stack-protector configure test and MIPS64

[Joseph S. Myers]

On Mon, 21 May 2012, Roland McGrath wrote:

> > The following patch does that.  It suffices to make the build work for 
> > MIPS64; also tested with a full build and testsuite run for x86_64.
> 
> My paranoia would be happier with a check that doesn't get confused by
> a version of -fstack-protector that emits some symbol other than
> __stack_chk_fail.  In fact, it looks like my check would be confused
> by an i386 or powerpc GCC that set -fPIC by default (because then it
> would call __stack_chk_fail_local).  Such confusion today is a hard
> error.  Your change would turn that error into a false negative, which
> troubles me.

I'm not aware of any distribution compilers with that property, and this 
test is really about working well with known distribution compilers (we 
don't try to allow for every possible strange way a compiler could be 
configured, just those that seem likely to arise in practice).  A false 
negative would simply mean a build failure at some later point rather than 
at configure time; I'm not convinced it's worth complicating this test to 
avoid that for a hypothetical compiler.

-- 
Joseph S. Myers
joseph@codesourcery.com