fix: [WIP] mixing offset and limit with Range header

[taimoorzaeem]

Fixes #3007.

@steve-chavez @laurenceisla @wolfgangwalther This is an attempt to fix this issue. There is still a lot of refactoring and testing left in this PR. For now, I would like you to review this redesign. Am I going in the right direction with this?

[laurenceisla]

All in all, It's looking good! Separating Ranges and treating limit and offset as query strings should be the right choice. I'd say 👍 on continuing with this PR.

Some considerations below:

[laurenceisla]

This should still allow doing something like

/clients?select=*,projects(*)&projects.limit=1

I think this change breaks that.

[laurenceisla]

This also may be related. It breaks limited updates/deletes which need order to be present in the query string.

[taimoorzaeem]

@laurenceisla @steve-chavez How should this change now reflect in Content-Range header in below tests? According to mdn the range start and end are zero indexed and inclusive. Shouldn't this mean that if the result has only 1 row, it should return Content-Range: 0-0/* instead of 1-1/*?

it "using the Range header" $
          request methodGet "/rpc/getitemrange?min=2&max=4"
                  (rangeHdrs (ByteRangeFromTo 1 1)) mempty
             `shouldRespondWith` [json| [{"id":4}] |]
              { matchStatus = 200
              , matchHeaders = ["Content-Range" <:> "1-1/*"]
              }

        it "using limit and offset" $ do
          post "/rpc/getitemrange?limit=1&offset=1" [json| { "min": 2, "max": 4 } |]
             `shouldRespondWith` [json| [{"id":4}] |]
              { matchStatus = 200
              , matchHeaders = ["Content-Range" <:> "1-1/*"]
              }
          get "/rpc/getitemrange?min=2&max=4&limit=1&offset=1"
             `shouldRespondWith` [json| [{"id":4}] |]
              { matchStatus = 200
              , matchHeaders = ["Content-Range" <:> "1-1/*"]

Edit: Also, now that limit and offset are separated from range header, how about we also separate their test cases? RangeSpec.hs for range header and a new LimitOffsetSpec.hs for testing limit and offset query params? WDYT?

[laurenceisla]

Shouldn't this mean that if the result has only 1 row, it should return Content-Range: 0-0/* instead of 1-1/*?

Not necessarily, remember that the Range can have an offset, e.g. Range: 1-1, which starts on the second item and ends in the same second item.
Then the response header is valid: Content-Range: 1-1/*. This is happening in the first example (rangeHdrs (ByteRangeFromTo 1 1) = Range: 1-1).

        it "using the Range header" $
          request methodGet "/rpc/getitemrange?min=2&max=4"
                  (rangeHdrs (ByteRangeFromTo 1 1)) mempty
             `shouldRespondWith` [json| [{"id":4}] |]
              { matchStatus = 200
              , matchHeaders = ["Content-Range" <:> "1-1/*"]
              }

The entire resource (without the Range) is {"id": 3, "id": 4}, then Range: 1-1 takes {"id": 4}.

The other requests with limit=...&offset=... without the Range header, must not return Content-Range (mixing both needs to be tested too, there's an example in the original issue on what it should return).

Edit: Also, now that limit and offset are separated from range header, how about we also separate their test cases? RangeSpec.hs for range header and a new LimitOffsetSpec.hs for testing limit and offset query params? WDYT?

I agree, they must be separated now.

[taimoorzaeem]

The other requests with limit=...&offset=... without the Range header, must not return Content-Range (mixing both needs to be tested too, there's an example in the original issue on what it should return).

@laurenceisla Alright, now with this, I am also assuming that Prefer: count=exact/planned/estimated will only be applied when Range: .. was sent in the request. Is my assumption correct?

[laurenceisla]

Alright, now with this, I am also assuming that Prefer: count=exact/planned/estimated will only be applied when Range: .. was sent in the request. Is my assumption correct?

Yes that would be correct. This will make PostgREST a bit more HTTP compliant, so this PR is also partially related to #1089. There, it mentions:

[...] in RFC 7233 Content-Range is only a feature of special partial responses, which are indicated by status code 206 (Partial Content), which in turn can’t happen unless the request has a Range. RFC 7233 § 4.2

[taimoorzaeem]

it "updates with limit/offset using table default values(field-with_sep) when json keys are undefined" $ do
          request methodPatch "/complex_items?select=id,name&columns=name,field-with_sep&limit=1&offset=2&order=id"
            [("Prefer", "return=representation"), ("Prefer", "missing=default")]
            [json|{"name": "Tres"}|]
            `shouldRespondWith`
            [json|[
              {"id":3,"name":"Tres"}
            ]|]
            { matchStatus  = 200
            , matchHeaders = ["Preference-Applied" <:> "missing=default, return=representation"]
            }

I am having trouble figuring out what this one test is NOW supposed to return and what could be the problem here? @laurenceisla WDYT?

Failures:

  test/spec/Feature/Query/UpdateSpec.hs:356:13: 
  1) Feature.Query.UpdateSpec, Patching record, PATCH with ?columns parameter, apply defaults on missing values, updates with limit/offset using table default values(field-with_sep) when json keys are undefined
       body mismatch:
         expected: [{"id":3,"name":"Tres"}]
         but got:  []

[laurenceisla]

Alright, now with this, I am also assuming that Prefer: count=exact/planned/estimated will only be applied when Range: .. was sent in the request. Is my assumption correct?

Yes that would be correct. This will make PostgREST a bit more HTTP compliant, so this PR is also partially related to #1089.

Wait, I think I made a mistake here. I mean, this is correct and needs to be changed, but it would be better to do so in a different PR. I didn't take in consideration that the Content-Range was returned on ALL requests right now.

Sorry about this, you made considerable changes to the tests and the process that generates the Content-Range. You have these options here:

1. Revert the changes in Content-Range (that is, make all the requests return the header like before).
2. Keep the changes, but they definitely need to be on a different PR. The change is "Do not return Content-Range header unless Range is sent", which is a standalone issue, and it may need a confirmation for the rest of the team.

[taimoorzaeem]

@laurenceisla @steve-chavez @wolfgangwalther One test case that is remaining to be fixed is:

test/spec/Feature/Query/UpdateSpec.hs:368:13: 
  1) Feature.Query.UpdateSpec, Patching record, PATCH with ?columns parameter, apply defaults on missing values, updates with limit/offset using table default values(field-with_sep) when json keys are undefined
       body mismatch:
         expected: [{"id":3,"name":"Tres"}]
         but got:  []

Any idea how is this test giving this result?

[wolfgangwalther]

test/spec/Feature/Query/UpdateSpec.hs:368:13: 
  1) Feature.Query.UpdateSpec, Patching record, PATCH with ?columns parameter, apply defaults on missing values, updates with limit/offset using table default values(field-with_sep) when json keys are undefined
       body mismatch:
         expected: [{"id":3,"name":"Tres"}]
         but got:  []

Any idea how is this test giving this result?

This PR is too big for me to look at - I don't even know where to start. Is there any chance to somehow split this one commit into multiple pieces, where each piece can stand alone, make sense and have passing tests?

[taimoorzaeem]

This PR is too big for me to look at - I don't even know where to start. Is there any chance to somehow split this one commit into multiple pieces, where each piece can stand alone, make sense and have passing tests?

I think not. The range header and limit and offset were too intermingled that it's almost impossible (at least for me). Also, I wrote this patch quite a while ago, so it would be kinda hard for me to even understand the changes.