fix: republish @amritk/mini-lynx-native with its dist, and remove the development export condition #102

Workflow file for this run

	name: CI

	on:
	pull_request:
	push:
	branches:
	- main

	concurrency:
	group: ${{ github.workflow }}-${{ github.ref }}
	cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}

	jobs:
	check:
	name: Check
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: oven-sh/setup-bun@v2

	- name: Install dependencies
	run: bun install --frozen-lockfile

	- name: Lint
	run: bun run check

	- name: Check mini reactivity footgun
	run: bun run check:reactivity

	# Regenerating `llms.txt` catches a stale bundle, but a stale source
	# regenerates perfectly: an `AI.md` that never mentions the subpath a
	# release added is wrong in a way no diff can see. This reads what each
	# package actually exports and asks whether its `AI.md` says so.
	- name: Check AI docs match what each package publishes
	run: bun run check:ai-docs

	# `llms.txt` and `llms-full.txt` are generated from the packages' `AI.md`
	# files and committed, so they go stale in exactly the way a committed
	# build artifact always does: silently, and only for the audience that
	# cannot complain. They had drifted far enough to describe a design that
	# no longer existed. Regenerating and diffing is the whole guard.
	- name: Check generated LLM docs are current
	run: \|
	bun run generate-llms
	git diff --exit-code llms.txt llms-full.txt

	# Ahead of the type check, not after it. The packages used to resolve each
	# other's SOURCE while type-checking, through a `development` condition in
	# their exports maps — which meant the manifests shipped a condition
	# pointing at `./src/*.ts` inside a tarball that also ships `src`, so
	# anything honouring it handed consumers raw TypeScript. The condition is
	# gone; cross-package imports resolve to `dist` like a consumer's do, and
	# `dist` has to exist by the time anything reads it.
	- name: Build
	run: bun run build

	- name: Type check
	run: bun run types:check

	# Unaffected by the ordering above: vitest.config.ts aliases every
	# workspace package to its `src/`, so the suite has never needed a build.
	- name: Test
	run: bun run test

	# Loads every compiled module under plain Node, drives the built
	# mini-lynx runtime end-to-end through its fake Element PAPI, and
	# installs packed tarballs like an npm consumer — catches build-, pack-
	# and manifest-level corruption the src-aliased tests cannot see.
	- name: Smoke and e2e test built artifacts
	run: bun run test:dist

	# `android` below — and `ios`, while it was enabled — are the only things
	# anywhere that compile the native halves of the `@amritk/lynx-*` packages —
	# notifications, location, dialogs and deep linking.
	# Everything in `Check` above runs in JavaScript and would pass just as
	# happily against Kotlin and Objective-C that do not build —
	# `src/native-contract.test.ts` narrows that gap by comparing the two native
	# method surfaces against the TypeScript one, but comparing signatures is not
	# compiling.
	#
	# They are separate jobs rather than steps because they need toolchains the
	# main job has no use for, and a macOS runner costs several times an Ubuntu
	# one — keeping them apart means a failure here does not hide a failure there.

	android:
	name: Android native
	runs-on: ubuntu-latest
	steps:
	- uses: actions/checkout@v4
	- uses: oven-sh/setup-bun@v2

	- uses: actions/setup-java@v4
	with:
	distribution: temurin
	java-version: '17'

	- uses: android-actions/setup-android@v3

	# The runner image ships several platforms but the set moves; asking for
	# the one `build.gradle.kts` compiles against is cheap and removes a
	# failure mode that would look like a code problem.
	- name: Install the compile SDK
	run: sdkmanager "platforms;android-35" "build-tools;35.0.0"

	- uses: gradle/actions/setup-gradle@v4

	- name: Install dependencies
	run: bun install --frozen-lockfile

	# `--require-sdk` turns the "no Android SDK, skipping" path into a
	# failure: skipping is right for a contributor's laptop and would be a
	# silently green CI job.
	- name: Compile the Android libraries
	run: bun run check:android --require-sdk

	# DISABLED — the iOS job below was the entire wall clock of this workflow:
	# 81 minutes, against 2m17s for `Android native` and 50s for `Check`.
	#
	# Not because the code is large. The four pods are ~2,600 lines of
	# Objective-C between them. Each `pod lib lint` builds an eight-target graph
	# — `Lynx`, `LynxBase`, `LynxServiceAPI`, `Lynx-LynxResources`, `PrimJS` and
	# a generated host `App` — and compiling the Lynx engine is essentially the
	# whole twenty minutes per pod. Because every lint gets its own throwaway
	# sandbox, that compile happened four times per run, sequentially, to check
	# four small files. On a macOS runner's 10x multiplier that is ~810 billed
	# minutes for a check whose Android counterpart costs two.
	#
	# What is lost while this is off: nothing anywhere compiles the Objective-C.
	# `src/native-contract.test.ts` still compares the native method surfaces
	# against the TypeScript one, so a renamed or dropped method is still caught
	# — but a syntax error, a bad selector, or a header that no longer exists
	# will reach a published tarball. Treat a release touching
	# `packages/lynx-/ios/*` as unverified and lint those pods by hand:
	#
	# cd packages/lynx-notifications/ios
	# pod lib lint MiniLynxNotifications.podspec --allow-warnings --platforms=ios
	#
	# Re-enabling as-is brings the 81 minutes back with it. The two fixes worth
	# making first, in order of payoff: gate the job on `packages/lynx-/ios/*`
	# and the podspecs changing (use a job-level `if:`, not a workflow-level
	# `paths:` — a workflow that never runs never reports, and a required status
	# check that never reports blocks the PR forever), and collapse the four
	# lints into a single Podfile carrying all four as development pods so Lynx
	# compiles once instead of four times.
	#
	# ios:
	# name: iOS native
	# runs-on: macos-15
	# steps:
	# - uses: actions/checkout@v4
	#
	# # `pod lib lint` builds the podspec's sources against the real iOS SDK,
	# # resolving `Lynx` from CocoaPods so the Lynx headers the module imports
	# # are the actual ones. It validates from the files on disk and never
	# # fetches the podspec's `source`.
	# #
	# # `--allow-warnings` because CocoaPods warns about things a library
	# # published from a monorepo subdirectory cannot avoid; it does NOT
	# # suppress compiler errors, which is the point of running this.
	# - name: Lint and compile the notifications pod
	# working-directory: packages/lynx-notifications/ios
	# run: pod lib lint MiniLynxNotifications.podspec --allow-warnings --platforms=ios
	#
	# # A separate step rather than a second podspec in the same command, so a
	# # failure names which pod broke without reading the log.
	# - name: Lint and compile the location pod
	# working-directory: packages/lynx-location/ios
	# run: pod lib lint MiniLynxLocation.podspec --allow-warnings --platforms=ios
	#
	# - name: Lint and compile the dialogs pod
	# working-directory: packages/lynx-dialogs/ios
	# run: pod lib lint MiniLynxDialogs.podspec --allow-warnings --platforms=ios
	#
	# - name: Lint and compile the deep-linking pod
	# working-directory: packages/lynx-deep-linking/ios
	# run: pod lib lint MiniLynxDeepLinking.podspec --allow-warnings --platforms=ios

Navigation Menu

fix: republish @amritk/mini-lynx-native with its dist, and remove the development export condition #102

Workflow file

fix: republish @amritk/mini-lynx-native with its dist, and remove the development export condition #102

Uh oh!

Workflow file for this run