🚀 go-pugleaf

Matthew L Martin <mxm@proteon.com> writes:

|> I have reduced this to what I think is the minimum needed to reproduce:
|> The gcc used is 2.7.2 with no local modifications, targeted for
|> m68k-unknown-linux, hosted on a sparc running sunos-4.1.4.

|> I first noticed this on gcc-2.7.2.1, I then backed down to 2.7.2, but it
|> appears to be a m68k-elf bug.

[...]

|>> cat inconf.c

|> /*
|>  * External declarations
|>  */
|> extern int    zero();

|> extern unsigned int tnets;

|> void  *ALP(unsigned  int  size);
|> extern  inline void *alp(unsigned  int  size)
|> {
|>   return((void  *)(
|>     {
|>     register  unsigned  short  __SCOPS ;
|>     register  unsigned  long  __RETURN asm("%d0");
|>     asm volatile ("move %%SR, %0; move #0x2700, %%SR" : "=md" (__SCOPS));
|>     ;
|>     ALP ( size );
|>     asm volatile  ("move %0, %%SR": : "md" ((unsigned  short)(__SCOPS)));
|>     ;
|>     __RETURN;
       ^^^^^^^^ this is not initialized.
|>     })
|>   );
|> }

This is not a bug in gcc: __RETURN is never initialized, so gcc can put
there anything it sees fit.  Actually the real bug is that gcc does not
warn about this use of an uninitialized variable.
-- 
Andreas Schwab                                      "And now for something
schwab@issan.informatik.uni-dortmund.de              completely different"