715
edits
WebDriver/RemoteProtocol/Meetings/2019/07/23: Difference between revisions
Jump to navigation
Jump to search
WebDriver/RemoteProtocol/Meetings/2019/07/23 (view source)
Revision as of 16:56, 23 July 2019
, 23 July 2019→Minutes: do
(→Roster: fill in) |
(→Minutes: do) |
||
| Line 18: | Line 18: | ||
== Minutes == | == Minutes == | ||
=== Puppeteer client workarounds and changes === | |||
The extent of the changes we’ve made to the Puppeteer client so far are limited to the necessary workarounds for getting the tests barely running with Firefox. The changes are almost exclusively limited to the lib/Launcher.js file, which deals with finding the browser binary, profile/<code>user-data-dir</code> tweaks, and starting it as a subprocess. | |||
Beyond this we’re expecting the Firefox CDP implementation not to diverge on the interface level from Chrome. For certain domains—such as console- and network logging—we may see the server return messages with varying ''content'', but the ''interaction'' from the client’s point of view should remain 1:1 comparable with Chrome. | |||
Maintaining a single client, with abstractions for Chrome’s and Firefox’ launcher code, would likely be beneficial. We may however have to maintain a fork of Puppeteer until we have something more tangible to upstream, since we don’t want to block upstream Puppeteer from releasing newer versions. We should however work in such a way that we maintain Chrome support in the fork, rather than replacing it. | |||
=== Brief update from security review meeting === | |||
The security testing meeting went fine, and we answered some questions. Not heard back after this, but we expect sec-review to not be a problem. | |||
=== Gutenberg progress === | |||
See [https://docs.google.com/spreadsheets/d/1rCKRDW4WNZXs2f9-Bm9CMuV9EAUXoHGHbD3GHYY6zdo/edit#gid=15969737 spreadsheet for currently failing Gutenberg tests]. | |||
Many Gutenberg tests failing because of unexpected console logs. Unlike WebDriver, Puppeteer/CDP does not have ''classes'' or ''variant types'' of errors, which means tests typically rely on parsing the error message string. | |||
Since we can’t change the messages for these errors without risking web compat problems for the web platform, we must find a way to work around this in order to pass more Gutenberg tests. One particular worry is that Gutenberg is not an isolated case, and that other test suites in the wild do the same thing. | |||
Ideally, it was argued in the meeting, the different non-platform errors, for example from the CSS parser, would be uniquely identifiable and defined in a web standard. This would enable users of Puppeteer to work with console errors in a more reliable way across browsers. It is worth airing this idea with other vendors at a future point, but does not offer any practical, immediate solution. | |||
Other alternatives are to hide messages on the server side, or the push the responsibility on to the test writer. The conclusion from the meeting was to do the latter, which likely means in this case that we will have to upstream some patches to the WordPress project to include a couple of Firefox-specifc error ignores in order for us to pass more tests. | |||
=== Current work/coming up next === | |||
* ato still working on fixing up Puppeteer-tests-on-try patches to address code review issues raised by build peers | |||
* ochameau has ceased work on the remote agent until new developers from Interop are joining, in order to maximise value of knowledge transfer | |||
* jdescottes has one patch that still needs review, dealing with JavaScript dialogues, but expects to transition now to work on Fission in DevTools | |||
== Changelog == | == Changelog == | ||