Re: Tail Recursion (Why it appears that Gnat 3.15p does support it)

[Duncan Sands <baldrick@free.fr>]

Hi Chad, first I commented out the Put in Fib:

begin
--Put(Int'Image(Next) & ", ");
Fib(Second,Next);

then I compiled this with gcc from CVS as follows:

> gnatmake -O2 tail
gcc -c -O2 tail.adb
tail.adb:12:01: warning: possible infinite recursion
tail.adb:12:01: warning: Storage_Error may be raised at run time
gnatbind -x tail.ali
gnatlink tail.ali

The disassembly is then:

> objdump -r -d tail.o

tail.o:     file format elf32-i386

Disassembly of section .text:

00000000 <tail__fib.365>:
   0:   55                      push   %ebp
   1:   89 e5                   mov    %esp,%ebp
   3:   eb fe                   jmp    3 <tail__fib.365+0x3>
   5:   8d 74 26 00             lea    0x0(%esi),%esi
   9:   8d bc 27 00 00 00 00    lea    0x0(%edi),%edi

00000010 <_ada_tail>:
  10:   55                      push   %ebp
  11:   b8 08 00 00 00          mov    $0x8,%eax
                        12: R_386_32    .rodata
  16:   89 e5                   mov    %esp,%ebp
  18:   ba 00 00 00 00          mov    $0x0,%edx
                        19: R_386_32    .rodata
  1d:   83 ec 08                sub    $0x8,%esp
  20:   89 04 24                mov    %eax,(%esp)
  23:   89 54 24 04             mov    %edx,0x4(%esp)
  27:   e8 fc ff ff ff          call   28 <_ada_tail+0x18>
                        28: R_386_PC32  ada__text_io__put__4
  2c:   31 c0                   xor    %eax,%eax
  2e:   89 e9                   mov    %ebp,%ecx
  30:   89 44 24 04             mov    %eax,0x4(%esp)
  34:   31 d2                   xor    %edx,%edx
  36:   b8 01 00 00 00          mov    $0x1,%eax
  3b:   c7 04 24 01 00 00 00    movl   $0x1,(%esp)
  42:   e8 b9 ff ff ff          call   0 <tail__fib.365>
  47:   c9                      leave
  48:   c3                      ret


which is a bit too optimised!

Comments:
(1) I commented out the Put because I expected the "&" (string concatenation) to cause
trouble, since it uses the secondary stack.
(2) Maybe the fact that Fib is a local procedure causes problems for older gcc.

All the best,

Duncan.