Re: [PATCH 2/2] tracing/filters: remove error messages

[Tom Zanussi <tzanussi@gmail.com>]

On Thu, 2009-06-18 at 09:17 +0800, Li Zefan wrote:
> Frederic Weisbecker wrote:
> > On Wed, Jun 17, 2009 at 04:47:15PM +0800, Li Zefan wrote:
> >> Now we restore original filter is the new one can't be applied,
> >> and no long show error messages, so we can remove them totally.
> > 
> > Why?
> > These messages are very powerful to point a user to its mistakes
> > in filters syntaxes or semantics.
> > 
> > I really think we are removing a very useful feature in this patch.
> > 
> 
> I think it's better done by providing a user-space program/script.
> 
> So what's the criterion to decide what should be in kernel and
> what should be in user-space?
> 

I thought there wasn't supposed to be any userspace for non-binary
tracing.  If there is, then yeah, you can get rid of the error messages
in the kernel because your userspace program can arrange to never submit
an erroneous filter.

> btw, this feature is not full-fledged, that it can't point to the
> exact position where error occured, and the implementation will add
> complexity and I'm sure it's worthy or not.
> 

It could, it just doesn't yet.  But even without exact position
information, the messages should still be useful in most cases.

Tom

> > May be we can keep the previous filter in case of new filter string
> > inserting failure, though if the user wanted to insert a new one, there
> > are few chances that the previous one is still relevant for him.
> > I don't know.
> > 
> >> Another reason is, I don't think it's good to show error messages
> >> when reading a control file.
> > 
> > So, why not create a filter_error file in this case? One for each
> > event and subsys that would print the last error?
> > 
> 
> Yeah, if we do want to keep this feature in the kernel.