v2: GraphQL peers and type packages.

[jquense]

I think we need to come up with a better strategy for the duplicate graphql issues. The main one is well known, regardless of version mismatch graphql will not run if there are multiple copies, which is dumb. The concern on the horizon (especially with theming) is that any custom types, inadvertently have this problem as well. case in point, both gatsby-transformer-remark and gatsby-transformer-react-docgen use graphql-type-json, but if the package isn't deduped that will break, with gql complaining of an existing type named 'JSON' already.

The only sort of approach i can think that might help here is to do what we do with webpack and provide the graphql object to plugins as well as a set of custom scalars like JSON. That may not super scalable tho :/

[KyleAMathews]

I think this makes sense — it should be scalablish — people wanting new ones can just PR in the new field type — same as React did for years with DOM property support.

[pieh]

What is the best way to pass graphql module?

We do pass graphql as a function to run queries in createPages/createPagesStatefully node hooks, so either pass graphql module under different field or export it from gatsby package?

[pieh]

I started looking into doing this and noticed that gatsby-plugin-remark tests rely on importing from graphql ( https://github.com/gatsbyjs/gatsby/blob/v2/packages/gatsby-transformer-remark/src/__tests__/extend-node.js ). It would also require some refactoring in gatsby-transformer-sharp/gatsby-transformer-sqip.

Right now I'm leaning towards exporting graphql (with added JSON type and any other custom type I will notice in plugins) from gatsby, so plugins can do

-const { GraphQLString } = require(`graphql`)
+const { GraphQLString } = require(`gatsby/graphql`)

Didn't try it yet but it shouldn't conflict with runtime stuff gatsby "export" using webpack alias - https://github.com/gatsbyjs/gatsby/blob/v2/packages/gatsby/src/utils/webpack.config.js#L343

Thoughts on this approach?

[m-allanson]

Done in #5415