`Query` object resolvers type should require only those fields not already required on the mapped type from resolvers about graphql-code-generator HOT 4 CLOSED

Hi @loucadufault ,
On top of the "fields missing from the mapped type must be required" scenario, there are a few cases we need to consider:

1. If the type of a mapper's field does not match the schema type's field with the same name, the field resolver must be impl. In the example below, isAdmin resolver is required in the following case because yes | no must be converted to boolean to avoid runtime error.

// schema
type User {
  isAdmin: Boolean!
}

// mapper
type UserMapper = {
  isAdmin: "yes" | "no"
}

2. Mapper can be anything e.g. string, types from node_modules, inferred type, even a class

To be able to handle more complex cases effectively, we need to be able to do static analysis to compare the mapper against the generated TypeScript type. This can be done most effectively using the TypeScript compiler.

However, adding static analysis into the typescript-resolvers plugin may not be as effective as it sounds. One reason is performing static analysis in TypeScript could be complex, and adding it to the base plugin could make it much harder to maintain.

This is why Server Preset (@eddeee888/gcg-typescript-resolver-files) was created. It uses the generated types from typescript-resolvers plugins and compare it against the mapper types, handling the mentioned missing fields and other complex cases.

In your example, it'd automatically generate resolver files and mark the resolvers that need mappings, and tell the developer why it needs to be implemented:

export const Book: BookResolvers = {
  author: () => {
    /* Book.author resolver is required because Book.author exists but BookMapper.author does not */
  },
 }

Here's another demo where server preset adds required resolvers on mapper type changes:

There's more to features as well, you can read more on how it works and how to set it up here: https://the-guild.dev/graphql/codegen/docs/guides/graphql-server-apollo-yoga-with-server-preset

from graphql-code-generator.

Closing this because this is better addressed in Server Preset. If you have issues related to Server Preset, please create an issue in https://github.com/eddeee888/graphql-code-generator-plugins (It's in a separate repo but I'm working closely with the Codegen team)

from graphql-code-generator.

Thanks for the reply

If the type of a mapper's field does not match the schema type's field with the same name, the field resolver must be impl.

Are these actual graphql-js semantics, that it will attempt to call the next resolver if the current one returned an unexpected type?

2. Mapper can be anything e.g. string, types from node_modules, inferred type, even a class

To be able to handle more complex cases effectively, we need to be able to do static analysis to compare the mapper against the generated TypeScript type. This can be done most effectively using the TypeScript compiler.
However, adding static analysis into the typescript-resolvers plugin may not be as effective as it sounds. One reason is performing static analysis in TypeScript could be complex, and adding it to the base plugin could make it much harder to maintain.

Can you elaborate on this supposed complexity? To me it just seems like a deep type equality check, I can't imagine that is an unsolved problem in the TypeScript world. As far as checking for all the possible types, since TS types just wrap JS primitives, in this case an object, it comes down to checking properties.

I'm mainly not understanding the argument as to why this must be in a separate project; in my view this feature of codegen is fundamentally broken as it stands.

from graphql-code-generator.

that it will attempt to call the next resolver if the current one returned an unexpected type?
The next resolver in the chain will always be called if provided. If the last one in the chain return a type that does not match the final output type, it'd send an error.

Can you elaborate on this supposed complexity?

Sure, here's one of the issues of doing it in base resolver plugin:

Generally, to effectively run TypeScript checks we need to load files into compiler's Program with compiler options. Then, we load files (virtual or physical), and do property checks, etc. Using the TypeScript compiler helps us do inferred type, type alias, interfaces, etc. mappers correctly

So, here's what we'd need to do:

• The base resolver generates some output, but before we generate this output to physical files...
• We need to load that output into the Program, load all mappers into the Program...
• Do statical analysis to find missing fields...
• Re-run codegen logic with missing fields to make them required...
• Return the output to Codegen engine to generate the physical files

Apart from running the same logic twice, we need to use typescript or ts-morph as a hard dependency for the base resolver.

This will have performance impact to all existing users.

in my view this feature of codegen is fundamentally broken as it stands.

If you have ideas how to fix issues at the base plugin level, feel free to create a PR 🙂

from graphql-code-generator.