Re: Reduce Dwarf Debug Size

[Ian Lance Taylor <iant at google dot com>]

"Devang Patel" <devang.patel@gmail.com> writes:

> > C++ debug info size is a problem for real uses of gcc.  We've got a
> > patch on the table.
> 
> We had patch few years ago that follows more or less what Dan described
> earlier. And it was ignored/rejected in favor of unbuilt system without anyone
> cosidering  "is it useful, does it work, is it well-written." ;)

Do you have a link to the submission to gcc-patches?


> > We don't have to use it.  But while I would love
> > to see a symbol database, I don't see that as a viable alternative for
> > all people.  And it doesn't exist and it's not trivial to implement
> > and nobody is working on it.
> 
> and after couple of years GCC is still here.
> 
> Despite what you said earlier, I understand that you'd like to see
> Lawrence's patch included in mainline GCC in one form or other :)

I wouldn't put it that strongly.  If it is rejected, I would prefer
that it be rejected for what I consider to be a good reason.  (Of
course as a middle-end maintainer I could force it in, but I don't
like to work that way.)


> I do not blame you either :). If you are thinking in this direction
> then one additional thing to consider : "Avoid emitting debug info. that
> GDB does not use it and user can not take advantage of it through
> GDB." For example, C++ constructor/destructor info.

Sounds like a good idea to me.  And an orthogonal one.


> If you have not talked with Apple GDB folks then give them a call.
> They have dealt with seriousness of this issue and found resonable
> approach that works on very large system.

My understanding of Apple's approach is that it wouldn't work for us.
But it may well be that I don't understand it.

Ian