Atspi: do we need to keep track of "when a AtspiAccessible has been killed by the application"

[spectranaut]

For example, Orca does it here: https://github.com/GNOME/orca/blob/34ba2393fc64c0e82f47da3900c7de398a89c2cb/src/orca/ax_object.py#L162

For use, we need to know when an object has been killed as well. Should we check preemptively, in IsNull? otherwise, will we throw a specific error every time? Seems a little annoying, because you will have to try/catch for whether the object still exists every time you use it.

[spectranaut]

Notes:

• when the node no longer exists (the web pages changes), we get getRoleName() == invalid
  • do we get role = invalid in other scenarios?
  • seems like this might be browser specific.
• when you kill an application (like chrome), then try to use the accessible for the root, get child count return -1
  • does this happen at other times?
• when you kill an application and try to get the attributes, you get a seg fault, which we can fix:
  (process:1140712): GLib-CRITICAL **: 12:09:21.592: g_hash_table_iter_init: assertion 'hash_table != NULL' failed Segmentation fault
  • fixed in: Make sure attribute hash exists #224

[spectranaut]

Probably this won't happen frequently in testing, but it's still worth thinking about a little bit more.

[alice]

I think maybe we can do something like #226 here too - i.e. have the toString() method never throw/crash.

I like the idea of folding checking for dead-ness into isNull() - maybe then if it is dead, you could set the internal pointer to null as well, so future calls to isNull() will just return true without needing to check for deadness.

I guess we already somewhat handle dead-ness given that we throw a runtime error if we get an error back from one of the library functions, so maybe we just need to do what Orca does and inspect the error string to see whether it's because of a dead object, and null out the pointer if so?