[llvm-commits] [PATCH] update getPointerTo to handle multiple address spaces

[Villmow, Micah]

> -----Original Message-----
> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
> bounces at cs.uiuc.edu] On Behalf Of Duncan Sands
> Sent: Tuesday, October 30, 2012 12:46 AM
> To: llvm-commits at cs.uiuc.edu
> Subject: Re: [llvm-commits] [PATCH] update getPointerTo to handle
> multiple address spaces
> 
> Hi Micah,
> 
> On 29/10/12 22:48, Villmow, Micah wrote:
> > I would prefer to wait on doing any major refactoring. This is way
> outside of the original scope of the work that I had original
> planned/agreed on doing. This original scope of the work was basically:
> > 1) Move TargetData to DataLayout.
> > 2) Update all clients.
> > 3) Add in support for multiple address spaces to DataLayout.
> > 4) Update clients to add in support for multiple address spaces having
> varying pointer sizes.
> >
> > Now, ChrisL wants DataLayout removed from being a pass and moved into
> the Module(Bug11936), and now you want a refactor of the pointer types.
> Now, after having dealing with this for the last few months, I agree,
> the code is not as clean as it could be, but stopping everything just to
> re-factor, I don't agree with, when most of the changes have already
> gone in.
> >
> > Otherwise I'll just be stopping one set of work when it is almost done
> to move on to another project that again is not guaranteed to not be
> affected by the same scope creep. LLVM 3.2 branches in about 2 weeks,
> and as it currently is, multiple address space support is not complete,
> but almost finished. I would highly prefer to get it working by 3.2 and
> then refactor the code that has shown to be problematic.
> 
> personally I would rather see the whole thing reverted rather than
> having a poor implementation in 3.2.  At least the API should be correct
> (even if not all bugs are shaken out), I will try to give you a hand
> with this (i.e. cleaning up your points 3 and 4).  I think changing
> DataLayout from being a pass to something
> else can wait until after 3.2.   I'm not sure what you mean by Chandler
> wanting
> a refactor of the pointer types.  
[Villmow, Micah] Maybe this was a misunderstanding on my part.
If you mean changing your datalayout
> methods to assert when not passed a pointer (as opposed to using an
> address space of 0) then I agree with Chandler that it is essential to
> fix this ASAP.  The more you add places that make use of the current
> bogus behaviour the harder it will get to fix later.  This is the main
> thing I will try to help you out with.
[Villmow, Micah] I have a patch coming, just trying to get rid of the last few places where the assertion is being triggered in the code.
> 
> Ciao, Duncan.
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits