look the block up by type, as other rules do (e.g. output_parameters.rules, control_flow.rules):

  <action>meth_block_id = getAst(lastid).getAncestor(-1, java.method_def).getImmediateChild(java.block, null).id</action>

Fix is in 11913a/rev. 16772. Please review.

That's a nice catch!

Indeed, the AST node for attaching the code for initialisation of the old value is not the very first child of METHOD_DEF (which in this case is the LPARENS for parameter declaration) but the BLOCK/BOGUS where the record buffer for target buffer is also instantiated.

A single code style issue: the current line is 138 chars long, well beyond the 110 character limit.

I completed testing and found just one issue; the session triggers require another form of identifying the target body; because in their case there was no proper .getAncestor(-1, java.method_def).getImmediateChild(java.block, null) resolution. Check 11913a/rev. 16773.

Another nice catch!. The fix makes sense.

The fix is regression tested across several examples of schema triggers and session triggers in different combinations.
I will proceed with merging.

Branch 11913a was merged into trunk as rev. 16772 and archived.