Henry Robinson commented on ZOOKEEPER-510:

Hm, I had interpreted the Python documentation differently, by analogy with 
PyModule_AddIntConstant. I also tested but couldn't force a GC error. This is 
what I just tried:

>>> import zookeeper
>>> import gc
>>> gc.collect()
>>> zookeeper.set(0,"/","test")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
zookeeper.ZooKeeperException: zhandle out of range

I also tried removing the reference from the module before doing the gc, via 
del zookeeper.ZooKeeperException. 

However, I think you're right, by looking at other usage examples. I think it's 
acceptable to take the examples in the Python manual as canonical, so I support 
reinstating that line. 

The patch is simple, but I'm not sure how best to test it.  Perhaps forcing an 
exception pointer to NULL via a C module call, but then that would pollute the 
object's namespace. A test module that wasn't shipped, but built for testing? 

> zkpython lumps all exceptions as IOError, needs specialized exceptions for 
> KeeperException types
> ------------------------------------------------------------------------------------------------
>                 Key: ZOOKEEPER-510
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-510
>             Project: Zookeeper
>          Issue Type: Bug
>          Components: contrib-bindings
>    Affects Versions: 3.2.0
>            Reporter: Patrick Hunt
>            Assignee: Henry Robinson
>             Fix For: 3.3.0
>         Attachments: ZOOKEEPER-510.patch, ZOOKEEPER-510.patch, 
> ZOOKEEPER-510.patch, ZOOKEEPER-510.patch
> The current zkpython bindings always throw "IOError("text")" exceptions, even 
> for ZK specific exceptions such as NODEEXISTS. This makes it difficult (error 
> prone) to handle exceptions in python code. You can't easily pickup a 
> connection loss vs a node exists for example. Of course you could match the 
> error string, but this seems like a bad idea imo.
> We need to add specific exception types to the python binding that map 
> directly to KeeperException/java types. It would also be useful to include 
> the information provided by the KeeperException (like path in some cases), 
> etc... as part of the error thrown to the python code. Would probably be a 
> good idea to stay as close to java api as possible wrt mapping the errors.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to