[🐛 Bug]: When taking a screenshot fails, the resulting NameError is misinterpreted as a lack of driver support for screenshots

[DanielHeath]

What happened?

Chromedriver sometimes returns {value: null} when you GET /session/<id>/screenshot instead of returning a string. I think that's probably a chromedriver bug.

In the selenium ruby bindings, in lib/selenium/webdriver/common/takes_screenshot.rb , any NameError is treated as "driver does not support screenshots".

When nil.unpack1('m') is called (because a null value was returned from chromedriver), this is reported as "driver does not support screenshots. This makes diagnosing the actual issue significantly more difficult!

How can we reproduce the issue?

Use chromedriver

Load a page with an iframe

Switch to the iframe

Take a screenshot

Relevant log output

[53, 62] in /home/daniel/.gem/ruby/3.2.2/gems/selenium-webdriver-4.12.0/lib/selenium/webdriver/common/takes_screenshot.rb
   53:         when :base64
   54:           full_page ? full_screenshot : screenshot
   55:         when :png
   56:           as = screenshot_as(:base64, full_page: full_page)
   57:           debugger unless as
=> 58:           as.unpack1('m')
   59:         else
   60:           raise Error::UnsupportedOperationError, "unsupported format: #{format.inspect}"
   61:         end
   62:       rescue NameError
(byebug) as
nil
(byebug) screenshot_as(:base64, full_page: full_page).length
253940

Operating System

Ubuntu

Selenium version

4.12.0 ruby

What are the browser(s) and version(s) where you see this issue?

Chrome 115

What are the browser driver(s) and version(s) where you see this issue?

Chromedriver 115

Are you using Selenium Grid?

No

[github-actions]

@DanielHeath, thank you for creating this issue. We will troubleshoot it as soon as we can.

---

Info for maintainers

Triage this issue by using labels.

If information is missing, add a helpful comment and then I-issue-template label.

If the issue is a question, add the I-question label.

If the issue is valid but there is no time to troubleshoot it, consider adding the help wanted label.

If the issue requires changes or fixes from an external project (e.g., ChromeDriver, GeckoDriver, MSEdgeDriver, W3C), add the applicable G-* label, and it will provide the correct link and auto-close the issue.

After troubleshooting the issue, please add the R-awaiting answer label.

Thank you!

[DanielHeath]

Looking further into this - it appears that if you use Timeout.timeout, the Net::HTTP client used to communicate with chromedriver can get into a broken state.

This is because Timeout.timeout asynchronously raises an exception in the calling thread, which interrupts operations which were not expected to be able to fail.

In ruby-3.2.2/lib/ruby/3.2.0/net/http.rb line 1855, the request is written to the socket; on line 1862 the response is read from the socket.

However, if a timeout exception occurs before the response arrives, the response is not read from the socket. This causes the next request made with that Net::HTTP client to read the response from the interrupted request (since it's queued on the socket and nothing else has read it).

[titusfortner]

So what code reproduces this?

Getting null back from the screenshot endpoint needs to be raised as a bug to Google.
Http client timeouts should end the session completely.

[DanielHeath]

So far, I can confirm (via mitmproxy) that chromedriver sends the correct response, so no bug needs to be raised to google.

By step-through debugging with mitmproxy in the other window, I can confirm that net/http returns the wrong response.

This is fixed if anything causes the socket to be closed and re-opened (eg if at least 2 seconds pass, as that's the keep-alive timeout in net/http).

Because Timeout uses ::Kernel.catch(instance) and throw(instance), the exception handlers in net/http are skipped.

I'm still working to get a minimal reproduction on that, but I expect to have it figured out soon.

[DanielHeath]

This is fixed in timeout 0.4.0 (see also rails/rails#29333 , ruby/timeout#30 )

[titusfortner]

So this is a bug in net/http? Presumably that is the library that is using the timeout method correctly.

[diemol]

Is there something for us to do here or can this be closed?

[titusfortner]

I went down a rabbit hole last time I looked at this trying to parse which libraries were supposed to be doing which things. What I recall is that net/http was the library that people agreed was properly handling the timeout issues. Can the user just require the latest version of timeout for this to work?

[titusfortner]

closing as I don't have enough information to know what needs to be done to fix something in selenium code.