Switching to JS file while Live Development is connecting causes it to fail

[peterflynn]

I could swear we fixed a bug just like this recently, but this occurs for me on master now...

1. Open an HTML file and a JS file in the working set
2. Go to the HTML file and click the Live Preview lightning bolt
3. While Chrome is launching, switch to the JS file

Result:
Chrome stays on the gray "interstitial" page forever, and Brackets eventually shows a connection error dialog.

Expected:
Should still connect, since I was on a valid HTML file when I originally clicked it.
And if I wait on the HTML file until the connection succeeds, switching to the JS file at that point doesn't cause it to disconnect... so we're putting the user in a sort of race condition.

This is easy to hit if you're using Live Preview but what you actually want to work on is a CSS or JS file. You can't launch it from the file you're in (because of the "you must select an HTML file first" issue), so you quickly switch to the HTML file and then back to the file you actually care about.

Also, it's easier to hit on Win than Mac, because every Live Preview invocation means waiting for an instance of Chrome to launch from scratch -- much longer time period in which to hit the bug.

[peterflynn]

This didn't really work in earlier sprints either -- in Sprint 29, it causes Chrome to show the JS file's source in the browser tab, and in Sprint 30 the current bug (above) repros. So maybe not a high priority...

[lkcampbell]

@peterflynn, I've actually run into this issue a few times myself, performing the exact scenario you describe above. When I did it last, I was using Sprint 29, and I thought the display of the source code was an odd, but intentional, way to address the problem.

Your Expected result make a lot more sense though. Especially since this behavior is now causing an error in the connection.

[njx]

Medium priority to @couzteau

[couzteau]

I observed reviewing the launch code that the helper function _doLaunchAfterServerReady is calling _getCurrentDocument again - after the open function has already retrieved a ref top the current document.

I have set up a branch with a fix: jhagenst/fix-5110

I think the fix is good, but I have only seen a small fraction of the LiveDebugger code, hence I have the feeling I don't have sufficient insight. Thanks for a comment on wether the fix seems to do the right thing.

[couzteau]

posted PR #5268

[njx]

FBNC to @peterflynn

[peterflynn]

Waiting until the "index.html-finding" code lands before regressing this

[pthiess]

@peterflynn Please regress.

[peterflynn]

@couzteau When I follow the repro steps now, Chrome closes when I do step 3. I would expect it to continue connecting to the HTML file to that the outcome is the same as if I waited a few seconds more before clicking the JS file.

This is less nasty than getting stuck on the loading screen forever, so changing priority to Low... but reopening. Leaving Sprint unassigned for now.

[peterflynn]

Btw, it's possible this was working better when @couzteau's PR #5268 landed, but was broken subsequently when #5414 landed.

[couzteau]

@peterflynn I cannot repro what you describe. Can you show me. I do agree that it is not unlikely that there is relationship to #5414