Problem with variable inspection and variadic functions?

Follow

Brian Myers

• May 24, 2010 02:52

I am using Crossworks for ARM v2.0.5.

I have observed that the inspection of variables seems to not work correctly inside variadic functions - at least sometimes; maybe always.

This is in debug mode, with no optimisations turned on. Hovering over a variable, the watch window, and locating a variable in memory all seem to point to the wrong place (but the code functions as expected, it's a debugger problem).

We can get around the problem by putting functionality which we wish to debug inside a sub-function, which the variadic function calls.

Below is a test function I have adapted from a similar one which we actually use. When stepping through this code, the variable "index"  does not change (hover, watch-window, or memory window); it is reported as 0x0800f378. Its address, as reported by the memory window, is 0x200027f8. Observing nearby memory reveals that "index" appears to actually reside at address 0x200027ec, in this example.

Am I doing something wrong? Or have I found a bug?

 

 

static void testing_printf( bool console_operation, const char *fmt_string, ... )
{
    va_list args;
    uint32_t index;

    if ( !console_operation )
    {
        debug_putchar( '#' );
    }

    va_start( args, fmt_string );    // Finds beginning of variadic arguments (after fmt_string).

    // Special version that takes va_list type (instead of an explicit list of args).
    for ( index = 0; index < 5; index++ )
    {
        debug_vprintf( fmt_string, args );
    }

    va_end( args );

    if ( !console_operation )
    {
        debug_putchar( '$' );
    }
}

0

• 

  Kelly Painter

  • May 24, 2010 22:08

  Did you try getting rid of the static function declaration?  I don't know if that is the problem but it is something I would suspect.

  0

  Comment actions Permalink
• 

  Brian Myers

  • May 24, 2010 22:14

  Didn't try that - already have a work-around.

  Variable inspection should still work in a static function, in any case. Functions which can be declared "static", certainly should be (IMHO).

  0

  Comment actions Permalink
• 

  Paul Curtis

  • May 25, 2010 14:47

  There is a problem in variadic debug information generation in GCC.  This has recently been addressed and may well be fixed.  I'm sure that EABI will produce even more problems.

  0

  Comment actions Permalink

Please sign in to leave a comment.