Walter is right about transitive readonly - here's the alternative

[Steven Schveighoffer]

"Sean Kelly" wrote
>> My implementation wasn't re-entrant for write locks. I never found it
>> necessary. That said, I know that it is possible to make re-entrant
>> write locks, but probably not by the means that you suggest.
>
> It's possible but to do so I think you'd have to maintain a list of which 
> threads held which lock.  This doesn't seem terribly efficient, and just 
> not worthwhile for something that is of marginal utility.

Since write-lock is exclusive, you only need to know a single thread which 
holds the lock.  The thread ID can be stored with the lock, so it is easy to 
tell/efficient.  Basically (in lock() method):

if(locked_id != my_id)
{
  lockSysMutex();
  locked_id = my_id;
  _reentrant++; // initialized to 0
}

and in unlock():

if(--_reentrant == 0)
{
  locked_id = 0;
  unlockSysMutex();
}

I'd disagree that it is of marginal utility ;)  but that's my opinion.  I 
use it a lot in C# which has reentrant write locks.

-Steve