Re: [CIL users] "const"

[Ben L. <li...@cs...>]

John Regehr wrote:
> separating the declaration and initialization of variables is a bit
> obnoxious

I'd love to see CIL keep declarations and initializations together, and 
simply have a visitor that can be applied to rewrite them as separate 
declarations followed by initializing assignments.  Then it's up to each 
of us to decide whether want to apply that transformation or not in our 
own CIL-based applications.

But this is all getting a bit away from the original proposal.  The 
original proposal suggested keeping the consts, but just not printing 
them.  If you do this, then *inside* of CIL, the variables will still be 
const and therefore assignments to them are incorrect.  I.e., you are 
asking CIL to manipulate invalid code.  Yes, if you omit the consts when 
printing the code out later, then the code you printed will type check. 
  But if the consts are present inside CIL's data structures, then CIL's 
data structures describe assignments to const-qualified data.  I worry 
that could create problems for other analyses within CIL.