|Summary:||[GTK] Suspicious warning in http/tests/xmlhttprequest/connection-error-sync.html|
|Product:||WebKit||Reporter:||Alicia Boya García <aboya>|
|Severity:||Normal||CC:||bugs-noreply, calvaris, mcatanzaro|
|Version:||WebKit Nightly Build|
Description Alicia Boya García 2018-05-23 07:42:23 PDT
The following test has started failing on GTK in r230852:r230858: http/tests/xmlhttprequest/connection-error-sync.html This is probably a consequence of https://trac.webkit.org/changeset/230857/webkit The additional log we're getting is this: --- /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/http/tests/xmlhttprequest/connection-error-sync-expected.txt +++ /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/http/tests/xmlhttprequest/connection-error-sync-actual.txt @@ -1,3 +1,4 @@ +CONSOLE MESSAGE: line 22: XMLHttpRequest cannot load http://127.0.0.1:8000/xmlhttprequest/resources/infinite-loop.php due to access control checks. Test the behavior of a sync XMLHttpRequest that encounters an infinite redirection loop. Exception NetworkError; code=19; number=undefined (0); message=' A network error occurred.' Is that message expected?
Comment 1 Michael Catanzaro 2018-05-23 20:26:13 PDT
Well it's not a regression, according to Youenn's changelog: "There is a small change of behavior in the way we return an error. Instead of returning a platformBadResponseError, we are now returning the error as computed by NetworkLoadChecker. This allows getting some more error logging in the JS console." But as for whether the test is expecting its resource load to be blocked, I'm going to say no, that looks like a bug. It's frustrating that we don't know why the load was blocked. The same error occurs on tons of websites, but "access control checks" is a very vague reason. I remember talking to Alexey about this last year, we need better logging.
Comment 2 Michael Catanzaro 2018-05-23 20:27:09 PDT
Well, the load should be blocked, but because it is an infinite loop, not due to an access control check... right? Maybe that's an implementation detail leaking?