Subj : Re: Can C++ local static objects be made thread safe?
To   : comp.programming.threads
From : David Schwartz
Date : Thu Feb 03 2005 02:13 am


"Gianni Mariani" <gi2nospam@mariani.ws> wrote in message 
news:B6GdnRWW98G7A5zfRVn-2w@speakeasy.net...

> I'm not sure how the GCC compiler implemented this, but I did show some 
> code that requires a mutex check only on the first time throught the code, 
> otherwise there is no check at all.  C++ already checks for prior 
> construction on every pass through the code, this check can be used to 
> also eliminate the need to lock a mutex.  The net effect is virtually 0 
> overhead.

    Your code is not portable. It relies on slippery and complex semantics 
that may or may not be available on future processors. There has been a 
major effort not to heap features onto C or C++ that require specific 
intricate platform details to be supportable. C and C++ are supposed to run 
on elevator controllers, remember?

    The big problem is that you have no way of knowing what optimizations 
might be possible (and vital) in the future and what will be able to do 
those optimizations. It may not be just the compiler and the processor, it 
may be the memory hardware. To design a language feature around things that 
just happen to work on particular platforms today is just terribly bad 
practice, and had it been followed in the early days, C would be a dead 
language today.

    If you want to make C and C++ support threads, then support threading 
primitives and don't add language features that can't be implemented with 
the primitives you support. And for the love of god, don't rely upon code 
that nobody can understand.

    DS

.