Bugzilla

My current understanding why the tests are failing:

* `test_preload_csp_imgsrc_none`:
  * The early hint response with the hint to the image resource and the final http response are sent with 0ms delay in-between.
  * The final http response doesn't need to load the image in the html page, because of `Content-Security-Policy: img-src 'none'`.
  * the page load is completed before the early hint request is sent.
  * Tab gets closed (and the early hint request gets cancelled/or requesting the count is faster than requesting the image in the preloader).
  * No early hint preload is counted.
  * Possible workarounds:
    * allow either one or zero preload requests - might have influence on other test cases if the preload is made when the next test case is started, might be unlikely or impossible
    * add delay in page load to allow the preload to be made within the time frame (10-15ms) - code is still racy, but might avoid the intermittent for now
* Other test failures, where a normal response is made instead of an early hint response:
  * If the channel is created first from the main document (here again because it is available instantly after the early hint)
  * AFAIK the main document is loaded in the child process
  * The early hint request is made on the parent process
  * The test failure occurs if the child process is faster in creating the channel to load the resource.

My current understanding why the tests are failing:

* `test_preload_csp_imgsrc_none`:
  * The early hint response with the hint to the image resource and the final http response are sent with 0ms delay in-between.
  * The final http response doesn't need to load the image in the html page, because of `Content-Security-Policy: img-src 'none'`.
  * the page load is completed before the early hint request is sent.
  * Tab gets closed (and the early hint request gets cancelled/or requesting the count is faster than requesting the image in the preloader).
  * No early hint preload is counted.
  * Possible workarounds:
    * allow either one or zero preload requests - might have influence on other test cases if the preload is made when the next test case is started, might be unlikely or impossible
    * add delay in page load to allow the preload to be made within the time frame (10-15ms) - code is still racy, but might avoid the intermittent for now
* Other test failures, where a normal response is made instead of an early hint response:
  * AFAIK the main document is loaded in the child process and requests the image
  * The early hint request is made on the parent process
  * The test failure occurs if the child process is faster in creating the channel to load the resource (here again because it is available instantly after the early hint)

My current understanding why the tests are failing:

* `test_preload_csp_imgsrc_none`:
  * The early hint response with the hint to the image resource and the final http response are sent with 0ms delay in-between.
  * The final http response doesn't need to load the image in the html page, because of `Content-Security-Policy: img-src 'none'`.
  * the page load is completed before the early hint request is sent.
  * Tab gets closed (and the early hint request gets cancelled/or requesting the count is faster than requesting the image in the preloader).
  * No early hint preload is counted.
  * Possible workarounds:
    * allow either one or zero preload requests - might have influence on other test cases if the preload is made when the next test case is started, might be unlikely or impossible
    * add delay in page load to allow the preload to be made within the time frame (10-15ms) - code is still racy, but might avoid the intermittent for now
* Other test failures, where a normal response is made instead of an early hint response:
  * AFAIK the main document is loaded in the child process and requests the image
  * The early hint request is made on the parent process
  * The test failure occurs if the child process is faster in creating the channel to load the resource (here again because it is available instantly after the early hint)
* I've no explanation why the iframe test doesn't fire any loads.