No longer able to catch unexpected errors after removal of UnforgivingExecutionContext

まず、削除された UnforgivingExecutionContext テストと pull request 1255 および 1357 の変更を確認し、その後、graphql-core==3.1.6 と動作を比較します。リンクされた gist の失敗ケースを再現します。予期しない非 GraphQL エラーが実行を停止し、トップレベルのエラーを 1 つ生成し、バックエンドの stack trace を保持する一方で、GraphQL エラーは変更されない状態になれば完了です。

• What is the current behavior?
  Since the introduction of https://github.com/graphql-python/graphene/pull/1255 by @AlecRosenbaum it was possible to raise unexpected errors to the backend processing, as examplained in https://github.com/graphql-python/graphene/issues/902#issuecomment-672952043. This is critical so you get a decent traceback in the log and no error message for each record in the result set, ie

{
  "errors": [
    {
      "message": "An unknown error occurred.",
      "locations": [
        {
          "line": 5,
          "column": 5
        }
      ],
      "path": [
        "testObject",
        0,
        "testField"
      ]
    },
    {
      "message": "An unknown error occurred.",
      "locations": [
        {
          "line": 5,
          "column": 5
        }
      ],
      "path": [
        "testObject",
        1,
        "testField"
      ]
    }
]
}

Changes in graphql-core started to let the UnforgivingExecutionContext tests fail, as recorded by @mweinelt in https://github.com/graphql-python/graphene/issues/1346 and noted by @weilu https://github.com/graphql-python/graphene/pull/1255#issuecomment-857877151 with a suggestion for a potential fix.

https://github.com/graphql-python/graphene/pull/1357 by @codebyaryan brought in welcome changes for query validation, but also removed UnforgivingExecutionContext in https://github.com/graphql-python/graphene/pull/1357#issuecomment-902573578

• If the current behavior is a bug, please provide the steps to reproduce and if possible a minimal demo of the problem via
  a github repo, https://repl.it or similar.

If we bring back the Test case for UnforgivingExecutionContext without specifying UnforgivingExecutionContext per this configuration https://github.com/alexhafner/graphene/commit/0aaa3fcfac05ae5243d0a65faa562334ae5f37e3, we get those tests failing because unexpected errors are no longer being raised up: https://gist.github.com/alexhafner/5215ef726b356eef6571efb969f6a3e7

• What is the expected behavior?

Be able to stop processing a graphql operation when an unexpected error occurs, return a top-level error message to the GraphQL user, and get a full stack trace in the backend to be able to act on the error message further.

For example:

If we bring back UnforgivingExecutionContext, and create a similar solution to the one in https://github.com/graphql-python/graphene/pull/1255#issuecomment-857877151, the tests succeed again as shown here: https://gist.github.com/alexhafner/a78f493f1b869df0ba112b4acb5da5e9. The approach in https://github.com/alexhafner/graphene/commit/f7655378709244a3e24f00605b47056237ff77d9 raises the original errors for non-GraphQL Errors, and handles GraphQL Errors unchanged by using super() on handle_field_error().

other solutions are welcome

• What is the motivation / use case for changing the behavior?

catching unexpected issues is system and security critical, and a full stack trace is needed.

• Please tell us about your environment:

  • Version: graphene master with graphql-core==3.1.6, graphql-relay==3.1.0 graphql-server==3.0.0b4
  • Platform: OS X

• Other information (e.g. detailed explanation, stacktraces, related issues, suggestions how to fix, links for us to have context, eg. stackoverflow)
  N/A