101268 2012-11-05 15:24:40 -0800 Fixed flaky stack overflow test: fast/xmlhttprequest/xmlhttprequest-recursive-sync-event.html 2014-12-22 13:20:41 -0800 1 1 1 Unclassified WebKit Tools / Tests 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED https://bugs.webkit.org/show_bug.cgi?id=139840 P2 Normal --- 1 mark.lam mark.lam oldest_to_newest 758956 0 mark.lam 2012-11-05 15:24:40 -0800 Test fast/xmlhttprequest/xmlhttprequest-recursive-sync-event.html is flaky. Here's how we can make it not flaky: // This test is flaky because it depends on the size of the available native // stack, the size of native stack frames, and the layering of function calls // in the JavaScript VM code. Together, these variables determine when a // StackOverflowError will be thrown in the course of this test of unbounded // recursion. // // From experience with this test, StackOverflow error tends to be thrown in // 2 possible scenarios: // // Scenario 1: Overflowed while calling xhr.open() // =============================================== // In this scenario, the VM is executing xhr.open() and succeeds in setting // up the new request. In response to having opened the request, it tries to // dispatch an event with readyState 1 (OPENED). Unfortunately, there is not // enough stack and we get an overflow error and 1 console error message here. // // The error is thrown but is then eaten up by the EventDispatcher. Hence, // onreadystatechange() back in JS land does not even see this error. // // open() eventually returns to onreadystatechange(), and we call xhr.send() // next. xhr.send() overflows the stack again, and we get a 2nd error and // a 2nd console error message. // // Note: the lastReadyStateObserved in this case is 4 because the event // dispatch for OPENED failed to execute onreadystatechange(). // // The total number of error messages seen is 2. // // Scenario 2: Overflowed while calling xhr.send() // =============================================== // In this scenario, the VM is executing xhr.send() and a readyState change // to 4 (DONE) occurs. An event is dispatched and the VM tries to call // onreadystatechange() but overflows the stack. Here, we see 1 error and // 1 console error message. // // Since we never succeeded in reentering onreadystatechange(), we did not // not set up anymore requests, and will not attempt to re-enter // onreadystatechange() after this. Hence, we will not see anymore overflow // errors nor console error messages. // // In contrast with scenario 