[PATCH] Add attribute "artificial"

Daniel Berlin dberlin@dberlin.org
Thu Aug 11 02:08:00 GMT 2005


On Wed, 2005-08-10 at 18:21 -0700, Devang Patel wrote:
> On Aug 10, 2005, at 5:48 PM, Daniel Berlin wrote:
> 
> >>  Stuart has answered all
> >> questions/doubts regarding with ICC does in this situation.
> >>
> >
> > As Stuart has pointed out, it *has no user visible attribute* to do
> > this.
> 
> C'mon Danny. At that time Stuart did not answer any question  
> regarding user visible attributes. It was not a issue at that time.  
> Stuart pointed out that your arguments regarding how ICC behaves were  
> not correct.

I'm happy to admit that i was incorrect about what ICC did.
(However, I did file something with the Intel using premier support, and
they claim they are considering changing it for future versions.  So it
is possible they are having the same argument inside Intel right now :)

> 
> > So how does this help your argument?
> 
> Would this be your last argument in this thread ? :-)

I got to keep in shape somehow :)

> Irrespective of  
> how ICC handles this, I still think it is good to have user visible  
> attribute.

I don't actually disagree with you. 

In fact, I was willing to see what others did with your patch as posted,
until Mark jumped in and restarted this discussion.  :)

My basic concerns boil down to these:

1.  As you can tell, I'm just extremely worried about adding attributes
that affect debug info.  Every attribute we've ever added has been
misused an abused to *our* detriment (IE bugs filed, etc).  Debug info
is a very tricky area for us to get right.  Adding knobs just makes it
that much harder.

while at the same time:

2. Making decisions *for* the users about what it going to get displayed
in their debug info (which is what putting this attribute in the
*compiler header* will do), and having them have to edit compiler
headers to get what they want, bothers me quite a bit.  I'd much rather
provide a way for whatever tool is consuming the information to filter
out things they think users don't want to see.

That's why, at least for DWARF, it was requested that you use the
artificial attribute, if you were going to use it in the compiler
header.

Again, if you can convince Mark or Richard or Jason or Geoff or whoever
that a user visible attribute is right, i won't stand in your way.
It's easy enough for me to auto-assign bug-reports about it to you or
them :):):)




More information about the Gcc-patches mailing list