This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Need doc of debug output produced by -fdump-translation-unit
- From: Jim Wilson <wilson at specifixinc dot com>
- To: Don Lindsay <lindsayd at cisco dot com>
- Cc: Holger Peine <peine at iese dot fraunhofer dot de>, gcc at gcc dot gnu dot org
- Date: 16 Jun 2004 16:50:49 -0700
- Subject: Re: Need doc of debug output produced by -fdump-translation-unit
- References: <Pine.LNX.4.44.0406091534120.8012-100000@dlindsay-lnx.cisco.com>
On Wed, 2004-06-09 at 16:00, Don Lindsay wrote:
> Umm, Holger didn't ask for a way to do that. He wants the information to
> go out, and not come back. I doubt he needs completeness, either: I don't.
Doesn't matter. A one way transfer of info can still be used to subvert
the GPL. In this case, emitting complete info from the front ends but
not having a way to read it back in would make it easy for someone to
use the gcc front ends with a proprietary compiler back end. Of course,
someone has already done this, namely the SGI Open64 compiler, but it
was hard for them to do it and maintain it.
I think the next great hope for getting the policy changed would be
intermodule interprocedural analysis and optimization (IPA/IPO).
Currently, we have cgraph which allows for some intramodule IPA/IPO at
the RTL level, and tree-ssa which allows for some intramodule IPA/IPO at
the tree level. Before long, people will start demanding intermodule
IPA/IPO, and then there will be likely be pressure on the FSF to change
their policy.
Meanwhile, we can only allow incomplete dumps which are not useful to
proprietary compilers. And we can't do much about this, as this is an
FSF policy not a GCC policy.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com