tst-aio - new version - reveals a bug in glibc

[Mark Brown]

At 1:32 PM -0700 7/27/00, Ulrich Drepper wrote:
>Mark Brown <bmark@us.ibm.com> writes:
>
>> >Mark: it's probably something that has to be brought up in the AG.
>>
>> The current draft (AG Draft 3) for aio_cancel(), nor the SUSv2 (UNIX98)
>> spec for aio_cancel() mention specific fields.
>>
>> I'm not sure what document you are accessing...
>
>The aio_fsync() man page (in XBDd3) mentions aio_fildes and
>aio_sigevent and explicitly says that all other fields must be
>disregarded.

Please ignore my previous mail, which *almost* said what I wanted to say.

The AD Draft 3 of the XSH book, in aio_fsync(), does say this.

But this is not a "general" statement for all APIs, just a
statement for aio_fsync(), that only the filedes and the signal
are relevant.

This statement does not apply to aio_cancel(), aio_error() etc.

However, this statement has not been retracted or questioned in
1003.1-1996 or anywhere else -- the intent is that an aio_fsync()
happens right when you ask for it, I guess.

[I didn't write the AIO specs, I've just had years of experience
interpreting POSIX wording....]

Mark
-- 
-------------------------
Mark S. Brown                                                   bmark@us.ibm.com
IBM RS/6000 AIX System Architecture                    512.838.3926  T/L678.3926
IBM Corporation, Austin, Texas