Re: [pr 41246] Fix incorrect values of DECL_NO_STATIC_CHAIN

[Richard Henderson <rth at redhat dot com>]

On 09/16/2009 12:50 PM, Eric Botcazou wrote:
> > While looking into fixing the PR, I discovered that we can set
> > DECL_NO_STATIC_CHAIN incorrectly.  When nested functions call other
> > nested functions, they may need to pass along the original static chain.
> >    This discovery is by nature iterative; we had been performing a single
> > pass over the nested function tree and setting DECL_NO_STATIC_CHAIN
> > based on that.
>
> I think it boils down to what "incorrectly" means here.  Starting with a
> conservatively correct set of DECL_NO_STATIC_CHAIN flags, the old algorithm
> will generate another conservatively correct set, not maximal of course.

"Incorrectly" meaning a static-chain value passed to a function with 
DECL_NO_STATIC_CHAIN set.  "Incorrectly" meaning no static chain passed 
to a function when it ought to have gotten one.

I don't see how you say the old algorithm produced a "conservatively 
correct" set of values at all, except that *somehow* the Ada compiler 
was functioning.

> Write a testcase in GNU C if you don't like Ada. :-)

It's not that I don't like Ada, it's just that it seemed like we ought 
to have been generating incorrect code.  So I don't know how the 
compiler was working.

> Thanks, you presumably fixed PR tree-opt/30927.

So I did.


r~