Comment 1 by mdcurran on 2009-11-27 06:26
I think I am for a try ... except around UIAHandler.initialize in core.main. Its possible we actually should do the same for java access bridge really. Rather than just pretending it worked.

Comment 2 by jteh on 2009-11-27 06:29
So if UIA really isn't available at all, should it raise an exception too, rather than just pretending it succeeded as it does now? One issue is that UIA failing because of "class not registered" or some such is really an error, whereas it not being available isn't an error. We don't want to log an error in the latter case, but do we want to for the former?

Comment 3 by mdcurran on 2009-11-27 06:33
I assume you think that class not registered is more of an error than the dll just not existing because if the dll exists it should be registered?
I personally think it should also throw an error if the dll does not exist, perhaps NotImplementedError? and make initialization of APIs a little more standard.

Comment 4 by jteh (in reply to comment 3) on 2009-11-27 06:36
Replying to mdcurran:

I assume you think that class not registered is more of an error than the dll just not existing because if the dll exists it should be registered?

Correct.

I personally think it should also throw an error if the dll does not exist, perhaps !NotImplementedError?

That's fine with me. However, the question then is: should we special case !NotImplementedError and not log an error in that case? (Perhaps log info or warning, but not an error.) So, the code would look something like:

try:
 UIAHandler.initialize()
except NotImplementedError:
 log.info("UIA not available")
except:
 log.error("Error initializing UIA", exc_info=True)

Comment 5 by jteh on 2009-11-27 07:57
We're working around broken systems here, but it could still be considered a defect in NVDA.

Fix implemented in r3390.
Changes:
State: closed