Currently, we use Github Actions to run our Testcafe suites, but the problem is running Safari tests on the macos-latest image, which is 10.15 Catalina, gives the following error:
UnableToAccessScreenRecordingAPIError: The find-window process cannot access the Screen Recording API.
This is because MacOS images don't come with that permission granted by default. Apparently you can't handle this programmatically, and there's a feature called SIP that is turned on in 10.15 Catalina that stops you from getting around it. This is discussed here.
Since we're relying on Github to provide images, they've not been able to provide us with a non-SIP image as discussed here. The workarounds provided there have been of no help.
Any other ideas?
As you already mentioned, TestCafe doesn't capture screenshots and resize them because of SIP enabled on machines with MacOS. However, it is possible to use the remote mode since the platform specific functionality is disabled by default. You can execute these steps:
testcafe remote test.js --hostname ${HOSTNAME} --ports ${PORT1},${PORT2} &
pid=$!
open -a Safari http://${HOSTNAME}:${PORT1}/browser/connect
wait $pid
Related
When I try to debug it shows this dialog:
Error dialog
And when I clock "fix" It shows this:
Web Browsers dialog
I can't solve this problem, any suggestion? Also I am developing a react native app. But i don't think that's the problem...
I am running Intellij Idea in Elementary OS.
This is expected - debugging is only supported in Chrome and browsers of the Chrome family, as it's clearly stated in Help. So, when you start the debugger, the IDE will try to open Chrome regardless of the browser chosen in run configuration. You need to make sure that Chrome is installed and the correct path to it is specified in Settings | Tools | Web Browsers.
We used to support Firefox Remote Debugging (but without source maps) through the FireFox Remote run configuration, but our solution doesn't work in the latest Firefox versions due to changes in the protocol, and there doesn't seem to be an easy way to fix it. We've made some progress recently, but there are still some blocking issues
related ticket: WEB-45986
I am aware that there is no official support for DRM-protected content (e.g. Spotify webplayer, netflix, etc.) when using headless chromedriver.
Has anyone here gotten it to work despite the aforementioned limitation?
I went as far as using a custom profile in headless chromedriver (one that should include WidevineCdm), but so far no luck.
Solution: replacing Chromiuim's default libwidevine binary with one extracted from ChromeOS.
Download the latest ChromeOS libwidevine binary and extract it.
Copy the extracted libwidevinecdm.so to /usr/lib/chromium-browser.
Change the access permissions to 755 (chmod 755 libwidevinecdm.so)
Credits to vpetkovs's blog post for finding this bright solution.
I have a Windows 10 laptop running a Windows 7 Virtual Machine. Inside the Virtual Machine, when i start the WebDriver, it gives the error gpu_process_transport_factory.cc<1009> Lost UI shared context when it is started:
IWebDriver driver = new ChromeDriver() //This causes the 1009 Error
Also then varies for Failed to query stereo recording and Failed to send GpuChannelMsg_CreateCommandBuffer and command_buffer_proxy_impll.cc<115> Shared memory handle is not valid.
This has been working for months and no changes were made (this WebDriver is the sole purpose of the Virtual Machine), the laptop running it is running fine (ie. no GPU problems). The WebDriver still navigates through pages, but giving further errors and 10x slower.
EDIT: Updated to ChromeDriver to 2.35 and still the same behavior.
Headless Chrome was shipped in Chrome 59 by Google Team which introduced a way to run the Chrome Browser in a Headless Environment.
A note was added as :
Headless mode has been available on Mac and Linux since Chrome 59. Windows support came in Chrome 60.
The article Getting Started with Headless Chrome mentioned that :
--disable-gpu \ # Temporarily needed if running on Windows.
A note was added as :
Right now, you'll also want to include the --disable-gpu flag if you're running on Windows.
As per the discussion Headless: make --disable-gpu flag unnecessary it was clear that :
The --disable-gpu flag is no longer necessary on Linux or Mac OSX. It will also become unnecessary on Windows as soon as the bug SwiftShader fails an assert on Windows in headless mode is fixed.
What happened under the hood?
As per the discussion headless: Switch from osmesa to SwiftShader as Google/Chromium team decided to ship SwiftShader with Chrome, the team thought to start using it to render GL content in Headless Mode. This required a couple of changes as follows :
Skip GPU data collection in Headless Mode since SwiftShader isn't considered a software implementation by that code which lead to a failure when we tried to retrieve information from the Window System.
Only skip GL initialization in InitializeStaticEGLInternal if we intend to use osmesa. SwiftShader requires initialization like the other non-software implementations.
SwiftShader is currently not supported on Mac OSX, so the team decided to continue to use the physical GPU in Headless Mode on that platform (unlike on other platforms where everything is software rendered).
So, to disable WebGL support in Headless Mode they decided to use --disable-gpu and --disable-software-rasterizer
The idea to Support WebGL in headless is still under discussion but SwiftShader fails an assert on Windows in headless mode with an error as :
[0117/125830.649194:ERROR:gpu_process_transport_factory.cc(1043)] Lost UI shared context.
