svn commit: r294778 - in head: lib/libc/sys sys/kern

[Kubilay Kocak]

On 26/01/2016 10:38 PM, Bruce Evans wrote:
> On Tue, 26 Jan 2016, Kubilay Kocak wrote:
> 
>>> Log:
>>>   Restore flushing of output for revoke(2) again.  Document revoke()'s
>>>   intended behaviour in its man page.  Simplify tty_drain() to match.
>>>   Don't call ttydevsw methods in tty_flush() if the device is gone
>>>   since we now sometimes call it then.
>>> ...
>>>   This was first broken then fixed in 1995.  I changed only the tty
>>> ...
>>
>> Seems like
>>
>>>   This was next broken in 1997 then fixed in 1998.  Importing Lite2 made
>>> ...
>>
>> A fantastic
>>
>>>   This was next broken in 2008 by replacing everything in tty.c and not
>>> ...
>>
>> Regression test candidate :)
>>
>>>   It is now possible to fix this better using the new FREVOKE flag.
> 
> Regression tests for devices are difficult to write and more difficult
> to run.  Simpler for ttys than for networking or disks, but you still
> need at least 2 generic tty ports just to test things that are not
> very related to hardware.

Of course, though perhaps mocking external subsystems/components is
something we could do, as is relatively standard for testing only the
code you want to cover, rather than (requiring) the entire integration.

> Bugs in flushing and draining are sometimes obvious by observing if
> echo 123 >/dev/ttyXx works when it should fail or fails when it should
> work.
> 
> For more arcane bugs, I use the old NIST POSIX test suite.  This is
> badly written and hard to use and not very complete, but it finds about
> 30 regressions between FreeBSD-5 and FreeBSD-9.  30 over-counts for error
> cascades but undercounts for blocking and some other timing bugs, and
> of course strict POSIX tests don't get near FreeBSD features like
> revoke() or bidrectional devices.
> 
> Bruce