http/tests/security/cross-frame-access-put.html is racy

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Alexey Proskuryakov]

It failed for me under ASan.

[Alexey Proskuryakov]

Created attachment 243942 [details]
proposed fix

[WebKit Commit Bot]

Attachment 243942 [details] did not pass style-queue:


ERROR: LayoutTests/ChangeLog:1:  ChangeLog entry has no bug number  [changelog/bugnumber] [5]
Total errors found: 1 in 4 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Alexey Proskuryakov]

Created attachment 243944 [details]
with a fixed ChangeLog

[Darin Adler]

Comment on attachment 243944 [details]
with a fixed ChangeLog

View in context: https://bugs.webkit.org/attachment.cgi?id=243944&action=review

> LayoutTests/http/tests/security/cross-frame-access-put.html:30
> +    // Run the test in main frame after subframe part finishes.
> +    setTimeout(test, 0);

Does this comment mean that we are relying on the fact that the subframe test runs in a zero-duration timer that has already been scheduled, and that all such timers run in the order they are scheduled? If so, then I think the comment is too brief to be clear on this.

> LayoutTests/http/tests/security/resources/cross-frame-iframe-for-put-test.html:248
> +        // This complicates synchronization with main frame, which needs to wait for this
> +        // code to run.

I find this comment frustrating. It states that running on a timer “complicates” things, but not how we deal with the complexity. It would be better to state clearly what we do to make the test work, and optionally give a reason why other simpler techniques would not work. It’s not important for us to say this is complicated.

[Alexey Proskuryakov]

Committed <http://trac.webkit.org/r177880>.

Agreed about the comments, re-worded them before landing.