Nexus-future NPM

Documentation

Internal Development

Code Architecture

Overview

Roughly speaking we have three distinct levels of code:
1. src/{cli, framework} Top level modules coded directly against their respective domain.
2. src/utils/* Mid level modules that might provide some conveniences or encapsulate concerns shared across framework and cli.
3. src/lib/* Discrete modules that stand alone conceptually and technically.
Each level builds on the one below it.
One can think of code as evolving from level 1 down toward level 3.
Of course the natural place for some code is level 1. On the other hand we often don't know at first what a generic solution looks like. So level 3 tends to be grown into, rather than started from.
You could see a level 4 here as npm registry, where we fully extract a library. That is not an explicit goal, just a tip for the mental model.
Overall status of the codebase is in a state of refactoring. utils, watcher, and more are undergoing source restructuring in the near future.

Layout Overview

/docs         -- The website
/test         -- Integration tests
/src
  /cli        -- CLI codebase
  /framework  -- Runtime codebase
  /utils      -- Non-discrete modules (may have state, tight coupling)
  /lib        -- Discrete modules

`lib`

The layout of a typical lib module looks like so:

/lib
  /<module-name>
    /index.ts        -- Export-only module, the public interface
    /index.spec.ts   -- Tests against the public interface. Integration in the sense
                        that it is agnostic to the unit or units making up the lib.
    /*.ts            -- The modules making up the lib
    /*.spec.ts       -- Optional tests. Please prioritize `index.spec.ts`

Be careful about lib modules depending upon one another excessively. The more complex the dependency graph the harder it can become to reason about the modules. But if there is non-trivial re-use to be had and/or just a simple/clean and logical dependency then go for it.

The built-in exception to this heuristic is lib/utils which can be thought of as a bespoke lodash for our lib components. Use it for small utilities, which might be shared, are very generic, and are not numerous enough to justify their own dedicated lib module. For example there is a utility to make text span a given length using given pad character.

Testing

We use GitHub Actions.

Unit

yarn test:unit
yarn dev:test      # watch mode

Live under /src separated by and colocated with the respective module they test.
Unit tests run in CI against every commit.

System

yarn test:system

Live under /test/system
Almost like E2E but they work with the local package code (whereas E2E would work with an actually published package).
These are useful because they provide many of the functional checks of E2E with a lower barrier to running, namely needing a published package. For example pull-requests made by community members cannot trigger E2E tests because that would require publishing and GitHub actions has no way for PRs from forks to access secrets.
You must run yarn build right before running these tests.
These tests run by having the local package installed via folder-reference into the test project. This disables package hoisting. This in turn breaks typegen because its import of @nexus/schema will be unresolvable (it looks for node_modules/@nexus/schema but actually needs to look at node_modules/nexus-future/node_modules/@nexus/schema). The system tests encode a solution to this. You should only have to think about this when actually devloping the system tests.

The solution to this problem is an env hook exposed by nexus-future called NEXUS_TYPEGEN_NEXUS_SCHEMA_IMPORT_PATH. when set it will cause the typegen file node_modules/@types/nexus-typegen/index.d.ts to have the import from-value rewritten to whatever the env var value is. This env hook lives inside the typegen component but is leveraged by the system tests suite.

E2E

yarn test:e2e

Live under /test/e2e
E2E tests run in CI against every commit after the package has been published. These are preview and pr releases so its acceptable, and doing it this way provides a true smoke test of if the real user journey works end to end.
E2E tests can be run on your machine. They default to working with latest dist-tag. Use E2E_NEXUS_VERSION env var to set the desired version to test against.

Continuous Delivery

We use dripip to make releases.
Every PR commit results in:
1. Pre-Release of pattern:
```
0.0.0-pr.<pr-num>.<build-num>.<short-sha>`
```
2. Update to an npm dist tag of pattern
```
pr.<pr-num>`
```
Every trunk commit results in a
1. Pre-Release of pattern:
```
<next-version>-next.<build-num>
```
2. Update to an npm dist tag of pattern
```
next
```
Stable releases are cut manually.

Any release type can be run manually:

yarn release:preview
yarn release:stable
yarn release:pr

Website

We use docsifyjs/docsify.
There is no build step
Commits to master will trigger deployment (via gh-pages, no ci/cd on our part)
Navigation is manually managed in _sidebar.md
Cover page is managed in _coverpage.md
Configuration and significant styling customizations are kept in index.html

Getting started

Install docsify-cli
There is currently a bug with docsify-cli requiring the following manual fix after installation. To make this less painful, install globally so you should only have to do this once.
```
yarn global add docsify
```
```
vim /usr/local/bin/docsify
:se ff=unix
:wq
```
Boot docs dev to preview your changes locally
```
yarn docs:dev
```

Workflow Tips

Working With Example Apps via Linking

Refer to https://github.com/prisma-labs/nexus-future-examples

Working with create command

In any example you can use this workflow:

rm -rf test-create && mcd test-create && ../node_modules/.bin/nexus create

Hacks

Shebang / TS use strict / experimental-worker mess

We use workers to speed up extraction of context type from user's app
These are experimental in Node 10.x and thus require flag --experimental-worker
We support Node 10.x
Node only allows opting in via cli flag (e.g. no config file being read optionally)
Nexus is a node CLI and has a shebang so os knows to use Node to interpret the contents
It is not possible to pass flags to node here
Except with a weird shebang hack

The shebang hack is broken by the output of use strict by TS strict mode

Needs to be:

#!/bin/sh
':' //; exec node --experimental-worker "$0" "$@"

Actually:

#!/bin/sh
"use strict"
':' //; exec node --experimental-worker "$0" "$@"

So we use noImplicitUseStrict option to remove it from our emit
This fails because strict adds alwaysStrict and that is incompatible with noImplicitUseStrict
So we explor strict into all its parts except the one we want to remove, alwaysStrict
sigh
...
UDPATE
...

We discovered that TS also emits ; which breaks things:

Needs to be:

#!/bin/sh
':' //; exec node --experimental-worker "$0" "$@"

Actually:

#!/bin/sh
':'; //; exec node --experimental-worker "$0" "$@"

Went back on the no-strict-emit and now resoted to sed operations after build...
...
UDPATE
...
We discovered that npx gets broken by the weird shebang
reverted