fix(seekdb): DIMA-2026051300116077401

[ep-12221]

Task Description

Fix an issue where the detailed error message (including the object name) is lost when a DML statement referencing a non-existent object (e.g., SELECT t.notable.*) is executed inside a trigger or stored procedure body, resulting in a generic "Unknown table" error.

Solution Description

Root Cause: The regression was introduced by commit 676b4e2dc86 (2026-05-06), which added OB_ERR_BAD_TABLE to the deferred-error list in ObPLResolver::resolve. In the deferred-error branch, the failed statement is replaced with a PL_SIGNAL stub, which only saves the error_code_, ob_error_code_, and sql_state_, but does not save the formatted user message already present in the thread-local ObWarningBuffer. Consequently, at runtime, only the generic template from ob_errpkt_strerror ("Unknown table") is used, losing the specific object name.

Fix Details:

1. src/pl/ob_pl_resolver.cpp: When entering the deferred-error branch, before destroying the failed statement, if wb->get_err_code == save_ret, copy wb->get_err_msg into the PL parsing allocator and store it in a new ObPLSignalStmt::user_msg_ field.
2. src/pl/ob_pl_stmt.h: Added a common::ObString user_msg_ field and corresponding get_user_msg/set_user_msg/has_user_msg methods to the ObPLSignalStmt class.
3. src/sql/ob_spi.{h,cpp}: Added ObSPIService::spi_pl_set_user_error_msg(ctx, err_code, sql_state, msg, msg_len) to write the message (truncated to ObWarningBuffer::WarningItem::STR_LEN) back to the current thread-local warning buffer.
4. src/pl/ob_pl_code_generator.{h,cpp}: Registered the LLVM function in init_spi_service. In visit(ObPLSignalStmt), for signals where has_user_msg is true, generate a call to the new SPI function before generate_exception. The err_code parameter uses s.get_ob_error_code (the OB internal errno, e.g., -5201) to align with the check in send_error_packet.
5. src/pl/ob_pl.cpp: Registered the symbol spi_pl_set_user_error_msg via WRAP_SPI_CALL for LLVM JIT linking.

Behavior Comparison (macOS, mysql-mode):

• Before fix: ERROR 42S02: Unknown table
• After fix: ERROR 42S02: Unknown table 't.notable' (Matches Linux behavior and the expected result in tools/deploy/mysql_test/test_suite/ddl/r/mysql/drop.result:104)

Scope of Impact:

• Only affects the deferred-error branch for specific error codes (OB_ERR_FUNCTION_UNKNOWN, OB_ERR_SP_WRONG_ARG_NUM, OB_ERR_SP_DOES_NOT_EXIST, OB_ERR_GET_STACKED_DIAGNOSTICS, OB_ERR_RESIGNAL_WITHOUT_ACTIVE_HANDLER, OB_ERR_BAD_TABLE) when a matching user message exists in the warning buffer.
• When user_msg_ is empty, all code paths behave identically to before the fix.
• Normal SIGNAL/RESIGNAL paths (with explicit MESSAGE_TEXT) are unaffected.

Passed Regressions

Local Validation Environment: macOS arm64, build_release/src/observer/seekdb, port 12881.

1. Reproduced and verified the fix (focus test):

   mysqltest < focus.test
   ...
   INSERT INTO table1 VALUES (1);
   ERROR 42S02: Unknown table 't.notable' <-- Correctly includes object name after fix
   DROP TABLE table1,table2;
   ok
   

2. Full regression for ddl.drop:

   mysqltest --result-file=tools/deploy/mysql_test/test_suite/ddl/r/mysql/drop.result \
   < tools/deploy/mysql_test/test_suite/ddl/t/drop.test
   ddl.drop: exit=0 -- ok
   

3. Adjacent SIGNAL/RESIGNAL/trigger tests show no new regressions:
   • Running mysql8_t_1/signal.test and mysql8_t_1/signal_demo1.test before and after the patch showed diffs only in timestamps, confirming existing failures are unrelated and no new mismatches were introduced.
   • mysql8_t_1/signal_code.test: exit=0 (ok).
4. Runtime trace self-check (temporary build with DEBUG logs):
   • Parsing phase log: DEBUG: captured saved_user_msg for deferred SIGNAL(save_ret=-5201, saved_user_msg=Unknown table 't.notable') – Confirmed successful capture from warning buffer.
   • Runtime phase log: DEBUG: spi_pl_set_user_error_msg called(err_code=-5201, msg_len=25, ObString(msg_len, msg)=Unknown table 't.notable') – Confirmed LLVM JIT call to SPI with OB internal errno parameter.
   • These DEBUG logs were removed for the final commit.
     Commit: e4be58867b8 pushed to issue/2026051300116077401, awaiting orchestrator execution of ob flow checkin / merge-review / core-test.

Upgrade Compatibility

Other Information

DIMA: 2026051300116077401

Release Note

Fixed an issue where executing a DML statement referencing a non-existent object inside a trigger or stored procedure body resulted in a generic error message (e.g., "Unknown table") instead of the specific message containing the object name (e.g., "Unknown table 't.notable'").

[ep-12221]

The mapping Dima issue is: [macOS] [introduced after 20260331] [mysqltest] Trigger referencing non-existent table 't.notable' reports an error inconsistent with Linux seekdb.