GLSP-1682: Consume the glsp-core packages and migrate the VS Code e2e tests - #113
Merged
Merged
Conversation
- Add `@eclipse-glsp/playwright-vscode`, the published VS Code integration for the GLSP Playwright framework, and a private `workflow-vscode-e2e` package holding the Playwright configuration and specs, both under a new `e2e/` workspace directory - Consume the framework and the integration-agnostic Workflow suites from glsp-core (`@eclipse-glsp/playwright`, `@eclipse-glsp-examples/workflow-e2e`) as regular npm dependencies, so running the tests no longer requires glsp-playwright and glsp-server-node checkouts next to this repository. The VSIX under test is likewise resolved from this repository instead of through the GLSP CLI - Cover the desktop and the web extension variant as separate Playwright projects, replacing the `USE_WEB_EXTENSION` switch. Both contribute the same custom editor, so each variant is installed into an extensions directory of its own via the new `extensionsDir` option; without it the workbench would decide which one the tests exercise - Resolve the webview by class instead of by tag. A web extension is hosted in an additional hidden worker-host iframe, which made the previous selector ambiguous and the web variant untestable - Let Playwright start only the bundled Workflow node server and export the connection settings the extension reads, so it attaches instead of spawning its own. Any other server, including the Maven-built Java server, is expected to be running already, mirroring glsp-core - Reduce the e2e workflow to a single checkout accordingly. It stays red until glsp-core publishes its packages to npm Also: - Remove the obsolete `.yarnrc` left over from the migration to pnpm - Adapt to the reformatting and the obsolete lint suppressions reported by the newer `@eclipse-glsp/prettier-config` and `eslint-config` Part of: eclipse-glsp/glsp#1682 Co-authored-by: Haydar Metin <dev@haydar.metin.at>
- Drop the local verdaccio redirect and resolve `@eclipse-glsp` and `@eclipse-glsp-examples` from npmjs again, now that glsp-core publishes its packages - Migrate to pnpm 12, oxc and TypeScript 7, following glsp-core. This rides along rather than landing separately because the dependency update forces it: the shared `ts-config` sets `moduleResolution: bundler`, which TypeScript 5 rejects for CommonJS output, and `@eclipse-glsp/config` no longer provides ESLint or Prettier at all - Replace the ESLint and Prettier setup with `oxlint.config.mts` and `oxfmt.config.mts`, and lint from the repository root only, dropping the per-package lint scripts - Raise the floor to Node 24 across the engines field and the workflows, matching the toolchain glsp-core now requires Also: - Point the editor configuration at the oxc extension and the TypeScript Native Preview extension; TypeScript 7 ships no tsserver, so `typescript.tsdk` can no longer be used - Drop the unused `mocha-ctrf-json-reporter` dependency and the hoisting entries for the mocha toolchain, which this workspace no longer uses Part of: eclipse-glsp/glsp#1682
`glsp updateNext` no longer resolves a floating `next` dist-tag range at install time; it rewrites the manifests to the currently published exact version instead, so every nightly bump is a reviewable diff. - Replace the literal `next` ranges of the `@eclipse-glsp` and `@eclipse-glsp-examples` dependencies with the exact versions the command resolves, as produced by `pnpm upgrade:next` Part of: eclipse-glsp/glsp#1682
