Two tests either submit a form or make the driver controller refresh somehow, making the browser request a new test before finishing the current one.

LayoutTests/fast/dom/HTMLButtonElement_change-type_t01: Skip # Breaks the test infrastructure. Issue #18558
LayoutTests/fast/dom/shadow/form-in-shadow_t01: Skip # Breaks the test infrastructure. Issue #18558.

This shows up in the error messages from test.py with the error

"Browser requested next test before reporting previous result"
and the offending test is listed as "The test before that", not as "Old test", because the offending test finishes, reports its result, requests a new test, and the new test is returned from the browser controller, before the refresh of the driver page happens asynchronously, so one new test has already been loaded before the bad request comes in.

This problem seems inherent in any test that makes the parent window refresh. We have some ideas to change the driver page to avoid this, but they may not fix this.

Removed Priority-High label.
Added Priority-Low label.

Issue #20578 has been merged into this issue.

There are new tests causing this problem. I will use divide-and-conquer on Android builder, to find the tests that are doing it. Suppressing

LayoutTests/fast/filesystem and
LayoutTests/fast/css-generated-content

to start with, since they are the two sets of tests that precede this problem happening.

Removed Priority-Low label.
Added Priority-Medium label.

First, we will suppress three suspicious tests that reload the window, to see if that solves the problem: LayoutTests/fast/

filesystem/file-after-reload-crash_t01.dart:
css-generated-content/malformed-url_t01.dart
css-generated-content/bug91547_t01.dart

There is still a general suppression for LayoutTests/fast/dom, which needs to be removed, after the problematic tests have been found.

We have found the problem in our test scripts that makes them vulnerable to this - it was actually a problem with the test starting twice, because it refreshed itself. This could allow it to finish twice, and if one of these "done" reports got a "Wait" when it fetched a new test, the new test timer fired again after half a second during another test.

The test scripts now will print an error message that a test has reloaded itself and started twice, and fail the test. Only 6 tests in co19 are now doing this, and a co19 issue has been filed on them. They are skipped on dartium in the status file. When we run browser tests on them, they should be skipped in those files too.

Added Fixed label.