Re: Read-write mutex sometimes fails on deadlock

["Randy Brukardt" <randy@rrsoftware.com>]

I wrote:
...
> If the new aspect Nonblocking is used, the delay is statically illegal. 
> (That *really* should have been the default, but it took us 20 years to 
> figure out the right set of rules -- too late to mandate.)

"New" here means Ada 2020; it was just approved at our most recent ARG 
meeting and isn't yet in the draft RM. So it might not be in your favorite 
compiler for a while.

The aspect is a big win for most things, as it makes turns the "potentially 
nonblocking" rules into static checks, so the portability issues [can you 
use Text_IO in a protected object or not?] and "tripping hazards" 
(Program_Error being raised unexpectedly) are eliminated.

To show how this works, let's look at the Eachus example using the aspect 
Nonblocking:

protected type Mutex
    with Nonblocking is
   entry Write(D: in Data):
   function Read return Data;
private
   The_Data: Data := No_Data;
end Mutex;

protected body Mutex is
   entry Write (D: in Data) is
   begin
      The_Data := D;
      Some_Data := True;
      Print(D); -- Illegal unless Print is declared with Nonblocking = True.
      delay 0.1; -- Illegal.
   end Write;

   function Read return Data is
      return The_Data;
   end Read;
end Mutex;

So this code only will compile if it is guarenteed to work on all 
implementations. (The original code would have raised Program_Error on 
Janus/Ada for the delay, Print may or may not have worked.)

                                       Randy.