1, we did not see an overflow error for xhr.open() // before we saw the error for xhr.send(). // // Note: lastReadyStateObserved in this case is 1 because the event // dispatch for OPENED succeeded in executing onreadystatechange(), and // lastReadyStateObserved was set to 1. // // The total number of error messages seen is 1. // // Flakiness // ========= // Since we can end up with scenario 1 or 2 depending on variables outside the // control of this test, the result of the test would be inherently flaky i.e. // we can see 1 or 2 console error messages. // // That is unless we do some compensation. In the case of scenario 2, we can // check if the lastReadyStateObserved is 1 which means we would have only seen // one console error message. If so, we can call xhr.send() again just to // trigger another one so that we'll have a total of 2. // // Note that in the compensation case, we'll need to call xhr.open() again // before calling xhr.send(). This is because the previous xhr request has // already been closed. Here, xhr.open() should not trigger an overflow error // just like it didn't before the previous xhr.send() triggered the error (the // fact that it succeeded without an error before means it will succeed this // time too). And this leaves it up to our compensating call to xhr.send() // to generate the 2nd expected overflow error and console error message. // // Note also that we'll need to make sure that we do this compensation only // once, and not repeatedly do the compensating xhr.send() call on every // onreadystatechange() frame on the return path. 758960 1 172420 mark.lam 2012-11-05 15:32:31 -0800 Created attachment 172420 Fix. 758987 2 172420 ggaren 2012-11-05 16:11:40 -0800 Comment on attachment 172420 Fix. Nice. I like this much better than just marking the test flaky in the testing harness. 759008 3 mark.lam 2012-11-05 16:25:41 -0800 Landed in r133540: <http://trac.webkit.org/changeset/133540>. 172420 2012-11-05 15:32:31 -0800 2012-11-05 16:11:40 -0800 Fix. bug-101268.patch text/plain 6174 mark.lam SW5kZXg6IExheW91dFRlc3RzL0NoYW5nZUxvZwo9PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBMYXlvdXRUZXN0cy9D aGFuZ2VMb2cJKHJldmlzaW9uIDEzMzUzNSkKKysrIExheW91dFRlc3RzL0NoYW5nZUxvZwkod29y a2luZyBjb3B5KQpAQCAtMSwzICsxLDEzIEBACisyMDEyLTExLTA1ICBNYXJrIExhbSAgPG1hcmsu bGFtQGFwcGxlLmNvbT4KKworICAgICAgICBGaXhlZCBmbGFreSBmYXN0L3htbGh0dHByZXF1ZXN0 L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2ZW50Lmh0bWwuCisgICAgICAgIGh0dHBz