Re: [PATCH v2 1/2] tracing/events: make __string() more general

[Li Zefan <lizf@cn.fujitsu.com>]

Steven Rostedt wrote:
> On Wed, 27 May 2009, Li Zefan wrote:
> 
>> Steven Rostedt wrote:
>>> On Tue, 26 May 2009, Li Zefan wrote:
>>>
>>>> This changes __string(..., src) to __string(..., len), thus makes it more
>>>> general, and then block TRACE_EVENT() can use it.
>>>>
>>>> Also introduce __fetch_str(), which is used in TP_assign() and returns
>>>> the address of the string, while __get_str() is used in TP_print().
>>>>
>>>> [ Impact: TRACE_EVENT api change ]
>>> What's the reason for the change?
>>>
>>> Is it to handle NULL pointers? If so, lets change the ftrace.h to handle
>>> it instead.
>>>
>> No, not for this purpose.
>>
>> With __string(..., src), we need to have a source string, but sometimes no
>> source string is available, and what we know is how big the string will be.
>>
>> Take block trace events for example, I want to convert an unsigned char array
>> into a string, after this patch, this can be done this way:
>>
>> TP_STRUCT_entry(
>> 	__string( cmd, rq->cmd_len * 3 )
>> )
>>
>> TP_fast_assign(
>> 	construct_str( cmd )
> 
> What does the construct_str do?
> 

Ah, it's pseudo code.. Should be construct_str(__entry->cmd, rq),
this helper function convert rq->cmd (unsigned char[]) to a string of
hex dump.

No source string is available, so I'm not calling assign_str() in
TP_fast_assign.

> 
>> )
>>
>> TP_prink("%s\n", get_str(cmd))
>>
>> Seems Christoph also runs into this problem while doing his XFS tracing.
>> (and NULL str ptr that I'm not suffering)
>>
>>> This is why I like the Impact line really be a "rational" line.
>>>  "TRACE_EVENT api change" is meaningless to me.
>>>
>> So what should this impact be. :(
> 
> That's the point. I don't know the purpose of the patch except that it 
> changes the API. But changing the API is the means to the end. I'm still 
> confused by what the end is.
>