[PATCH] tst-setcontext2: avoid bug from compiler optimization

[Joseph Myers]

On Wed, 25 Jan 2017, Chris Metcalf wrote:

> I'm not sure we need or want the returns_twice attribute, given that
> gcc already has special code for recognizing setjmp, vfork, etc. Currently
> glibc does not use the returns_twice attribute anywhere in its source tree.
> If getcontext or swapcontext is missing from gcc (and I don't know that
> it is) then presumably it's a gcc bug.

Recognizing based on function names and setting ECF_* is definitely an 
obsolescent approach in GCC, compared to the approach of GCC defining 
built-in functions using the appropriate attributes (and defining built-in 
functions is problematic for some of these functions because of the 
header-defined types involved in the function prototypes).  Defining 
built-in functions has the limitation of not providing the attribute with 
-fno-builtin.

Where an attribute is needed for correctness, as here, it should be 
included in the glibc headers.  GCC could then in future eliminate the 
special_function_p special-casing, replacing it either by built-in 
functions or by making fixincludes add the attribute to headers that don't 
already have it.

(Should clone have this attribute, just like vfork?  If so, that's a case 
missing from GCC.)

-- 
Joseph S. Myers
joseph@codesourcery.com