Oi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD0xMDEyNjguCisKKyAgICAgICAgUmV2 aWV3ZWQgYnkgTk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgKiBmYXN0L3htbGh0dHByZXF1ZXN0 L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2ZW50LWV4cGVjdGVkLnR4dDoKKyAgICAg ICAgKiBmYXN0L3htbGh0dHByZXF1ZXN0L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2 ZW50Lmh0bWw6CisKIDIwMTItMTEtMDUgIEFsaWNlIEJveGhhbGwgIDxhYm94aGFsbEBjaHJvbWl1 bS5vcmc+CiAKICAgICAgICAgVW5yZXZpZXdlZCBnYXJkZW5pbmcuIE1hcmsgZm9udHMvY3Vyc2l2 ZS5odG1sIGFzIGZsYWt5IG9uIE1vdW50YWluTGlvbiBhcyB3ZWxsIGFzIExpb24uCkluZGV4OiBM YXlvdXRUZXN0cy9mYXN0L3htbGh0dHByZXF1ZXN0L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1z eW5jLWV2ZW50LWV4cGVjdGVkLnR4dAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBMYXlvdXRUZXN0cy9mYXN0L3ht bGh0dHByZXF1ZXN0L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2ZW50LWV4cGVjdGVk LnR4dAkocmV2aXNpb24gMTMzNTMxKQorKysgTGF5b3V0VGVzdHMvZmFzdC94bWxodHRwcmVxdWVz dC94bWxodHRwcmVxdWVzdC1yZWN1cnNpdmUtc3luYy1ldmVudC1leHBlY3RlZC50eHQJKHdvcmtp bmcgY29weSkKQEAgLTEsNCArMSw1IEBACiBDT05TT0xFIE1FU1NBR0U6IFJhbmdlRXJyb3I6IE1h eGltdW0gY2FsbCBzdGFjayBzaXplIGV4Y2VlZGVkLgorQ09OU09MRSBNRVNTQUdFOiBSYW5nZUVy cm9yOiBNYXhpbXVtIGNhbGwgc3RhY2sgc2l6ZSBleGNlZWRlZC4KIFRoaXMgdGVzdHMgdGhhdCBo YXZpbmcgaW5maW5pdGUgcmVjdXJzaW9uIGluIFhNTEh0dHBSZXF1ZXN0IGV2ZW50IGhhbmRsZXIg ZG9lcyBub3QgY3Jhc2guIAogUEFTUwogCkluZGV4OiBMYXlvdXRUZXN0cy9mYXN0L3htbGh0dHBy ZXF1ZXN0L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2ZW50Lmh0bWwKPT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PQotLS0gTGF5b3V0VGVzdHMvZmFzdC94bWxodHRwcmVxdWVzdC94bWxodHRwcmVxdWVzdC1yZWN1 cnNpdmUtc3luYy1ldmVudC5odG1sCShyZXZpc2lvbiAxMzM1MzEpCisrKyBMYXlvdXRUZXN0cy9m YXN0L3htbGh0dHByZXF1ZXN0L3htbGh0dHByZXF1ZXN0LXJlY3Vyc2l2ZS1zeW5jLWV2ZW50Lmh0 bWwJKHdvcmtpbmcgY29weSkKQEAgLTcsNiArNyw4MSBAQCBmdW5jdGlvbiBsb2cocykKICAgICBs b2dEaXYuYXBwZW5kQ2hpbGQoZG9jdW1lbnQuY3JlYXRlRWxlbWVudCgiYnIiKSk7CiB9CiAKKy8v IFRoaXMgdGVzdCBpcyBmbGFreSBiZWNhdXNlIGl0IGRlcGVuZHMgb24gdGhlIHNpemUgb2YgdGhl IGF2YWlsYWJsZSBuYXRpdmUKKy8vIHN0YWNrLCB0aGUgc2l6ZSBvZiBuYXRpdmUgc3RhY2sgZnJh bWVzLCBhbmQgdGhlIGxheWVyaW5nIG9mIGZ1bmN0aW9uIGNhbGxzCisvLyBpbiB0aGUgSmF2YVNj cmlwdCBWTSBjb2RlLiBUb2dldGhlciwgdGhlc2UgdmFyaWFibGVzIGRldGVybWluZSB3aGVuIGEK Ky8vIFN0YWNrT3ZlcmZsb3dFcnJvciB3aWxsIGJlIHRocm93biBpbiB0aGUgY291cnNlIG9mIHRo aXMgdGVzdCBvZiB1bmJvdW5kZWQKKy8vIHJlY3Vyc2lvbi4KKy8vCisvLyBGcm9tIGV4cGVyaWVu Y2Ugd2l0aCB0aGlzIHRlc3QsIFN0YWNrT3ZlcmZsb3cgZXJyb3IgdGVuZHMgdG8gYmUgdGhyb3du IGluCisvLyAyIHNjZW5hcmlvczoKKy8vCisvLyBTY2VuYXJpbyAxOiBPdmVyZmxvd2VkIHdoaWxl