[PATCH xorg-gtest 06/16] xserver: move Terminate and Kill handling here

[Chase Douglas]

On 07/05/2012 05:50 PM, Peter Hutterer wrote:
> On Thu, Jul 05, 2012 at 02:09:13PM -0700, Chase Douglas wrote:
>> On 07/03/2012 10:42 PM, Peter Hutterer wrote:
>>> On Tue, Jul 03, 2012 at 10:33:50AM -0700, Chase Douglas wrote:
>>>> On 07/02/2012 11:44 PM, Peter Hutterer wrote:
>>>>> Signed-off-by: Peter Hutterer <peter.hutterer at who-t.net>
>>>>> ---
>>>>>   include/xorg/gtest/xorg-gtest-xserver.h |   13 ++++++++++++
>>>>>   src/environment.cpp                     |   31 +++-------------------------
>>>>>   src/xserver.cpp                         |   34 +++++++++++++++++++++++++++++++
>>>>>   3 files changed, 50 insertions(+), 28 deletions(-)
>>>>>
>>>>> diff --git a/include/xorg/gtest/xorg-gtest-xserver.h b/include/xorg/gtest/xorg-gtest-xserver.h
>>>>> index 52a2fd0..821b01f 100644
>>>>> --- a/include/xorg/gtest/xorg-gtest-xserver.h
>>>>> +++ b/include/xorg/gtest/xorg-gtest-xserver.h
>>>>> @@ -53,6 +53,19 @@ class XServer : public xorg::testing::Process {
>>>>>       void Start(std::string &program);
>>>>>
>>>>>       /**
>>>>> +     * Terminates this server process. Will signal the server to terminate
>>>>> +     * multiple times before giving up.
>>>>> +     *
>>>>> +     * @return false if the server did not terminate, true otherwise
>>>>> +     */
>>>>> +    bool Terminate(void);
>>>>> +
>>>>> +    /**
>>>>> +     * Kills the server. With a vengeance.
>>>>> +     */
>>>>> +    bool Kill(void);
>>>>
>>>> We don't need to recreate these functions. We've already inherited
>>>> them from xorg::testing::Process. Those implementations should work
>>>> automatically if we set up the XServer class properly.
>>>
>>> The implementation is different to the one in Process, this one does the
>>> server-specific bits like waiting for the process to shut down and then
>>> complaining to the log.
>>
>> I read too quickly to realize the difference between the
>> Environment/XServer implementation and the Process implementation,
>> but I'm wondering if we should just move the extra stuff to the
>> Process implementation. Then we won't need to do any redefinition or
>> overriding.
>
> I think we should keep Process basic. For most processes failing to kill it
> should be an issue in itself, it's just the server that takes too long to
> shut down and needs special handling.

I thought about that too. I think the question to ask is: do we think 
most users or subclasses of Process are going to want wait for the 
process to really die. The Process::Kill and Terminate methods just try 
once to raise the signal to the child process. They return true if the 
signal was sent, but don't check what happens after that. Perhaps a 
better approach to resolving this issue would be to add second versions 
of these methods that would take a timeout value and return true only if 
the signal was successfully sent and the process died.
KillAndCheck(timeout) and TerminateAndCheck(timeout) maybe?

-- Chase