Follow the layout glsp-core settled on, so that the instructions are shared by every agent rather than being Claude-specific. - Reduce `CLAUDE.md` to a reference to `AGENTS.md` and rewrite the latter as a short list of working rules instead of a project description that restates `README.md` and `package.json` - Move the skills to `.agents/skills/` and mirror them into `.claude/skills/` for discovery - Exclude both directories from the workflow path filters, so agent metadata changes no longer start CI or e2e builds Also, correcting what the skills claimed: - The fix skill now notes that linting type-checks too, and why the e2e suites are excluded - The changelog skill used the old Prettier list markers and named Prettier as the formatter; it is now oxfmt-correct. Changelog generation is no longer model-invocable, matching glsp-core Part of: eclipse-glsp/glsp#1682
The client, protocol, Node server, Workflow example and the Playwright framework are now consumed from `glsp-core`, but the documentation still pointed at `glsp-client` and `glsp-server-node`, which are about to be archived. - Replace those references in `README.md`, `CONTRIBUTING.md` and the package READMEs with `glsp-core`, keeping `glsp-server` for the Java server. The server-from-sources section now distinguishes which repo holds which variant - Describe the package layout in `README.md`, which `AGENTS.md` refers to for exactly that and which never mentioned the `e2e` packages. Document `pnpm test:e2e` alongside it Also: - Correct the prerequisites, which still asked for Node 22 and pnpm 11 Part of: eclipse-glsp/glsp#1682
`pnpm/action-setup` v6.0.8 predates pnpm v12 support and installs it into the v10 `PNPM_HOME` layout, where pnpm no longer exposes its bundled node-gyp to lifecycle scripts. Raising `packageManager` to pnpm 12 therefore breaks the install step of any workflow whose dependency graph builds a native module. - Update the action to v6.1.0, whose only change is pnpm v12 support, in all four workflows that set pnpm up Part of: eclipse-glsp/glsp#1682
`oxlint --format github` only emits inline annotations, so a failing Lint job gave no overview of what was wrong without scrolling the raw log. Adopt the `lint-report.mjs` script from glsp-core, which runs oxlint once with the JSON reporter and emits both the inline annotations and a Markdown table in the job summary. It fails on errors and passes on warnings, exactly like `pnpm lint`. Part of: eclipse-glsp/glsp#1682
Contributor
There was a problem hiding this comment.
Can we add a script for headless test:e2e in the package.json so that i dont need to remember the concrete code to run it?
Watch doesnt work:
❯ pnpm watch
$ concurrently --kill-others -n tsc,webview,extension,web-extension -c red,blue,yellow,green "tsc -b -w --preserveWatchOutput" "pnpm -sworkflow:webview watch:bundle" "pnpm -s workflow watch:bundle" "pnpm -s workflow:web watch:bundle"
[tsc] [10:56:46 AM] Starting compilation in watch mode...
[tsc]
[web-extension] error: unexpected argument '-s' found
[web-extension]
[web-extension] Usage: pnpm [OPTIONS]
[web-extension]
[web-extension] For more information, try '--help'.
[extension] error: unexpected argument '-s' found
[extension]
[extension] Usage: pnpm [OPTIONS]
[extension]
[extension] For more information, try '--help'.
[tsc] [10:56:46 AM] Found 0 errors. Watching for file changes.
[tsc]
[extension] pnpm -s workflow watch:bundle exited with code 2
--> Sending SIGTERM to other processes..
[web-extension] pnpm -s workflow:web watch:bundle exited with code 2
--> Sending SIGTERM to other processes..
[webview] error: unexpected argument '-s' found
[webview]
[webview] Usage: pnpm [OPTIONS]
[webview]
[webview] For more information, try '--help'.
[webview] pnpm -s workflow:webview watch:bundle exited with code 2
--> Sending SIGTERM to other processes..
[tsc] tsc -b -w --preserveWatchOutput exited with code SIGTERM
[ELIFECYCLE] Command failed with exit code 1.
