[llvm-commits] [llvm] r92006 - /llvm/trunk/include/llvm/Analysis/LoopInfo.h

David Greene dag at cray.com
Wed Dec 23 10:39:00 PST 2009


On Wednesday 23 December 2009 12:20, Chris Lattner wrote:
> On Dec 23, 2009, at 10:14 AM, David Greene wrote:
> > On Wednesday 23 December 2009 11:52, Chris Lattner wrote:
> >> Please pull the dump() implementation out of line.
> >
> > How?  LoopBase is a template.  I could go through various contortions
> > to produce a .ipp file with template implementation code and then
> > #include that everywhere LoopInfo.h is included, but is it worth it?
>
> Ugh, that's unfortunate.  Lets just remove dump() from loopinfo, I
> don't think anyone uses it in practice anyway.

LoopStrengthReduce.cpp.

I agree with you that conflicting definitions of DEBUG are problematic.
We ran into that here a long time ago.  I would like to get these issues
fixed but it either involves generating .ipp files and puting an additional
#include burden on clients or redesigning the classes themselves.  In both
cases, the code owner should probably decide what to do.

I don't want to go back to errs() for cases like this because it will present
inconsistent debug output.  And I don't want to remove dumps from clients
that use them for debugging purposes.

Hmm, what to do?  I'm ok with leaving the "bad" include of Debug.h in for now
with the understanding that the code owners will fix the problems.  We've
dealt with conflicting DEBUG for some time now and a little more workaround
work here isn't too big a burden for us.

                               -Dave



More information about the llvm-commits mailing list