Re: [dpdk-dev] Proposal: enable redirection of DPDK logs from the user app

["Montorsi, Francesco" <fmontorsi@empirix.com>]

Hi Olivier,

> On 10/04/2016 02:28 PM, Montorsi, Francesco wrote:
> > Yes, but to be honest, that seems a troublesome solution for something
> > as easy as logging a string; e.g. by using fopencookie() approach, you
> > don't have the concept of "log message", you just provide a function
> > that must write a block of bytes somewhere.
> > Typically instead, you need to know where a log message starts and
> > ends, to e.g., add prefixes/postfixes to it.
> 
> I'm not sure that true if you call setbuf(log_stream, NULL).
> 
> In that case, it looks easy to prefix / postfix messages with a fopencookie
> callback like:
> 
> /* example on stdout */
> ssize_t
> simple_write(void *c, const char *buf, size_t size) {
> 	ssize_t ret1, ret2, ret3;
> 
> 	ret1 = fwrite("<", 1, 1, stdout);
> 	if (ret1 == 0)
> 		return 0;
> 	ret2 = fwrite(buf, size, 1, stdout);
> 	if (ret2 == 0)
> 		return 0;
> 	ret3 = fwrite(">", 1, 1, stdout);
> 	if (ret3 == 0)
> 		return 0;
> 	return ret1 + ret2 + ret3;
> }
> 
I didn't know about setbuf()... but are we sure that in this way the simple_write() function will always receive a full string? I mean: in the manpage for setbuf() it says:

"... When the first I/O operation occurs on a file, malloc(3) is called, and a buffer is obtained. .... If the argument buf is NULL, only the mode is affected; a new buffer will be allocated on the next read or write operation."

But: is it true that 1 write operation corresponds to 1 vfprintf() call? Maybe if you have a "long" a single vfprintf() call may translate to several simple_write() calls... I don't know honestly.

> > Indeed, most of the C/C++ (open source) libraries have some simple
> > hook that allows the user to have more control on logging... I think
> > DPDK should be no exception... :)
> 
> I understand that the current API is a bit more complex, but I don't feel there
> is any blocking issue to do what you want. What do you think?

See above. Moreover, IMHO it would be much more user-friendly to have a simple function callback to implement vs having to dig into fopencookie()+setbuf()+etc etc . 

> Also, I know you've said your patch needs some rework, but as you've also
> said you are using it, maybe it would be useful for you to know:
> - it makes use of a global variable 'log_buffer', shared by all the pthreads,
> which can lead to crashes

That's a good point. I will turn it into a __thread variable. Thanks for pointing out this.

> - it strips the log messages to 4095 chars

Correct, but in my experience DPDK never creates such a long line of log message... 

Francesco