Re: POSIX semaphore for GNU/Hurd

[Michal 'hramrach' Suchanek]

Sending what I have so far. Hopefully I'll be able to read any comments next
wek.

On Thu, Nov 07, 2002 at 10:22:09AM -0500, Neal H. Walfield wrote:
> > > not conform to the GNU coding standards; this will have to be fixed.
It probably still does not, I'll have to read the GCS carefully when
I have time.
> > > You did not implement most of my suggestions in my previous emails on
> > > this subject.
> > It was intended as an improvemenet of the previos code, not complete
> > rewrite :)
> 
> It will not be going in to the main line until it is rewritten.
> 
> >   - it can be used anywhere (modulo cancellation modes) not only on
> >   the Hurd :)
> 
> The cancelation code conforms to POSIX.  What is this suppose to mean?
Setting cancellation options is part of the same option as threads, it should
really work everywhere. I have no idea if it is async-signal safe.
> 
> > I do not understand what is exported about the mutex_t.
> 
> What do you mean?  Can you give more details.

----neal@cs.uml.edu said on 02 Nov 2002 19:34:42 -0500:

I would rather have part of the sem_t structure exported (? la                  
pthread_mutex_t).  The structure should contain an id and a union: the          
former so that we can distinguish between semaphores created with               
sem_init and those with sem_open; and the latter so that sem_open and           
sem_close can store the file descriptor and other relevant                      
information.
----

> 
> > --- libpthread.org/include/bits/semaphore.h 1970-01-01 01:00:00.000000000 
> > +0100
> > +++ libpthread/include/bits/semaphore.h     2002-11-07
> > 14:02:01.000000000 +0100
> 
> > +#define SEM_VALUE_MAX (UINT_MAX - 1)
> 
> This is defined by limits.h, not here.  POSIX says the value may be
> indeterminate and thus the macro not defined.  As such, we will not
> define it.
> 
> > +#define SEM_VALUE_INVALID UINT_MAX
> I could not find a reference to this macro in POSIX.  Could please you
> point me to the location where it is specified.
It should prabably have some underscore(s). But I removed it entirely.
> 
> > +
> > +struct __sem_t 
> > +{
> > +  unsigned int count;
> > +  pthread_mutex_t count_lock;
> > +  pthread_cond_t count_cond;
> > +  char * ID; /* may be used in future to store semaphore name */
> 
> Why do you think this is useful?
I thought it was suggested in one of the emails. Anyway, the union is
more flexible.
> 
> Did you actually test this?  You cannot specify NULL for the oldtype
> argument.
I did not, and now I can not as the tests can no longer be linked. I may
try downgrading my Hurd tree.
> 
> SEM_VALUE_MAX may not be defined.  Conditionally compile this code in.
I expected my implementation defines it when it is used. But some value with
more underscores could be used instead.
> 
> > +/* Destroys the semaphore */
> > +int 
> > +sem_destroy (sem_t *sem)
> > +{
> > +  int res = 0;
> > +  pthread_mutex_lock (&sem->count_lock);
> 
> Why are you bother to lock this?
I'd probably like that if I used the semaphore :)

-- 
Michal Suchanek
hramrach@centrum.cz

semaphore-2.diff
Description: Text document