portable FastCriticalSection implementation

[Gennady Proskurin]

patch attached.
It may have benefits on win32, the old implementation used sleep(1) in "fast" critical section (Thread::yield).
Now Thread::yield seems to be unused, I think it should be removed, it's dangerous. Incorrectly using scheduler's yield may lead to unnecessary cpu consumption.

I tested only unix after patching this.

[Steven Sheehy (steven-sheehy)]

Gennady, these patches you've submitted modify both the windows and linux core, so they would need to be reviewed by the dcplusplus team. We cannot apply changes to linuxdcpp without them also being applied to upstream dc++. The only one that is linux specific is bug #617021, which me or Razzloss can review.

However, your patches/branch are for linuxdcpp which is not up to date with the latest dc++ core. You may have to redo your patches for the dc++ core. But before you do that, I would suggest waiting on a response from the dc++ team to see if they would be interested in your patches before you go through that effort.

Poy et al, can you take a look at these patches that Gennady submitted (this and the other bugs he filed)? They probably won't apply, but you can get an idea from the diff and if it's something that you approve of either Gennady or I can port it to the new core.

[Big Muscle (bigmuscle)]

I completely removed CriticalSection stuff from StrongDC++ and use only following typedefs:

typedef boost::recursive_mutex CriticalSection;
typedef boost::detail::spinlock FastCriticalSection;
typedef boost::lock_guard<boost::recursive_mutex> Lock;
typedef boost::lock_guard<boost::detail::spinlock> FastLock;

[Gennady Proskurin (gpr)]

I agree, these typedefs are better than my patch.
The only drawback I see is that recursive_mutex and lock_guard are not garanteed to be header-only.

[Steven Sheehy (steven-sheehy)]

Will be integrated with next core upgrade.

[poy (poy)]

Fixed in DC++ 0.780.