POSIX message queues, libmqueue: mq_open, mq_unlink
Quoted 6 lines · Alex
Ok. It's just that every provider of the _kernel_ interface to user space has now to take care of being posix-compliant. Write the code for checks, iow. That is not the case for "open", for instance. And besides, with the patch applied the kernel is also posix compliant, isn't it?
No. E.g. mq_notify(,&{.sigev_notify=SIGEV_THREAD) cannot be implemented in kernel space. And sys_mq_getsetattr isn't posix compliant either - the user space library must implement mq_getattr and mq_setattr on top of the kernel API. The kernel API was designed to be simple and flexible. Perhaps we want to extend the kernel implementation in the future, and then a leading slash could be used to indicate that we are using the new features.
--
Manfred
Manfred Spraul, Sat, Apr 17, 2004 12:49:08 +0200:
Quoted 8 lines
Quoted 6 lines · Alex
Ok. It's just that every provider of the _kernel_ interface to user space has now to take care of being posix-compliant. Write the code for checks, iow. That is not the case for "open", for instance. And besides, with the patch applied the kernel is also posix compliant, isn't it?
Quoted 4 lines
No. E.g.
mq_notify(,&{.sigev_notify=SIGEV_THREAD)cannot be implemented in kernel space. Andsys_mq_getsetattrisn't posix compliant either - the user space library must implementmq_getattrandmq_setattron top of the kernel API.
Ok. It's inevitable, as it looks.
Quoted 3 lines
The kernel API was designed to be simple and flexible. Perhaps we want to extend the kernel implementation in the future, and then a leading slash could be used to indicate that we are using the new features.
and userspace will need to be updated (the slashes are cut off in libmqueue)