Decoding 'Maybe Whatever' with .: doesn't throw an error at compile time

[kvanbere]

Decoding 'Maybe Whatever' with (.:) doesn't throw an error at compile time. It just lets you know that it fails at run time, throwing an error instead of returning Nothing when the key doesn't exist.

This really isn't ideal. Could the API be 'safe-by-default' and only allow you to decode Maybe Whatever with (.:?) and not (.:) ? In my use cases I've never wanted to fail at runtime if the key doesn't exist and I intended to decode 'Maybe Whatever' but forgot to update the FromJSON instance.

[phadej]

there are good semantics for FromJSON (Maybe a), null decodes to Nothing, and something else decodes to Just x. This means that key&value should exist in the JSON object. .:? would also succeed with Nothing when key doesn't exist.

In fact, we have .:! to return Nothing if key&value doesn't exist, but otherwise parse as intended.
Therefore, one can parse to Maybe (Maybe Whaverer) with .:!`, distinguishing between successful, non-existing and null cases.

I'm afraid there is no simple way to change aeson so it would be "safe-by-default", somebody will be unhappy with a change! I recommend you to use generic/TH deriving, it's very capable. Maybe instance is useful, yet generics or TH can omitNothingFields and rename fields with fieldLabelModifier.

[phadej]

EDIT: we could make the default better by reworking whole aeson API: we could have FromJSONField type class, then FromJSONField (Maybe a) could "succeed" with Nothing if key doesn't exist. That's however very major change. Something like that is in #564 e.g.

[bergmark]

I think we can close this.