Open ericproulx opened 3 months ago
We've relied on multi_json
to swap JSON serializers between the default one and oj
, so removing it would be backwards incompatible and definitely create problems for users. It's possible that the paths where it was meaningful were all with large amounts of JSON, so nobody noticed that the error formatter wasn't using it? Which means it's a bug.
To code must call Grape::Json::Dump
or to_json
when formatting JSON ? The docs says
Built-in formatters are the following.
:json: use object's to_json when available, otherwise call MultiJson.dump
This means that even if you have multi_json
, the latter is not used when formatting JSON. Grape::ErrorFormatter::Json
and Grape::Parser::Json
are ok since its using Grape::Json.dump
and Grape::Json.load
respectively.
Its true that Multijson
facilitates the usage of Oj but anybody could simply call Oj.mimic_JSON()
and Oj
would override all to_json
.
Its true that
Multijson
facilitates the usage of Oj but anybody could simply callOj.mimic_JSON()
andOj
would override allto_json
.
It could be a relic of the past. A change like this is still breaking, so it's got to be worth it.
GitLab has their own JSON Formatter which is using dump
instead of to_json
. I think we should remove the to_json
part and just keep the Grape::Json.dump
.
Am I wrong to say that Grape::Formatter::JSON will never call
Grape::Json::Dump
since every object responds toto_json
? On the other hand, Grape::ErrorFormatter::JSON is not usingto_json
.I'm just concerned about the
multi_json
usage which doesn't seem to be consistent. Should we keep it ?