[avr-gcc-list] Re: More interrupt oddity

[Peter N Lewis]

> On 21 Oct 2001, at 12:01, address@hidden wrote:

>  > The goofy part is that the linker didn't give any errors or warnings
> >  that something may be wrong. Shouldn't it fail with an undefined
>  > symbol or something?

If you look in the docs, aka sig-avr.h, you see that:

#define SIG_INTERRUPT0 _interrupt0_

SIG_INTERRUPT0 is just defined to be the name of the routine 
(_interrupt0_), and that name (_interrupt0_) happens to be used in 
the jump table, so the linker knows it can be called.

#define SIGNAL(signame)                                 \
void signame (void) __attribute__ ((signal));           \
void signame (void)

So your code just ended up being:

void SIG_INTERRUPT0 (void) __attribute__ ((signal));
void SIG_INTERRUPT0 (void)

This is just a function which preserves more registers and flags than 
normal.  And it is never called, so I guess the linker would be free 
to remove it entirely or optimize away all the code which has no 
effect.

There isn't ny good way for the code to defend against this problem, 
unless there was perhaps some what to add to the #define a warning if 
the input parameter was SIG_* since that would presumably indicate a 
misspelling.

Enjoy,
   Peter.

--
Sungroper Solar Car Association <http://www.sungroper.asn.au/>