IGNhbGxpbmcgeGhyLm9wZW4oKQorLy8gPT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT0KKy8vIEluIHRoaXMgc2NlbmFyaW8sIHRoZSBWTSBpcyBleGVjdXRpbmcg eGhyLm9wZW4oKSBhbmQgc3VjY2VlZHMgaW4gc2V0dGluZworLy8gdXAgdGhlIG5ldyByZXF1ZXN0 LiBJbiByZXNwb25zZSB0byBoYXZpbmcgb3BlbmVkIHRoZSByZXF1ZXN0LCBpdCB0cmllcyB0bwor Ly8gZGlzcGF0Y2ggYW4gZXZlbnQgd2l0aCByZWFkeVN0YXRlIDEgKE9QRU5FRCkuIFVuZm9ydHVu YXRlbHksIHRoZXJlIGlzIG5vdAorLy8gZW5vdWdoIHN0YWNrIGFuZCB3ZSBnZXQgYW4gb3ZlcmZs b3cgZXJyb3IgYW5kIDEgY29uc29sZSBlcnJvciBtZXNzYWdlIGhlcmUuCisvLworLy8gVGhlIGVy cm9yIGlzIHRocm93biBidXQgaXMgdGhlbiBlYXRlbiB1cCBieSB0aGUgRXZlbnREaXNwYXRjaGVy LiBIZW5jZSwKKy8vIG9ucmVhZHlzdGF0ZWNoYW5nZSgpIGJhY2sgaW4gSlMgbGFuZCBkb2VzIG5v dCBldmVuIHNlZSB0aGlzIGVycm9yLgorLy8KKy8vIG9wZW4oKSBldmVudHVhbGx5IHJldHVybnMg dG8gb25yZWFkeXN0YXRlY2hhbmdlKCksIGFuZCB3ZSBjYWxsIHhoci5zZW5kKCkKKy8vIG5leHQu IHhoci5zZW5kKCkgb3ZlcmZsb3dzIHRoZSBzdGFjayBhZ2FpbiwgYW5kIHdlIGdldCBhIDJuZCBl cnJvciBhbmQKKy8vIGEgMm5kIGNvbnNvbGUgZXJyb3IgbWVzc2FnZS4KKy8vCisvLyBOb3RlOiB0 aGUgbGFzdFJlYWR5U3RhdGVPYnNlcnZlZCBpbiB0aGlzIGNhc2UgaXMgNCBiZWNhdXNlIHRoZSBl dmVudAorLy8gZGlzcGF0Y2ggZm9yIE9QRU5FRCBmYWlsZWQgdG8gZXhlY3V0ZSBvbnJlYWR5c3Rh dGVjaGFuZ2UoKS4KKy8vCisvLyBUaGUgdG90YWwgbnVtYmVyIG9mIGVycm9yIG1lc3NhZ2VzIHNl ZW4gaXMgMi4KKy8vCisvLyBTY2VuYXJpbyAyOiBPdmVyZmxvd2VkIHdoaWxlIGNhbGxpbmcgeGhy LnNlbmQoKQorLy8gPT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT0KKy8vIEluIHRoaXMgc2NlbmFyaW8sIHRoZSBWTSBpcyBleGVjdXRpbmcgeGhyLnNlbmQoKSBh bmQgYSByZWFkeVN0YXRlIGNoYW5nZQorLy8gdG8gNCAoRE9ORSkgb2NjdXJzLiBBbiBldmVudCBp cyBkaXNwYXRjaGVkIGFuZCB0aGUgVk0gdHJpZXMgdG8gY2FsbAorLy8gb25yZWFkeXN0YXRlY2hh bmdlKCkgYnV0IG92ZXJmbG93cyB0aGUgc3RhY2suIEhlcmUsIHdlIHNlZSAxIGVycm9yIGFuZAor Ly8gMSBjb25zb2xlIGVycm9yIG1lc3NhZ2UuCisvLworLy8gU2luY2Ugd2UgbmV2ZXIgc3VjY2Vl ZGVkIGluIHJlZW50ZXJpbmcgb25yZWFkeXN0YXRlY2hhbmdlKCksIHdlIGRpZCBub3QKKy8vIG5v dCBzZXQgdXAgYW55bW9yZSByZXF1ZXN0cywgYW5kIHdpbGwgbm90IGF0dGVtcHQgdG8gcmUtZW50 ZXIKKy8vIG9ucmVhZHlzdGF0ZWNoYW5nZSgpIGFmdGVyIHRoaXMuIEhlbmNlLCB3ZSB3aWxsIG5v dCBzZWUgYW55bW9yZSBvdmVyZmxvdworLy8gZXJyb3JzIG5vciBjb25zb2xlIGVycm9yIG1lc3Nh Z2VzLgorLy8KKy8vIEluIGNvbnRyYXN0IHdpdGggc2NlbmFyaW8gMSwgd2UgZGlkIG5vdCBzZWUg YW4gb3ZlcmZsb3cgZXJyb3IgZm9yIHhoci5vcGVuKCkKKy8vIGJlZm9yZSB3ZSBzYXcgdGhlIGVy cm9yIGZvciB4aHIuc2VuZCgpLgorLy8KKy8vIE5vdGU6IGxhc3RSZWFkeVN0YXRlT2JzZXJ2ZWQg aW4gdGhpcyBjYXNlIGlzIDEgYmVjYXVzZSB0aGUgZXZlbnQKKy8vIGRpc3BhdGNoIGZvciBPUEVO RUQgc3VjY2VlZGVkIGluIGV4ZWN1dGluZyBvbnJlYWR5c3RhdGVjaGFuZ2UoKSwgYW5kCisvLyBs