On Mon, 5 Apr 2021 09:59:20 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> On Mon, 5 Apr 2021 12:33:57 +0300
> "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@xxxxxxxxx> wrote:
>
> > The warning() function is used in a lot of places in the trace-cmd
> > library, but there is no implementation. The function is implemented in
> > the trace-cmd application. Added a weak implementation in the library, in
> > case the function in not implemented in the application, using that
> > library.
> >
>
> Isn't the "warning()" function implemented in libtraceevent? That's where
> it would be used as it is weak there.
>
> But honestly, I think we should change the libtraceveent warning to
> "tep_warning()" if we haven't already done so.
>
> /me goes to look at the code.
OK, so what I think we need to do is have this:
void __weak tep_print_error(const char *fmt, const char *app, va_list ap)
{
if (errno)
perror(app);
fprintf(stderr, " ");
vfprintf(stderr, fmt, ap);
fprintf(stderr, "\n");
}
That gets the format and a va_list, and this is the weak function that
anything can overwrite (like KernelShark to have a pop up on error?).
Then we can have functions:
void __weak tep_vwarning(const char *fmt, va_list ap)
{
tep_print_error(fmt, "libtraceevent", ap);
}
void __weak tep_warning(const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
tep_vwarning(fmt, ap);
va_end(ap);
}
And the same for libtracefs and libtracecmd.
where it will have a tracefs_warning() and a tracecmd_warning() functions
defined. Then an app can overwrite how tep_print_error() works, as well as
tep_vwarning works (or tracefs_vwarning() etc).
Hmm, thinking about this, there's no reason to have tep_warning() weak,
because by overwriting tep_vwarning(), you have full control of
tep_warning().
-- Steve
|