Re: [PATCH] fully enable rpctrace to trace multitask programs.

[olafBuddenhagen]

Hi,

On Mon, Jul 06, 2009 at 02:26:10PM +0800, Da Zheng wrote:
> olafBuddenhagen@gmx.net wrote:

> Since you think it's not proper to wrap the first thread at the place
> where trace exec_startup_get_info() is called, I now wrap the first
> thread of a task when the first RPC request comes.

Actually, I wasn't quite sure whether it's proper -- I just don't
understand the implications well enough. But if you found a safer
approach, all the better :-)

>>> In that case, I have to always use error() to check whether RPCs
>>> return  successfully?
>>
>> Well, strictly speaking you should.
>>
>> However, as rpctrace already uses it wrongly all over the place, I'm
>> not sure what the best approach is. I tend to think it's still better
>> to do it right in the new code...
>>   
> I think we can do it in this way. I can keep all assert_perror() for
> now  and provide another patch to replace them with error().

I actually meant that it's probably better to use error() from the
beginning in all code you *add*, and keep assert_perror() in preexisting
code for now... An extra patch could "fix" the existing code.

-antrik-