Two findings from the review of the glsp-core migration. - Restore `pnpm watch`, which died on an argument error for each of the three bundler processes: pnpm 12 no longer takes `-s` as the global silent shorthand, on `run` it means `--sequential` now. Pass `--silent` instead - Add `test:e2e:headless` and `e2e test:headless`, which wrap the run in `xvfb-run -a`, so running the suites without a display no longer means remembering the incantation. The e2e workflow and the documentation use them Part of: eclipse-glsp/glsp#1682
`glsp releng publish next` publishes only the packages affected by the last commit. Following glsp-core, expose a toggle that overrides that baseline, for the cases where a publish has to cover everything. - Add a `full` input to the `workflow_call` and `workflow_dispatch` entry points and pass `--full` to `publish:next` when it is set - Drop the hand-written change detection, which predates the affected-only publishing in the CLI and duplicated it. It also only ever matched `packages/`, so a change confined to the published `e2e` package would have skipped the publish job entirely Part of: eclipse-glsp/glsp#1682
Concurrent Playwright workers made the hover-driven tests flaky, which is what turned the e2e job red: VS Code windows sharing one display and one pointer deliver the enter and leave events of a window that opens or closes to whichever window is underneath, cancelling a hover another worker is waiting on. The suites then time out waiting for the popup. Neither retrying the hover in the test framework nor dropping the webview's focus-lost dispatch removes it, so the interference is environmental rather than a client bug. - Start one `Xvfb` per worker process and launch that worker's VS Code on it, requested with `GLSP_VSCODE_ISOLATED_DISPLAYS` and set by the `:headless` scripts. The server picks a free display itself, so workers starting at the same time cannot collide - Derive the worker count from that isolation: half the cores, at most four, and one wherever the displays cannot be separated, which covers a plain run watched on the caller's display as well as macOS and Windows - Split the e2e workflow into one job per packaged variant, so the desktop and the web suites no longer run back to back The desktop and the web suite pass on four workers in 1.6 and 1.7 minutes respectively, against 5.2 and 5.3 minutes serially. Part of: eclipse-glsp/glsp#1682
Splitting the e2e workflow per variant broke it: each job packages only its own extension, and the configuration resolved the path of every declared variant eagerly, so the job failed on the VSIX it had no reason to build. It passed locally only because a full run leaves both files behind. - Accept a function for `vsixPath` and call it from the setup project, so the path is resolved for the variant that runs rather than for all of them - Declare the projects of both variants unconditionally and leave the filtering to Playwright's `--project`. Deriving the list from `process.argv` never worked in the first place: the configuration is re-loaded in every worker process, whose arguments are the worker's Part of: eclipse-glsp/glsp#1682
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What it does
Aligns this repository with the glsp-core consolidation: the client, protocol, Node server and the integration-agnostic parts of the Workflow example and the Playwright framework are now consumed as published packages, and the VS Code e2e tests live here.
e2e/workspace directory with@eclipse-glsp/playwright-vscode, the published VS Code integration for the GLSP Playwright framework, and the privateworkflow-vscode-e2epackage holding the Playwright configuration and specs. Running them no longer requires glsp-playwright and glsp-server-node checkouts next to this repositoryUSE_WEB_EXTENSIONswitch, each installed into an extensions directory of its own, and resolve the webview by class so the web variant is testable at allpnpm/action-setupto v6.1.0, which is the first release supporting pnpm 12Also:
scripts/lint-report.mjs, replacing theeslint-json-to-mdstepAGENTS.md+.agents/skills/layout, and updateREADME.md,CONTRIBUTING.mdand the package READMEs to point at glsp-core instead of glsp-client and glsp-server-nodePart of: eclipse-glsp/glsp#1682
How to test
test:e2ebuilds the workspace, packages both VSIX variants and runs the desktop and web suites; single variants viapnpm e2e test:desktop/pnpm e2e test:web. The Workflow example itself can be launched as before from theWorkflow GLSP Example Extensionlaunch configurations.pnpm check:allandpnpm test:e2everified locally (both variants green)Follow-ups
@eclipse-glsp/playwright-vscodeis published from this repository for the first time; the first release run should confirm the publish job picks it upnextdependency versions are now exact, so nightly bumps arrive as reviewable diffs frompnpm upgrade:nextinstead of floatingChangelog
Breaking: the minimum Node version is raised to 24 (pnpm 12 for development), and the GLSP client/protocol dependencies are now sourced from glsp-core.