DevTools listening on ws://127.0.0.1:37429/devtools/browser/1f0b2bf7-dfdd-44ac-9da7-f2659d352f0d
Conclusion
This error doesn't impact your #Test and you can ignore the error for the time being.
This appears to be an issue with the latest version of Chrome (65.0.3325.146). By rolling back to an earlier version of Chrome (64.0.3282.186) the issue disappeared for me.
Will investigate further and update here if I find any more info but as a temporary workaround you can uninstall chrome and install version 64.0.3282.186.
'Lost UI Shared Context' was mistaken for an error while it's actually a logging output.
This has been fixed since Chrome 72.0. no extraneous arguments needed.
I'm having troubles running Ruby/RSpec tests against Safari 11 when I'm trying to run tests via ssh manually or via Jenkins (where machine where Safari exists is remote slave).
When executing tests, I'm getting following error:
Selenium::WebDriver::Error::WebDriverError:
unable to connect to safaridriver 127.0.0.1:7050
What is weird is that I'm able to run tests when I'm logged to the Mac machine directly. This leads me to the conclusion that there could be some permission which, by default, disables execution from ssh session but not sure why?
Also, to my knowledge, Safari Driver is part of Safari 10+ and as such is not installed anymore as extension
Update: I've found out that safaridriver executable that should be spawned by tests cannot be spawned from some reason when I login via ssh.
Example:
/usr/bin/safaridriver -p 7050
It will just terminate with non-zero exit code while running same command directly on machine will run safaridriver in foreground. Since I expect Jenkins to be running test job on this machine, my idea to overcome this issue would be to launch safaridriver on this machine (by cron or launchctl) and then use this instance to connect to it with my tests. However, so far, I was not able to make my tests (Selenium/Capybara) re-use existing safaridriver instead of always trying to spawn new one on different port. Any idea on this would also be greatly appreciated.
This is my environment:
OS: MacOS Sierra 10.12.6
Browser: Safari 11.0.2
Thanks in advance
There is a possible workaround to do this. First you need to create a Automator Workflow or Apple Script that launches the SafariDriver
Then save this as a application. Let assume we name it SafariDriver7050.
Then from the SSH session you need to execute
open /Applications/SafariDriver7050.app
This will actually launch SafariDriver in the logged in session and it should work for you.
The caveat being shutting it down, you will need to first kill the SafariDriver7050 app and then you need to kill the safaridriver process. The order matters, else it will create a error dialog on UI
Edit-1:
As you suggested, it would be even easier to do this, when you wrap your test as a app and then it will be automatically be able to launch SafariDriver without any issues. The key to issue is using open command in a SSH session
Edit-2
Why does SafariDriver not work in SSH? Well if you look at the linux counterpart
In case of linux we can use the DISPLAY environment variable to launch an app in an existing display or we can use something like XVFB to launch the browser in a virtual display. That is the concept that most frameworks use in case of linux machines.
But Mac doesn't have such kind of feature, which is why this workaround is needed. Now why it doesn't have that, I am not sure. There may be some other workaround that I may not be aware of, so anyone who has valuable info, can help improve this part of the answer
For my case:
On OSX host side was created Automator app as described above and modified a little bit:
security unlock-keychain -p your_host_password /Users/$USER/Library/Keychains/login.keychain-db
safaridriver --enable
safaridriver -p 7050
On CCI/Jenkins side:
open /Applications/StartSafariDriver7050.app/
run pytest cmd
osascript -e 'quit app "StartSafariDriver7050"'
pkill safaridriver
In Python:
def safaridriver():
return webdriver.Safari(desired_capabilities=Caps.SAFARI, port=7050)
I seem to be having issues with integrating Xcode6 with jenkins, I currently have this setup and working with Xcode 5.
With xcode 6 running remotely via SSH the simulator time-out, when I run locally it succeeds.
Command
xcodebuild -workspace PROJECTNAME.xcworkspace -scheme BGO_Tests -destination 'platform=iOS Simulator,name=iPhone 5s' -derivedDataPath ./Build clean test
2014-08-19 10:46:36.591 xcodebuild[33966:381f] iPhoneSimulator: Timed out waiting 120 seconds for >simulator to boot, current state is 1.
Testing failed:
Test target BGO_Tests encountered an error (Timed out waiting 120 seconds for simulator to boot, current state is 1
Tested with recent Xcode 6 beta 6
Note: the device names changed in Xcode 7, so you no longer specify them using iPhone 5 (9.1 Simulator) but rather iPhone 5 (9.1).
Use xcrun instruments -s to get the current list of devices and then you can pre-launch it using:
xcrun instruments -w "iPhone 5 (9.1)" || echo "(Pre)Launched the simulator."
