Re: assert caveat

[Benjamin Franksen <benjamin.franksen@bessy.de>]

On Samstag, 11. Oktober 2008, Till Straumann wrote:
> Unless I miss something then it seems to me
> that EPICS uses 'assert()' in the same way I often
> do which AFAIK is incorrect:
>
> assert( expression_with_side_effect );
>
> because if the code is ever compiled with -DNDEBUG then
> the side effect (eg locking a mutex in epicsMutexMustLock())
> *goes*away*.
>
> 'assert' is a macro which evaluates to nothing if NDEBUG is defined.
>
> The correct way of using assert would be:
>
>   result = expression_with_side_effect;
>
>   assert(result);
>
> but it is inconvenient because the message printed by a failed
> assertion would only say 'assertion of 'result' failed' and it would
> not print the stringified expression.

My standard solution is to use a variable name that sufficiently describes 
the asserted invariant. For instance

#define epicsMutexMustLock(ID) {\
  epicsMutexLockResultIsOk_ ## ID =\
    (epicsMutexLock((ID))==epicsMutexLockOK);\
  assert(epicsMutexLockResultIsOk_ ## ID); }

Cheers
Ben