[llvm-commits] Trivial warning

[Roman Divacky]

On Fri, Nov 16, 2012 at 11:38:07AM -0600, Hal Finkel wrote:
> ----- Original Message -----
> > From: "Benjamin Kramer" <benny.kra at gmail.com>
> > To: "Joe Abbey" <jabbey at arxan.com>
> > Cc: "Hal Finkel" <hfinkel at anl.gov>, llvm-commits at cs.uiuc.edu
> > Sent: Friday, November 16, 2012 10:00:47 AM
> > Subject: Re: [llvm-commits] Trivial warning
> > 
> > 
> > On 16.11.2012, at 13:11, Joe Abbey <jabbey at arxan.com> wrote:
> > 
> > > On Nov 16, 2012, at 12:38 AM, Joe Abbey <jabbey at arxan.com> wrote:
> > > 
> > >> Hal (et al),
> > >> 
> > >> One of my build bots reported a warning:
> > >> PPCRegisterInfo.cpp:501:51: warning: cast from type ???const
> > >> llvm::MachineFunction*??? to type ???llvm::MachineFunction*??? casts
> > >> away qualifiers [-Wcast-qual]
> > >> Index: lib/Target/PowerPC/PPCRegisterInfo.cpp
> > >> ===================================================================
> > >> --- lib/Target/PowerPC/PPCRegisterInfo.cpp	(revision 168110)
> > >> +++ lib/Target/PowerPC/PPCRegisterInfo.cpp	(working copy)
> > >> @@ -498,7 +498,8 @@
> > >>      } else if (CRSpillFrameIdx) {
> > >>        FrameIdx = CRSpillFrameIdx;
> > >>      } else {
> > >> -      MachineFrameInfo *MFI = ((MachineFunction
> > >> &)MF).getFrameInfo();
> > >> +      MachineFrameInfo *MFI =
> > >> +        (const_cast<MachineFunction &>MF).getFrameInfo();
> > >>        FrameIdx = MFI->CreateFixedObject((uint64_t)4, (int64_t)-4,
> > >>        true);
> > >>        CRSpillFrameIdx = FrameIdx;
> > >>      }
> > >> 
> > >> Ok to commit?
> > > 
> > > Past Joe,
> > > 
> > > Make sure your current build is setup to do PowerPC builds.
> > > 
> > > Also, swapping a warning for an error is a horrible trade??? let's
> > > try that patch again:
> > > 
> > > Index: lib/Target/PowerPC/PPCRegisterInfo.cpp
> > > ===================================================================
> > > --- lib/Target/PowerPC/PPCRegisterInfo.cpp	(revision 168110)
> > > +++ lib/Target/PowerPC/PPCRegisterInfo.cpp	(working copy)
> > > @@ -498,7 +498,8 @@
> > >      } else if (CRSpillFrameIdx) {
> > >        FrameIdx = CRSpillFrameIdx;
> > >      } else {
> > > -      MachineFrameInfo *MFI = ((MachineFunction
> > > &)MF).getFrameInfo();
> > > +      MachineFrameInfo *MFI =
> > > +        (const_cast<MachineFunction &>(MF)).getFrameInfo();
> > >        FrameIdx = MFI->CreateFixedObject((uint64_t)4, (int64_t)-4,
> > >        true);
> > >        CRSpillFrameIdx = FrameIdx;
> > >      }
> > 
> > While this would be the right way to silence the warning, the whole
> > thing looks like a horrible hack to me. Shouldn't the FrameIdx
> > generated in a place where MF isn't const?
> 
> Roman, you had looked at these const issues, right? Do you have an opinion on this?

I actively avoided any const_cast<>s in my const rage :)

This code was contributed by wschmidt@ and iirc it needs to be
like this because PPC is special (addressing against fp).

I think Bill didnt see an easy way to fix this. I think const_cast<>
in here is better than C style cast.

Roman