Re: [avr-libc-dev] Re: snprintf bug ?

[Dmitry K.]

On Thursday 13 October 2005 21:46, Russell Strong wrote:
> Hi,
>
> The following is the full source and Makefile for my snprintf bug

[...]
> static void update_heading_string (int hdg_reading)
> {
>       long delta = 0;
>       unsigned char i;
>       static int base_hdg = 0;
> //    static char heading_tmp[5];     // This works
>       char heading_tmp[5];            // This doesn't work
[...]
>       snprintf (heading_tmp, 5, "%04X", base_hdg);
[...]
> int main () __attribute__ ((naked));
> int main ()
> {
[...]
>               update_heading_string (heading_byte_1 * 256 + heading_byte_2);
[...]
> OPTIMIZE       = -O2
[...]

I have look yours program by two compilers: 3.3 and 4.0.
Modern compiler (4.0) make function 'update_heading_string'
as inline function (with -O2). This reduce code size (call
is single) and speed up run.
   You have exclude frame pointer (r28/r29) initialization
by defining 'naked' for main.
   But with automatic string 'heading_tmp' this is not work:
frame pointer is used as base address.  Result is writing
to random place.  With static string all works:
address is absolute.

With old compiler it works also: 'update_heading_string'
is not inlined, frame pointer is initialized in prologue
of 'update_heading_string'.

I do not known is this bug single, I have not try to run.

Regards,
Dmitry.