Implement the new ExceptionHandling specification

[killenb]

The exception handling specification https://chimeratk.github.io/ApplicationCore/master/exception_handling_design.html already identifies issues where the implementation does not match the spec. It is strongly correlated with the initial value propagation (https://chimeratk.github.io/ApplicationCore/master/spec_initial_value_propagation.html) and the data validity propagation (https://chimeratk.github.io/ApplicationCore/master/spec_data_validity_propagation.html)

The tickets from exception handling can be grouped into 7 "threads". Each of them will become one or more tickets

1. Independent issues (can be addressed separately, maybe even to small for a single ticket if someone is working on the stuff anyway)
   • 11.1
   • 11.3
   • 11.4
   • 11.6
   • 11.11
   • 11.12
2. Read and writes must all block during recovery and initalisation
   • 11.2
   • 11.20
3. New data validity propagation using DataFaultCounter objects
   • 11.5
   • 11.9
   • Write spec for data validity propagation #122
   • Introduce a DataFaultCounter class #106
4. read/writes block when they should not, and don't block when there is no initial value
   • 11.7
   • 11.13
   • 11.17
   • 11.18
5. ThreadedFanout and TriggerFanout must not block on device errors
   • 11.8
   • 11.14
6. Remove special treatment for first opening (is not different from the others)
   • 11.10
7. Exceptions should be thrown in doPostRead/Write, not in doRead/WriteTransfer (also includes changes in DeviceAccess)
   • 11.16
   • 11.19

[killenb]

Status of the implementation as of 2020/04/24

Scope of the past focus week (calendar week 15) and the "rescue" week (calendar week 16)

Not addressed:

• Bi-directional variables
  • No tickets yet
• Initial values in transfer group
  • Ticket needed. Exception handling decorator re-tries full read in postRead(). This is wrong in TransfersGroups. The ExceptionHandlingDecorator has to learn that it is in a transfer group (only highest level knows at the moment) and must throw instead of recovery. An ExceptionHandlingTransferGroup catches this and does the recovery.
  • Quick mitigation: If no valid version number do a waitForRecovery() in preRead(). Like this it is clean for working devices. Broken devices still does not work cleanly.
• DataValidity flag propagation (see Set owner of ExceptionHandlingDecorator #102)
  • TriggerFanOut not set as "owner" of ExceptionHandlingDecorator because a different scheme is needed, wanted to avoid doing twice
    • Introduce a data fault counter class Introduce a DataFaultCounter class #106
    • The trigger fan out must hold one instance of the DataFaultCounter for each variable that is polled (ticket needed)
    • Invalid flag from the trigger should be propagated as well (Trigger fan out should propagate the data invalid flag from the trigger #109)
  • Consuming fanout has no owner set. The original design (Extend the ExceptionHandlingDecorator to have an optional ConsumingFanOut in addition to the owner #108) was too complicated and is not needed any more, now that the first read does not block any more in case of exceptions. (needs ticket)
  • The ThreadedFanout need implementation. There was a ticket (Implement data validitiy flag propagation in ThreadedFanOut #101) which has become obsolete with the DataFaultCounter scheme ConfigReader: allow building hierarchies of configuration variables #16. new ticket needed
• Documantation
• Code improvements without functional changes

Was in the scope but not done yet:

• No transfers must be started while DeviceModule is in recovery (Add startTransfer() and transferCounter to DeviceModule #131, ExceptionHandlingDecorator: for readAsync preRead() must block until the transfer is allowed #157, part of Proper implementation of RecoveryAccessor read operations #138) important

• No "values after construction" must be propagated (part of Proper implementation of RecoveryAccessor read operations #138) important

• Write function that does not block on exceptions (Introduce writeWithoutErrorBlocking() to ExceptionHandlingDecorator #134)

• ThreadedFanOut and TriggerFanOut should use non-blocking write (Introduce writeWithoutErrorBlocking() to ExceptionHandlingDecorator #134 -> ThreadedFanout and TriggerFanout should not block #137)

• First read after exception does not block to transport the data validity (part of Proper implementation of RecoveryAccessor read operations #138) important

• readLatest() and readNonBlocking() should only block if there is no initial value (part of Proper implementation of RecoveryAccessor read operations #138)

• ExceptionHandling in TriggerFanOut not needed any more (Adapt exception handling in TriggerFanOut #142)

• Tests

  • No tested yet: Write tests that only valid recovery accessors are written to the device #145, tests for asynchronous reads with exceptions (not ticket yet)
  • Need to be adapted to changed behaviour: Adapt tests to new exception handling scheme #155, Check tests for initial value propagation #156

[killenb]

The implementation of the new exception handling scheme is in place. All the specification references in this ticket are invalid by now because it has changed too much. There are tickets to write systematic tests with the current spec.
All implementation ticket references are in this ticket are closed, so I also consider this design as closed.