Usage on GitHub Actions with xvfb #147
Comments
|
I had to clean up old xvfp processes but this method has been working for me: Haven't had a chance to play with browser-run yet in actions. |
|
fwiw, its an action https://github.com/marketplace/actions/cleanup-xvfb. Let me know how it goes if you try it out. |
|
Would it be possible to create an action that wraps Before - run: sudo apt-get install xvfb
- name: npm install, build, and test
run: |
npm i
xvfb-run --auto-servernum npm test
env:
CI: true
- name: Cleanup xvfb pidx
uses: bcomnes/cleanup-xvfb@v1After - name: Test with xvfb
uses: run-with-xvfb
with:
run: npm test
env:
CI: true |
|
I recently set up xvfb with the "Before" method and it worked well. But yes that action looks really cool...I guess it would need to install |
|
I might give this a shot, but since it's your idea, if you want to, please go first! |
|
I’ve been using actions a bit more an the cleanup might not even be needed. Not sure what was happening before. Running in headless ci, you need to set up a frame buffer for chrome to work. Unless that responsibility is pulled into browser run, setting up per run is the way to go. |
Same for me, this currently works for headless electron: https://github.com/hypergraph-xyz/desktop/blob/master/.github/workflows/test.yml |
I just published an action like you suggested. To clean the process I used @bcomnes's script. I added an author header to it. Even tho the cleaning is not really necessary in practice, I wanted the action to feel isolated. Because I saw @juliangruber conditions his workflow based on platform I added that condition inside the action so your workflow ends up cleaner, if you're not using linux then the command will be executed without using xvfb. In this commit to @juliangruber's repo you guys can see the differences (what before was 7 lines, now can be done with 3). I added the action in the official electron documentation. Later I'll add a |
|
This is fantastic work @GabrielBB! |
|
Since I ended up here and I figured others might as well, this is my solution: jobs:
build:
env:
DISPLAY: :0
steps:
- name: Setup xvfb (Linux)
if: runner.os == 'Linux'
run: |
sudo apt-get install -y xvfb libxkbcommon-x11-0 libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-randr0 libxcb-render-util0 libxcb-xinerama0 libxcb-xinput0 libxcb-xfixes0
# start xvfb in the background
sudo /usr/bin/Xvfb $DISPLAY -screen 0 1280x1024x24 &I used this to test a PyQt5 application. |
We have 84 times this kind of exceptions in 'ajde' tests in our build logs on GitHub, even though the tests seem to pass: HeadlessException: No X11 DISPLAY variable was set, but this program performed an operation which requires it. I found this discussion: juliangruber/browser-run#147 And then this GitHub Action: https://github.com/marketplace/actions/gabrielbb-xvfb-action Signed-off-by: Alexander Kriegisch <Alexander@Kriegisch.name>
* update tslib dependency in packages/*/package.json
* add Makefile to .editorconfig
* update packages and lerna config
* update base tsconfig target to ES6
* add several Makefile targets and tools/build.mk with sample esbuild config
* rough-out makefiles with peer dependency compilation
* change peer dependencies to newlines that seem to work better with Makefile targets
* Update make targets for proper Makefile cross-package dependencies
* add the commonjs/esm targets
* add tools for the package template
* remove references to config.yaml
* update Makefile for repackaging; pull package version from lerna.json (for now)
* repackage all package.json files
.. removing `scripts`, `standard`, and other unused package.json fields
* remove `standard` from `repackage`
* improve logging and peer dependency checking
* Remove numerous references to rollup
... and the many workarounds that are now replaced by esbuild
* make) update the Makefile `test` targets
and add test_{package}
* update repackage to export "." and "./helpers"
* repackage with dist/index.js as esm import, dist/index.cjs commonjs
* Fix build/peer dependencies building
* Fix karma testing config for esbuild
* Fix top-level Makefile packaging
* Fix utils.jsx imports / dependencies
* make) add esm dependency to testing; add watch target
* change spec imports from ../src to ../dist
... so we don't have a double-import problem
* consolidate the karma config to tools/karma.conf.js
* fix builder missing karma/frameworks
* tools) fix broken ../{pkg}/dist/*.js
* Move packages/build.knockout to builds/knockout
* Move packages/build.reference to builds/reference
* Fix tests hanging b/c packages/builder had none
* set global export for ko; update knockout builds
* Add a stub for commonjs=>iife exports
* Add BUILD_VERSION to the exports
* tools) Update & simplify our repackaging with lerna
* add builds/* to the workspaces/clean target
also remove unneeded targets
* fix iife/browser build & BUILD_VERSION
* more fixes of BUILD_VERSION
* build/reference) add tests of iife/root imports
* root) add workspaces builds/; add lerna stream comments
* fix ReferenceError with globals
* make/tools) remove peer_dependenicies in lieu of lerna build
* ci) switch to the circleci orb for saucelabs
* Updated config.yml
* ci) add job
* ci) add `make ci` target and Github workflow
* ci) fix make target
* bind) clean up some imports
* ci) remove peer devDependencies from the monorepo that are used for tests
... because lerna seems to order its topology by dependencies and devDependencies as equivalent
* make) change lerna concurrency to a variable
* Merge branch '130-1-package-update' of github.com:knockout/tko into 130-1-package-update
* Revert "ci) remove peer devDependencies from the monorepo that are used for tests"
This reverts commit b2fcedf.
* ci) use concurrency=1 for build/exec
* ci) change devDependencies to dependencies or peerDependencies
... to avoid circular loops in the Lerna topology algorithm
* make) ensure that the node_modules is built
* workflows) switch to node 16
* ci) add sauce env vars
* ci) remove package-lock.json dupe; start to simplify SauceLabs config
* ci) disable startConnect by default
* ci) disable Safari as it always tseems to time out
* ci) disable SauceLabs, enable local testing
... so we're no longer blocked
* ci) disable SL on pull; set test concurrency=1
* ci) try xvfb-run on Github actions
per:
juliangruber/browser-run#147
* ci) install electron / npm
* ci) try GabrielBB/xvfb-action tests
* ci) switch back to SauceLabs w/o recording/screenshots
... also set test concurrency to 1
* ci) disable CI altogether until we DD it
I tried to install/configure
xvfbon GitHub actions to usebrowser-run/tape-runthere, but I was not successful.If anyone finds a solution, please advise. This is the simplest configuration that's only missing
xvfbI wrote all of this and it exits with
0butbrowser-rundoesn't output anything. The code is not run at all, invalid JS will also result inbrowser-runexiting successfully.The text was updated successfully, but these errors were encountered: