Re: Proper Tail Recursive C++

"Jerry" <jerpat@iastate.edu> wrote:
>> Does anybody know if there exists a C++ compiler that is properly tail
>> recursive? Please tell me where and how I could get my hands on one.


hbaker@netcom.com (Henry Baker) writes:
>There's a bit of a problem defining exactly what you mean by 'properly
>tail recursive C++'. If local storage is allocated in the stack frame
>and is still 'live' at the time of the tail-call, then you can't just
>'re-use' the stack frame.


Furthermore, a C++ compiler pretty much has to assume that any local
variable whose address is taken and passed to a function defined in a
different translation unit is live. Similarly any local variable
which is passed by reference to such a function must also be assumed
live. Furthermore, any local variable with a non-trivial destructor
is live -- C++ does not give compilers the freedom to invoke the
variable's destructor before the tail-recursive call.


>In the particular case where no storage is 'live' at the time of the
>tail call, then I've been told that some versions of GNU may do the
>right thing.


GNU C++ will do it only if a function has no local variables, and only
if there is an explicit tail-recursive `return' statement (a
tail-recursive void-valued function won't be optimized, unless you use
the GNU extension allowing `return' statements in void-valued
functions).


--
Fergus Henderson <fjh@cs.mu.oz.au>
WWW: <http://www.cs.mu.oz.au/~fjh>
PGP: finger fjh@128.250.37.3
--

Return to the comp.compilers page.
Search the comp.compilers archives again.