Re: g_try_new and g_try_new0 + error reporting for g_object_set

[Tim Janik <timj gtk org>]

On Thu, 12 May 2005, Stefan Kost wrote:

> Hi Jan,
>
> the problem is not the OOM case. The problem is that the user may has entered 
> a too large value for something and that is the base for an allocation.
> It is much better telling the user that this much memory can't be allocated, 
> rather that just crash and loose the work of the user.

you are right, and similar situations are actually what g_try_new() and friends
were added for. however the GObject proeprty system has a limitation by design
that conflicts with the scenrio you outlined above. that is, setting an object
property will *always* succeed, there is no mechanism to handle or report
errors in setting object properties through the g_object_set*() interface and
there never will be.

the options you have to implement you user feedback above are:
* in any case accept the value set on the object via g_boject_set () (since you
  have to and can't signal g_object_set() failing)
- simply don't change you internal value if the value set from the outside is
  too large (i.e. turn property setters of too large values into a NOP).
- add a current-error or similar property which is changed when setting too
  large values, so the user is notified (::notify::current-error) and can read
  out the error.
- add a signal that is emitted if too big values are set, the signal can carry
  an error description.
- use your own API instead of object properties which can have suitable
  mechanisms to report failure and report errors.

---
ciaoTJ