Semi-determinism in Exempi

[Hubert Figuiere]

Sorry I was meant to reply to this earlier and got sidetracked...

On Wed, 2008-04-09 at 01:24 -0400, Pipian wrote:
> In the process of fiddling a bit more with Exempi, I've worked around  
> a minor bug that I'm still trying to fix (I'll send a patch when I  
> can), and discovered a rather unpleasant side-effect of Exempi's  
> exception handling of the internal C++ code:  Many functions that  
> return boolean values will return false if the main function returns  
> false AND when an exception is caught.  This in and of itself would  
> not be a problem, but there is no way to determine when a genuine  
> error has occurred and when the function is returning false, as the  
> error code is not reset between function calls.
> 
> For example, if there is an error in obtaining a property value with  
> xmp_get_property(), it is indistinguishable from when no property is  
> found, as there is no way to know whether or not xmp_get_error() was  
> set during xmp_get_property().  This makes it difficult to write unit  
> tests when expecting no property to be found, as there is no way to  
> ensure that an error did not occur when reading the property.
>
> I thus propose that exempi add a new error code XMPErr_OK to which
the  
> error variable is set at the beginning of each function.  This way, it  
> is possible to determine if a genuine error has occurred or whether  
> the function is merely returning false.

For individual cases like that it is a bug. I'll recheck the code and
make sure it does it properly by resetting the error code. It is already
supposed to be 0 if there is no error. I'll see if this can be improved.

Also I have to make it thread safe, it isn't currently, ie if called
from different thread the error code may be overridden. Having it
per-thread should help.


Hub