Prelaunching
I got to a point where what I proposed down there wasn't working anymore. In addition to making the changes mentioned here, you need to launch the simulator xcodebuild is expecting BEFORE xcodebuild is ran:
# First get the UDID you need
xcrun instruments -s
# Then launch it
open -a "iOS Simulator" --args -CurrentDeviceUDID <sim device UDID>
# and wait some time....
sleep 5
# Then launch your unit tests
xcodebuild [...] -destination 'platform=iOS Simulator,name=<device name matching the UDID>'
Old post
This bug is fixed in Xcode 6.3 and above. If you are experiencing similar problems in newer Xcode, it's likely another bug.
Apple follow up regarding Bug ID# 18001199:
The context provided by LaunchDaemons is not supported for running GUI
applications. The SSH service, and the default setup for Jenkins, are
both implemented as LaunchDaemons. In earlier versions of Xcode 5
xcodebuild could run tests on the iOS simulator in this context, but
that was never a supported configuration, and as you have noted that
is no longer working as of Xcode 6.
Unlike LaunchDaemons, LaunchAgents provide a context where you can run
GUI applications - if the user is logged in at the time, with a window
server / Aqua session. Converting your Jenkins configuration from
being a LaunchDaemon to being a LaunchAgent would avoid the reported
issue. You can also use launchd for running tests on the iOS simulator
from a SSH session, either by crafting a LaunchAgent and manually
loading / starting that, or by using "launchctl submit”.
Ok, after some more digging around the comments around here (many thanks to Opal), I found out that launching the slave via JNLP instead works.
As many people mentioned, it is not currently possible to run the unit test over SSH, so you might want to turn towards the JNLP agent for now until Apple fixes it.
If connecting with JNLP still does not solve it, try the solution mentioned in this comment.
i.e.: Run these on command line:
DevToolsSecurity -enable
sudo dscl . -append /Groups/_developer GroupMembership "user-that-runs-the-sim"
security authorizationdb write system.privilege.taskport is-developer
See References here and here.
I've recently found out that if you install a new version of Xcode and do not launch it. The simulator might start timing out again. To solve this, I've had to manually launch Xcode, and install the additional tools it requested.
I ended up solving this on Xcode 5 by doing the steps here, essentially running:
sudo security authorizationdb write system.privilege.taskport allow
This will eliminate one class of these authentication popups. You’ll also need to run:
sudo DevToolsSecurity -enable
However, once I upgraded to Xcode 6, I now get an infinite hang when trying to run xcodebuild tests over SSH. They continue to run just fine as long as I'm logged into the console, and running them from the keyboard.
I ran into the same issue. My working theory is that SSH on OSX is started as a LaunchDaemon, and LaunchDaemons are not allowed to present a UI; Reference.
I was able to work around the issue by using Java Web Start to launch the Jenkins slave. I then installed the Jenkins slave as a launchd service.
Unfortunately the Jenkins slave then installs itself as a -you've guessed it- LaunchDaemon, leading to the exact same problem of not being able to launch the tests; Reference.
I worked around that issue by moving the Jenkins Slave LaunchDaemon plist and jar files in /System/Library/LaunchDaemons into ~/Library/LaunchAgents, and updated the paths inside the plist file.
That finally allowed me to run XCode6 (Beta6) tests on an OSX jenkins slave.
I finally managed to find a good simple solution. JNLP was causing numerous issues with our jenkins server.
Workaround for SSH timeout via https://corner.squareup.com/2015/07/ios-build-infrastructure.html
"Mavericks (10.9) and Yosemite (10.10) determine if a process can access accessibility hooks via the parentage of the accessing process. By putting launchd in the list of allowed processes, processes launched via SSH or Jenkins have access to the accessibility hooks across the system. To do this you can modify the TCC database, per this gist. A reboot is required to make the change take effect."
#!/bin/bash
# This will add lauchd to the list of allowed processes for accessibility access
sudo sqlite3 /Library/Application\ Support/com.apple.TCC/TCC.db "INSERT or REPLACE INTO access VALUES('kTCCServiceAccessibility','/sbin/launchd',1,1,1,NULL)"
# This outputs the rows in the TCC database
sudo sqlite3 /Library/Application\ Support/com.apple.TCC/TCC.db 'select * from access'
echo "Restart is required for these changes to take effect"
Update 8/02/2016
This is now fixed in Xcode 7.2.1 ("Command line tool ‘xcodebuild test’ will no longer time out waiting for Simulator.app to launch")
I've seen this error before, one possibility is that since you probably downloaded the Xcode6 Beta from the internet (not the appstore as its not available yet), the machine you are trying to run it on will show a pop up asking you if you really want to open this app as its from the internet.
The same will happen when xcodebuild tries to launch the iPhone simulator app.
One thing you might want to try is to share screen with the machine and click "Open" in that pop up.
If that still doesn't work, I would try to:
Reset the Content & Settings of the simulator
Reboot the machine and make sure no simulator is running on start up (you can just choose not to re-open any app when restarting)