[llvm-commits] my 2nd attempt at now-infamous PR889

[Chris Lattner]

On Feb 16, 2011, at 9:13 PM, pawel kunio wrote:

> Hello,
> In a meantime of preparing 2 tests for PR9067, I came with my 2nd and hopefully now more
> successful attempt at now-infamous PR889 devirtualize Value destructor.
> 
> Please find patches for llvm and clang attached to this mail. As I can see from mingw build
> environment, it will need moving PseudoSourceValue.cpp from lib/CodeGen to lib/VMCore
> after applying patch (seems not to be required on vc build env, as order of libraries given 
> to linker is not meaningful there).
> 
> Please note that destructors have been moved into protected part of interface as well as 
> empty destructors were implemented for all Value-derived classes to help in catching all
> destructions.

Hi Pawel,

Thank you for working on this!  Unfortunately, this patch is way too monolithic to review like this though, could you split it out into a series of obvious changes?  Doug recently "devirtualized" the clang decl hierarchy by devirtualizing one method at a time, which worked really well.

-Chris