WebDriver/Meetings/2018-09-17

Agenda

• Minutes from last week lost due to Etherpad EOL
• geckodriver 0.22.0 released
• Work week planning
• Intermittent failures
  • [fixed] EdgeNeedsSweepUnbarrieredSlow assertions and crashes (Bug 1482029)
  • [fixed] Crashes in web-platform-tests are now correctly identified (Bug 1485259)
  • [needs fixed] test_loader_global_sharing.py causes shutdown / startup hang of Firefox (Bug 1444600) - caused by Shutdown hang due to JavaScript error in SessionStore.jsm (Bug 1489960)
  • [needs fix] (Perma ccov) Assertion failure: m_refCount == 1 or pages_.empty() (Bug 1485985)
  • [needs investigation by devs] Firefox startup hang for ASAN builds on Linux64 (Bug 1487243)
  • [needs investigation] Assertions in ExecutableAllocator.cpp due to GC leaks (Bug 1486101)
  • Wdspec jobs now also run on OS X \o/ (so far only a few low intermittent failures!)
  • There are a couple of crashes in WdSpec which we currently cannot identify due to missing crash information (needs Bug 1490906 and Bug 1433495)

Minutes

Meeting minutes

ato

Were lost, but now Etherpad is working.

whimboo

Can continue using this Etherpad until it does not work.

Remember to copy minutes before closing.

geckodriver 0.22.0 released

ato

Problem with a change in mozprofile, because of relative paths for mozrunner’s mozprofile dependency.

It will not be a problem when we are going to release from TaskCluster.

whimboo

I reached out to barancev and we found one regression in PointerOrigin.

The Python tests that I ran was also fine.

We did not reach out to the Ruby client maintainers, and they didn’t specify the duration.

This was a regression, because duration is supposed to be optional.

Our binary sizes are smaller on TaskCluster than from Travis.

geckodriver on ARM

whimboo

The only platform we don’t support on TC is Android (ARM).

ato

Yes, I talked to AutomatedTester about this last week.

Since geckodriver is in a pre-relase state we have the option to drop ARM support before 1.0.

We can’t do it afterwards, because we’ve committed to a certain level of stability and continuation.

AutomatedTester

We’re also the only vendor who supports an ARM build.

whimboo

There are also more changes we need for GeckoView, but these should not block a geckodriver 1.0 release.

geckodriver signing

whimboo

What would be the next steps for releasing geckodriver from TC? Is it signing?

ato

Yes! :)

Signing part is two-folded. On Windows we sign the binary, and on OS X and Linux we sign the archive

We should let RelEng know that we have Windows builds which can be signed

Linux is more complicated, because ....

We have to refactor our build process to integrate it into the build process (build.rs)

whimboo

For nightly all geckodriver binaries are in the common tests package, whereby for releases we need our own packages

ato

Could we publish these as separate artifact/downloads?

whimboo

I would like that too.

ato

Packaging remains. We should get jlorenzo started on Windows signing.

whimboo

Lets remember that we do not run wdspec tests on Windows in TC yet due to problems with structured logging

ato

There’s four milestones to do:

Four milestones:

1. Signing

2. Packaging

3. Windows tests

4. 1.0 release

Work week planning

whimboo

Lets start 9am in the morning

left - end of the corridor

topics:

* webdriver spec (lets do with David - maybe Wednesday)

* window tracking refactoring

* maybe signing work

* get MINIDUMP_SAVE_PATH support in geckodriver

* bug triaging

ato

Especially of Testing :: Marionette P2 and P3 bugs that block WebDriver conformance.

Intermittent failures

whimboo

Three things are fixed.

(1) an assertion that caused WPT failure;

(2) global sharing in Marionette caused a hang during shutdown;

(3) ?

m_Refcount == 1 still waiting on info from kmag.

The biggest failure at the moment is hang on x86_64 ASAN startup.

… also fails on mochitest, really anything where Marionette is enabled.

wdspec tests are now enabled on WPT, but no intermittents

Wd jobs are now split in two chunks, which will get

I would really like to see a fix for https://bugzilla.mozilla.org/show_bug.cgi?id=1433495.

This would help getting crash dumps.

Status updates

(Spoken status updates in bold.)

• ato
  • geckodriver in default build (1471281)

• whimboo
  • [wdspec] Cleanup manifest files (Bug 1488347)
  • [wdspec] Modifier key not correctly set due to difference of platform name between sys.platform and capabilities (Bug 1488681)
  • [wdspec] Enable Wd job on MacOS (Bug 1480342) and split them into two chunks for all platforms (Bug 1488682)
  • [geckodriver] Hardening handshake checks for Marionette connections (Bug 1461608)
  • [geckodriver] Regressions in Java Selenium test suite for Serde changes in geckodriver (Bug 1489141)
  • [geckodriver] "Pause" action primitive should support undefined as duration (Bug 1491751)
  • [marionette] Check for crashes if no connection can be established in raise_for_port() (Bug 1488820)
  • [marionette] Missing focus events for element interaction when Firefox is in background (Bug 1398111)
  • [marionette] Provide JSON serialisation of Window object (Bug 1274251)
  • [mozprofile] Remove testing related preferences (Bug 905404)
  • [Firefox] Investigated missing Symbol.iterator on Arguments in sandbox (Bug 1453057)
  • [Firefox] Keep focus on Chrome window when switching to a different application with focusmanager test mode enabled (Bug 1489967)

PTO/travel (🏖)

• ato away Thursday 20th