Breaking change in version 2.5 about ijson HOT 5 CLOSED

A new 2.5.1 version has been uploaded to PyPI to add the automatic reader conversion to the other backends, restoring backwards compatibility for users of the default ijson backend. A warning is shown when users access this functionality to indicate they should move to byte-based readers.

from ijson.

Maybe related to #12

Look #12 (comment)

from ijson.

Ok. Thanks. I've rebased tabulator on [email protected] - frictionlessdata/tabulator-py@16af3de - and it works. My main point was that the change like this is not what you expect on the minor release 😃 I would consider preserving compatibility at least until ijson@3. But thanks anyway.

PS.
Probably this library - https://github.com/reubano/meza/search?q=ijson&unscoped_q=ijson - can also stop working after [email protected]

from ijson.

I hadn't realised the python backend was different in this aspect from the other ones -- all other backends (which I tend to focus more on) work with bytes and have always failed when a string reader is passed, while the python backend is the only one that can take either. This definitely wasn't an intentional change, but an effect on changing the default backend from python to the fastest available.

Regarding breaking backwards compatibility and how to handle it now that this new version is out there, I'm a bit torn. On the one hand I understand that the default backend that gets selected behaves now differently, but on the other the change on default backend selection is documented, and explicitly selecting a backend is still possible (and their behaviours haven't changed), so users relying on the behavior of the python backend can easily continue using it. I think at the end of the day I'll probably end up releasing an ijson 2.5.1 with the warning you suggest to avoid further confusion.

from ijson.

Caught this breakage in our CI when docker images updated today. Will update our tabulator-py version pin to fix.

from ijson.