Re: [Emacs-diffs] trunk r117941: Default to stack objects on non-GNU/Linux, non-DOS_NT platforms.

[Paul Eggert]

On 09/25/2014 11:12 AM, Stefan Monnier wrote:
> I saw this extra checking in
> local_cons, I figured this is going to kill too much of the
> performance gain.

I agree that the performance gains are iffy, and would like to see some 
measurements here.

If it's any consolation, in some cases the new MAX_ALLOCA checking is 
optimized away.  For example, other_buffer_safely calls 
build_local_string twice, both with constant literals, and in both cases 
my compiler (GCC 4.9.1 x86-64) deduces that there's no need to check for 
MAX_ALLOCA, and generates code that creates the Lisp string on the stack 
unconditionally.