WebDriver/Meetings/2018-07-30

Agenda

dburns: We could make it pretty much at any time; If we choose London we can make use the office, and it is easy to travel; ACTION: Set up doodle

We discussed why geckodriver isn’t updated automatically on wpt.fyi.

dburns: Fixing some test failures, and after that it should be ready to land.; I also keep hitting different flake8 issues that I can't reproduce.
whimboo: Maybe different versions?
dburns: Could be.
ato: Are you going to release a new version after that?
dburns: Possibly.; I've asked people I know to run their tests with the patch, and things are looking good.
ato: It makes geckodriver look bad and we have a hard time communicating that this isn't a geckodriver bug.; New version would be really good.
whimboo: Does all Selenium clients have to make a new release when you ship a new Python client?
dburns: No, not for patch/dot releases.
whimboo: We have another failure with Windows, with the 10053 socket error.
ato: Yes, quite likely this is related. Systems have different networking stacks.; Let's reassess and see if the client patch fixes the issue.

whimboo: Do we have to propose discussion items?
dburns: Yes, I can create the agenda.
whimboo: I have some things to discuss that I want to have done before October.
ato: It is a problem that some vendors are unresponsive and that does make it hard to make decisions.; To not stagnate, we have to make a judgement call in some of these cases.; “Is this good for the web?”; jgraham made such a decision about one of jimevans’ PRs last week.

whimboo: We have a test which has two user prompts in sequence.; The test expects the first alert to close and to get the return value of the second.; It is not clear from the WebDriver specification if also the current substep of an algorithm should be aborted when a user prompt appears.
ato: It would be problematic to be more specific about how browser vendors implement this.; They might use techniques for intercepting the user prompt that are implementation specific.; Also doing it in the script execution environment using e.g. window.onerror is dangerous because the injected script could remove it.; The test seems to test the old behaviour of WebDriver, where it would invoke the user prompt handler.
whimboo: So the test is wrong?
ato: I think so.
whimboo: Maybe a good idea to keep the test but change the assertion, so that it is clear we abort after the first of the two alerts?
ato: That is a very good idea.

(Spoken status updates in bold.)