Add recursive_mutex to the header file. I need this for the LRU cache implementation.

[cykoo1, 2009-11-20 22:56:51]

Let me know if you prefer the implementation looks something like the following
or something along the line. 

class mutex_basic(type) {
  ....
}

class mutex() {
   ...
  private:
     mutex_basic(default);
}

class recursive_mutex() {
   ...
  private:
     mutex_basic(recursive);
}

[Jeffrey Yasskin, 2009-11-20 23:57:13]

http://codereview.appspot.com/157118/diff/1/2
File include/mutex.h (right):

http://codereview.appspot.com/157118/diff/1/2#newcode48
include/mutex.h:48: assert(pthread_mutexattr_init(&mutex_attribute_) == 0);
Consider switching to handle_err_return(pthread...) from posix_errors.h.
try_lock will have to be a little different since result==EBUSY isn't an error.

http://codereview.appspot.com/157118/diff/1/2#newcode50
include/mutex.h:50: PTHREAD_MUTEX_RECURSIVE) == 0);
I do somewhat prefer sharing the basic mutex implementation and just having the
specific types pass up their differences. mutex could pass up
PTHREAD_MUTEX_DEFAULT. (I doublechecked, and the undefined cases in the C++
standard are the same, assuming that violating a precondition causes undefined
behavior.)

Sharing the implementation would also make your small test for the new type more
complete. :)

http://codereview.appspot.com/157118/diff/1/2#newcode86
include/mutex.h:86: pthread_mutexattr_t mutex_attribute_;
I'm pretty sure you can make the mutexattr a local variable in the constructor.

[cykoo1, 2009-11-21 23:21:11]

http://codereview.appspot.com/157118/diff/1/2
File include/mutex.h (right):

http://codereview.appspot.com/157118/diff/1/2#newcode48
include/mutex.h:48: assert(pthread_mutexattr_init(&mutex_attribute_) == 0);
On 2009/11/20 23:57:13, Jeffrey Yasskin wrote:
> Consider switching to handle_err_return(pthread...) from posix_errors.h.
> try_lock will have to be a little different since result==EBUSY isn't an
error.

Done.

http://codereview.appspot.com/157118/diff/1/2#newcode50
include/mutex.h:50: PTHREAD_MUTEX_RECURSIVE) == 0);
On 2009/11/20 23:57:13, Jeffrey Yasskin wrote:
> I do somewhat prefer sharing the basic mutex implementation and just having
the
> specific types pass up their differences. mutex could pass up
> PTHREAD_MUTEX_DEFAULT. (I doublechecked, and the undefined cases in the C++
> standard are the same, assuming that violating a precondition causes undefined
> behavior.)
> 
> Sharing the implementation would also make your small test for the new type
more
> complete. :)

Done.

http://codereview.appspot.com/157118/diff/1/2#newcode86
include/mutex.h:86: pthread_mutexattr_t mutex_attribute_;
On 2009/11/20 23:57:13, Jeffrey Yasskin wrote:
> I'm pretty sure you can make the mutexattr a local variable in the
constructor.

Done.

[Jeffrey Yasskin, 2009-11-22 01:31:53]

Looks good! The only thing I'd change is to rename basic_mutex to something
starting with an underscore to reflect that it's an implementation type. Maybe
_posix_mutex? And/or put it into an "internal" namespace.

[cykoo1, 2009-11-23 02:54:55]

I have renamed basic_mutex to _posix_mutex.

I've put the class basic_mutex into a namespace. Or do you mean something
different when you say "internal" namespace?

[Jeffrey Yasskin, 2009-11-23 04:21:14]

I meant "internal" to be the name of the namespace, but your choice is fine too.
Please commit. :)

[cykoo1, 2009-11-23 18:26:26]

committed and pushed.