User interruption handling

Allow the end user to cancel a dialog or a long running procedure.

Basics

To properly understand how Genero handles user interruption, we first need to identify the different events that can occur for a Genero process, and how they are triggered:

The Unix "interrupt signal" (SIGINT): Can be triggered by a kill -s INT command, or by a CTRL-C key in a terminal (see stty -a)). To detect the CTRL-C key on Microsoft Windows platforms, the concept is a bit different and managed by Console Control Handlers. Note that a Unix SIGINT signal can occur in TUI or GUI mode.
The Unix "quit signal" (SIGQUIT): Can be triggered by a kill -s QUIT command, or by a CTRL-\ key in a terminal (see stty -a)). Like SIGINT, a SIGQUIT signal can occur in TUI or GUI mode on Unix.
The 'cancel' (and 'close') predefined actions of singular dialogs, when the runtime system is waiting for a user action.
The asynchronous user interruption event, fired by a action view (button) with the name 'interrupt', when the runtime system is in processing mode (FOR loop, SQL query)

If the program executes an interactive instruction, the GUI front-end can send action events triggered by the user such as 'accept' and 'cancel'. But when the program performs a long process like a loop, a report, or a database query, the front-end has no control. In the second case, user interruption can be detected by using the int_flag register and the 'interrupt' action name.

Tips for SIGINT and SIGQUIT signals

Typical Genero programs should use DEFER INTERRUPT and DEFER QUIT, otherwise any SIGINT or SIGQUIT signal will just stop the process.

Sending a SIGINT when a program is in a SLEEP will cancel the sleep, because this is a system call that is interruptible.

In GUI mode, one can still send SIGINT signals to the fglrun process, but that's not good practice and will not stop the current dialog as in TUI mode.

Detecting dialog cancellation

When executing a singular dialogs such as INPUT BY NAME, the runtime system creates automatically the 'accept' and 'cancel' predefined actions, to let the user validate or reject the dialog. When executing a multiple dialog, there is no predefined 'cancel' action created, and int_flag is not set, even for user-defined ON ACTION cancel handlers.

With singular dialogs, the int_flag register is typically tested after the dialog block (or in the AFTER INPUT block for example), to detect if the user wants the dialog input to be validated or rejected. Note that int_flag is not reset to FALSE when the automatic 'accept' action is fired. Therefore, it is good practice to initialize int_flag to FALSE, before starting a singular dialog.

In TUI mode, to cancel a singular dialog, user presses the CTRL-C key in the terminal, which produces a SIGINT signal and leaves the dialog instruction, after setting int_flag to 1/TRUE. Since the trigger to cancel a singular dialog in TUI mode is the SIGINT signal, the dialog can also be canceled by sending the SIGINT signal to the fglrun process. This is however not good practice. Dialog cancellation with SIGINT is supported for legacy TUI applications.

In GUI mode, to cancel a singular dialog, user fires the 'cancel' (or 'close') predefined action, to leave the dialog instruction, and int_flag is set to 1/TRUE. There is no interrupt signal or interrupt event used in this case: This is triggered by a regular GUI action event.

Detecting asynchronous user interruptions

To detect user interruptions coming from a GUI front-end while a program is in processing mode, define an action view with the name 'interrupt':

BUTTON sb: interrupt, TEXT="Stop";

When the runtime system takes control to process program code or execute a long running SQL query, the front-end automatically enables the local 'interrupt' action to let the user send an asynchronous interruption request to the program.

A program (the runtime system) can also receive a SIGINT interruption signal from the operating system. The interruption request that comes from the front-end is a different source. However, the runtime system handles both type of interruption events the same way.

When receiving an interrupt event from the front-end (from the particular 'interrupt' action), or from the system (SIGINT) the runtime system sets the int_flag register to TRUE.

Consider using DEFER INTERRUPT and test the int_flag register to properly handle user interruptions, and avoid immediate program termination. If the DEFER INTERRUPT instruction is not used, the program will stop immediately when an interruption event is caught. With

DEFER
INTERRUPT

, the program continues, and can test int_flag to check if an interruption event occurred. It is good practice to reset int_flag to FALSE after detecting interruption:

WHILE ...
   IF int_flag THEN
      LET int_flag=FALSE
      ERROR "Procedure was interrupted by the user"
      EXIT WHILE
   END IF
   ...
END WHILE

SQL queries can be interrupted too, if the target database supports this feature and the OPTIONS SQL INTERRUPT ON instruction is used. However, since the control is on the database server side while the SQL statement is running, it is not possible to execute program code to check int_flag. In order to detect an SQL interruption, check the sqlca.sqlcode register after the query for SQL error -213, indicating that the last SQL statement was interrupted.

Tip:

It is good practice to surround the SQL statement with OPTIONS SQL INTERRUPT ON/OFF and WHENEVER ERROR CONTINUE/STOP (or a TRY/CATCH block): You don't want other SQL statements to be interruptible, and if they fail but should not (for example if the table does not exist), the program must stop.

OPTIONS SQL INTERRUPT ON
WHENEVER ERROR CONTINUE
SELECT COUNT(*) INTO cnt FROM ...  -- Long running SQL statement
WHENEVER ERROR STOP
OPTIONS SQL INTERRUPT OFF
IF sqlca.sqlcode == -213 THEN
   ERROR "Database query interrupted by user"
   ...
END IF

When not using DEFER INTERRUPT, if the program enters in a long running procedure, a button with the action name 'interrupt' will become active. The user can then press that button, and the runtime system will stop the program, since DEFER INTERRUPT is not used. However, this will not happen when a dialog is active, because the 'interrupt' button will be automatically disabled in that context. Such situation can confuse the end user, expecting that the 'interrupt' button can stop the program in any context.

Note that the front-end can not handle interruption requests properly, if the application code generates a lot of AUI traffic, by doing many ui.Interface.refresh() calls. This refresh method should not be called more often than once in one second.

Implementing interruption of a long running SQL query

-- db_busy.per
LAYOUT
 GRID
 {
  Database query in progress...
        [sb            ]
 }
 END
END
ATTRIBUTES
 BUTTON sb: interrupt, TEXT="Stop";
END

MAIN
  DEFINE oc INT
  DEFER INTERRUPT
  DATABASE stores
  OPEN FORM f FROM "db_busy"
  DISPLAY FORM f
  CALL ui.Interface.refresh()
  OPTIONS SQL INTERRUPT ON
  WHENEVER ERROR CONTINUE
  SELECT COUNT(*) INTO oc FROM orders
  WHENEVER ERROR STOP
  OPTIONS SQL INTERRUPT OFF
  IF sqlca.sqlcode == -213 THEN
    ERROR "Database query has been interrupted..."
  END IF
END MAIN