Re: Safety of unprotected concurrent operations on constant objects

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Tue, 13 May 2014 17:46:58 +0100, Simon Wright wrote:

> Brad Moore <brad.moore@shaw.ca> writes:
> 
>> That may be why Ada has PO's in the standard, but does not provide any
>> mutex libraries like the one you suggest; They are too error
>> prone. I'd be surprised if mutexes and scope locks like the one in
>> your example wouldn't have been considered for inclusion in the
>> standard at some point during the history of Ada.  Ada does provide
>> the building blocks that programmers can create such abstractions
>> however, which programmers are then free to use as they wish.
> 
> I've several times seen Randy recommend use of the Containers as a way
> of avoiding problems with memory allocation/deallocation, and I'd have
> thought simple mutexes/locks would be just as much candidates to support
> the general user.
> 
> Are they error-prone? If Joe Programmer reinvents locking, what's the
> chance that there's something wrong with it?

Usually to mutex' seize and release there are other procedures and entries
added, e.g. when mutex is used with a container like a queue. So it is not
much advantage having them in the standard library. Then there are mutex
variations like reentrant mutexes, read-write mutexes, arrays of mutexes,
ordered mutexes (in order to prevent deadlock), global (system-wide) named
mutexes etc. This is why I am sceptical that mutexes must be in the
standard library.

Much more useful IMO would be atomic operations like atomic increment,
compare and swap, and a requirement to support pragma Atomic for every
scalar type.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de