Interrupt Handling fails on Linux

[simonjwright]

As reported in issue #4, the Interrupt Handling example, which is supposed to catch and report C-c, doesn’t do so on Linux Mint (or, it turns out, on Debian stretch).

Works fine on macOS Catalina.

[simonjwright]

Simple (and less simple) reproducers failed to actually reproduce the issue, so I’ve started a new branch issue_5, with changes that work (on my Debian Stretch VMware Fusion vm).

[AlexProudfoot]

OK. The example now works for me.

alex@mint-vm$ ./interrupt_handling-harness

Interrupt_Handling.Harness.
---------------------------

This program demonstrates handling a Signal, which is the
nearest thing to an interrupt available in standard OSs.

On Mac OS X and Linux systems, run the program and type C-c
(ctrl-c) to see the interrupt being reported.

On Windows, this doesn't work from a Cygwin Bash shell;
instead, run from a Command Shell started from Start/Run.

entered state IDLE
Interrupt_Handling.Device.T running
Interrupt_Handling.Device.T started
^CInterrupt_Handling.Device.T released
entered state INTERRUPTED
entered state IDLE
entered state IDLE
entered state IDLE
^CInterrupt_Handling.Device.T released
entered state INTERRUPTED
entered state IDLE
entered state IDLE
^CInterrupt_Handling.Device.T released
entered state INTERRUPTED
entered state IDLE
entered state IDLE
^CInterrupt_Handling.Device.T released
entered state INTERRUPTED
entered state IDLE
^Z
[1]+  Stopped                 ./interrupt_handling-harness

Did you find out what was causing the problem?

[simonjwright]

Did you find out what was causing the problem?

No; in gdb, I could see the interrupt triggering the task blocked on the PO, but then nothing happened.

[AlexProudfoot]

Hmmm. I guess this would be something to report to AdaCore.

[simonjwright]

Situation as of 2022-04-09: on Debian 11, macOS x86_64 the original (unfixed) demo works with GCC 10 but fails with GCC 11.

On macOS x86_64 and aarch64, it fails with GCC 12.

In all these cases, the simple Attach_Handler version works; and, this is the approach that anyone working on an embedded platform is going to be using, I think.

Not sure this demo is worth preserving in its present form. I certainly wouldn’t want to present it to AdaCore without some better idea of what’s going on.

[simonjwright]

Well, the problem appears to happen when the handling PO is allocated rather than being declared directly; this works

   type Handler is new Ada.Finalization.Limited_Controlled with record
      Handling_PO        : Handling;
      Attached_Interrupt : Ada.Interrupts.Interrupt_ID := 0;
      Old_Handler        : Ada.Interrupts.Parameterless_Handler;
   end record;

provided you use ’Unrestricted_Access:

   procedure Attach (H  : in out Handler;
                     To :        Ada.Interrupts.Interrupt_ID) is
   begin
      Ada.Interrupts.Exchange_Handler
        (Old_Handler => H.Old_Handler,
         New_Handler => H.Handling_PO.Trigger'Unrestricted_Access,
         Interrupt   => To);
      H.Attached_Interrupt := To;
   end Attach;

Now I might be able to produce a demo!