Don't fail on RPC with`Accept: text/plain, application/json`

[steve-chavez]

A request like :

GET /rpc/jsonfunc
Accept: text/plain, application/json

Gives 406 Not Acceptable. We prefer text/plain in this case because both q values are 1.0 by default and there's an order preference. If it were Accept: text/plain; q=0.2, application/json; q=0.8, then application/json would be picked and the request would succeed.

The RFC doesn't specify that ordering is relevant according to this so answer. So we could give preference to application/json over text/plain, because the latter is only supported on certain conditions while the former is supported on every request.

The problem comes from #2162 (reply in thread).

This might require patching parseHttpAccept.

[wolfgangwalther]

I think we need to separate two issues here:

• The bug in Can't open postgrest service in QGIS #2162 seems to be, that we are treating content negotiation (deciding which mediatype to respond to) and "finding the right function to call for this request" separately - while they need to merged into one. Before deciding for any of the mimetypes listed in the Accept header, we need to see which of those are actually actionable for the given set of arguments etc. - in this case, it would only be application/json, because there is no function found that can handle a text/plain type. But we settle on text/plain way too early right now - before we even look into the list of functions.
• The other question is what should happen, when both of the given mimetypes are indeed acceptable, i.e. we have overloaded functions that could handle both mimetypes and they do have the same priority. You suggest that we should give priority to application/json in this case - and I don't think we should. From a set of equally weighted types, it would be absolutely fine to choose the first one that is possible to choose.

[steve-chavez]

Before deciding for any of the mimetypes listed in the Accept header, we need to see which of those are actually actionable for the given set of arguments etc
From a set of equally weighted types, it would be absolutely fine to choose the first one that is possible to choose.

Agree with both 👍