I'm making pretty heavy use of JSON parsing in an app I'm writing. Most of what I have done is already implemented using Android's built in JSONObject library (is it json-lib?).
JSONObject appears to create instances of absolutely everything in the JSON string... even if I don't end up using all of them.
My app currently runs pretty well, even on a G1.
My question is this: are the speed and memory benefits from using a stream parser like Jackson worth all the trouble?
By trouble, I mean this: As far as I can tell, there are three downsides to using Jackson instead of the built in library:
I am also using the build-in JSON parser in most cases, but recently stepped into a scenario where it doesn't fit: For some web service requests I receive JSON documents of more than 1 MB. Loading these with the build-in JSON parser requires tremendous amounts of main memory and resulted in OutOfMemoryException
several times. For these scenarios a streaming parser is the better choice (even though it is more inconvenient in use) and the built-in JSON parser does not provide streaming, but only the DOM-like style.
For anyone looking for a streaming JSON parser for Android I can strongly recommend to use Google's GSON. I've tried Jackson JSON at first and it worked fine until I tried to build the release version of my app: ProGuard reported several issues and the running app crashed with mysterious NullPointerException
in the constructor of Jackson's ObjectMapper
(though everything worked fine in the debug version). Even after a few hours of trying around I wasn't able to fix this. I then switched to GSON then and everything worked like a charm.
BTW: The GSON streaming-only jar has a size of only 14kB -- so nothing to really worry about.
Guess you've pretty much answered your own question. :)
Using the built-in JSON parser myself and have never looked for an alternative.
EDIT: Now I'm using a thin annotation-based wrapper from DroidParts.
I think question is whether built-in one is good enough. If it is, sure, minimizing dependencies is often a good strategy. Good enough can refer to both efficiency and ease of use.
For what it's worth, Jackson also has a decent tree model as well as full data binding. Tree model is significantly faster than default parser (parsing is 3x-5x faster on J2SE, probably similarly on other platforms like Android, tree model itself is probably more efficient as well). Or: if you don't want dependency to second jar (mapper is needed for tree model and data binding), writing your own tree to cover your use cases is simple too. Either use basic HashMap/ArrayList/wrappers, or have your own classes if you prefer. Builder would be maybe 40 lines of code top.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With