Re: [PATCH 01/14] tools lib traceevent: Get rid of malloc_or_die() in show_error()

[Namhyung Kim <namhyung@kernel.org>]

On Mon, 9 Dec 2013 14:23:50 -0500, Steven Rostedt wrote:
> On Mon, 9 Dec 2013 16:14:39 -0300
> Arnaldo Carvalho de Melo <acme@ghostprotocols.net> wrote:
>
>> Em Mon, Dec 09, 2013 at 02:03:42PM -0500, Steven Rostedt escreveu:
>> > On Mon, 9 Dec 2013 15:30:09 -0300
>> > Arnaldo Carvalho de Melo <acme@ghostprotocols.net> wrote:
>> > 
>> > 
>> > > > +		error = malloc(MAX_ERR_STR_SIZE);
>> > > > +		if (error == NULL) {
>> > > > +			/* no memory */
>> > > > +			*error_str = "failed to allocate memory";
>> > > > +			return;
>> > > 
>> > > Can *error_str point to either malloc'ed or constant strings? Who
>> > > releases the allocated memory?
>> > > 
>> > 
>> > Good question. Perhaps we should have a flag that states if the string
>> > is allocated or not. Or better yet, since the only reason it would be
>> > pointing to a static string is if the string for error_str itself
>> > failed to allocate. Then we could use a string within pevent for it:
>> > 
>> > static char *pevent_failed_error_alloc = "failed to allocate memory";
>> > 
>> > Then in the freeing of error str:
>> > 
>> > void pevent_free_error_str(error_str)
>> > {
>> > 	if (error_str != pevent_failed_error_alloc)
>> > 		free(error_str);
>> > }
>> 
>> That is a possibility, yes, then any other routine that works in such a
>> way could check against this string, but what is wrong with returning a
>> value to that function and checking against < 0?
>
> Then everyone has to check if show_error() failed. Then report a bug if
> it did. Egad, then we need to check if that error function failed, and
> then that one and that one and that one :-)

What about returning error code rather than string?  This way we won't
worry about the allocation of the error string itself.

But the downside of it is loosing a positional info of the error.
Hmm.. what about using a static buffer in pevent for it then?

Thanks,
Namhyung