Race conditions in critical.c?

[Fawzi Mohamed]

On 2008-10-01 06:49:11 +0200, Sean Kelly <sean at invisibleduck.org> said:

> Sean Kelly wrote:
>> Bartosz Milewski wrote:
>>> 
>>> If you think this is bizarre bordering on reckless I'm with you.
>> 
>> Yup, it's completely useless given that POSIX only provides guarantees 
>> for visibility, etc, for threads entering the *same* mutex.  So one 
>> couldn't even argue that this approach would work as some strange sort 
>> of memory barrier.
> 
> Oh, and I haven't fixed critical.c for the same reason.  Given how they 
> work, does *anyone* use this feature?
> 
> 
> Sean

I think that it should either throw an exception "unsupported" or be corrected.

it is possible to use it correctly if you have a function that is 
called in parallel with only one critical section, or when the whole 
function is synchronized.

I checked my code and I prefer explicit locks, so indeed I haven't 
used, but both tango and phobos have some synchronized statements.

Fawzi