YXN0UmVhZHlTdGF0ZU9ic2VydmVkIHdhcyBzZXQgdG8gMS4KKy8vCisvLyBUaGUgdG90YWwgbnVt YmVyIG9mIGVycm9yIG1lc3NhZ2VzIHNlZW4gaXMgMS4KKy8vCisvLyBGbGFraW5lc3MKKy8vID09 PT09PT09PQorLy8gU2luY2Ugd2UgY2FuIGVuZCB1cCB3aXRoIHNjZW5hcmlvIDEgb3IgMiBkZXBl bmRpbmcgb24gdmFyaWFibGVzIG91dHNpZGUgdGhlCisvLyBjb250cm9sIG9mIHRoaXMgdGVzdCwg dGhlIHJlc3VsdCBvZiB0aGUgdGVzdCB3b3VsZCBiZSBpbmhlcmVudGx5IGZsYWt5IGkuZS4KKy8v IHdlIGNhbiBzZWUgMSBvciAyIGNvbnNvbGUgZXJyb3IgbWVzc2FnZXMuCisvLworLy8gVGhhdCBp cyB1bmxlc3Mgd2UgZG8gc29tZSBjb21wZW5zYXRpb24uIEluIHRoZSBjYXNlIG9mIHNjZW5hcmlv IDIsIHdlIGNhbgorLy8gY2hlY2sgaWYgdGhlIGxhc3RSZWFkeVN0YXRlT2JzZXJ2ZWQgaXMgMSB3 aGljaCBtZWFucyB3ZSB3b3VsZCBoYXZlIG9ubHkgc2VlbgorLy8gb25lIGNvbnNvbGUgZXJyb3Ig bWVzc2FnZS4gSWYgc28sIHdlIGNhbiBjYWxsIHhoci5zZW5kKCkgYWdhaW4ganVzdCB0bworLy8g dHJpZ2dlciBhbm90aGVyIG9uZSBzbyB0aGF0IHdlJ2xsIGhhdmUgYSB0b3RhbCBvZiAyLgorLy8K Ky8vIE5vdGUgdGhhdCBpbiB0aGUgY29tcGVuc2F0aW9uIGNhc2UsIHdlJ2xsIG5lZWQgdG8gY2Fs bCB4aHIub3BlbigpIGFnYWluCisvLyBiZWZvcmUgY2FsbGluZyB4aHIuc2VuZCgpLiBUaGlzIGlz IGJlY2F1c2UgdGhlIHByZXZpb3VzIHhociByZXF1ZXN0IGhhcworLy8gYWxyZWFkeSBiZWVuIGNs b3NlZC4gSGVyZSwgeGhyLm9wZW4oKSBzaG91bGQgbm90IHRyaWdnZXIgYW4gb3ZlcmZsb3cgZXJy b3IKKy8vIGp1c3QgbGlrZSBpdCBkaWRuJ3QgYmVmb3JlIHRoZSBwcmV2aW91cyB4aHIuc2VuZCgp IHRyaWdnZXJlZCB0aGUgZXJyb3IgKHRoZQorLy8gZmFjdCB0aGF0IGl0IHN1Y2NlZWRlZCB3aXRo b3V0IGFuIGVycm9yIGJlZm9yZSBtZWFucyBpdCB3aWxsIHN1Y2NlZWQgdGhpcworLy8gdGltZSB0 b28pLiBBbmQgdGhpcyBsZWF2ZXMgaXQgdXAgdG8gb3VyIGNvbXBlbnNhdGluZyBjYWxsIHRvIHho ci5zZW5kKCkKKy8vIHRvIGdlbmVyYXRlIHRoZSAybmQgZXhwZWN0ZWQgb3ZlcmZsb3cgZXJyb3Ig YW5kIGNvbnNvbGUgZXJyb3IgbWVzc2FnZS4KKy8vCisvLyBOb3RlIGFsc28gdGhhdCB3ZSdsbCBu ZWVkIHRvIG1ha2Ugc3VyZSB0aGF0IHdlIGRvIHRoaXMgY29tcGVuc2F0aW9uIG9ubHkKKy8vIG9u Y2UsIGFuZCBub3QgcmVwZWF0ZWRseSBkbyB0aGUgY29tcGVuc2F0aW5nIHhoci5zZW5kKCkgY2Fs bCBvbiBldmVyeQorLy8gb25yZWFkeXN0YXRlY2hhbmdlKCkgZnJhbWUgb24gdGhlIHJldHVybiBw YXRoLgorCitsYXN0UmVhZHlTdGF0ZU9ic2VydmVkID0gMDsKK2hhc0NvbXBlbnNhdGVkQWxyZWFk eSA9IGZhbHNlOworCiBmdW5jdGlvbiB0ZXN0KCkKIHsKICAgICBpZiAod2luZG93LnRlc3RSdW5u ZXIpIHsKQEAgLTE0LDkgKzg5LDE3IEBAIGZ1bmN0aW9uIHRlc3QoKQogICAgIH0KICAgICB2YXIg eGhyID0gbmV3IFhNTEh0dHBSZXF1ZXN0KCk7CiAgICAgeGhyLm9ucmVhZHlzdGF0ZWNoYW5nZSA9 IGZ1bmN0aW9uKCkgeworICAgICAgICBsYXN0UmVhZHlTdGF0ZU9ic2VydmVkID0geGhyLnJlYWR5 U3RhdGU7CiAgICAgICAgIGlmICh4aHIucmVhZHlTdGF0ZSA9PSA0KSB7CiAgICAgICAgICAgICB4 aHIub3BlbigiR0VUIiwgInJlY3Vyc2UuaHRtbCIsIGZhbHNlKTsKICAgICAgICAgICAgIHhoci5z ZW5kKG51bGwpOworCisgICAgICAgICAgICAvLyBDb21wZW5zYXRlIGZvciB0ZXN0IGZsYWtpbmVz cyBpZiBuZWVkZWQ6CisgICAgICAgICAgICBpZiAoIWhhc0NvbXBlbnNhdGVkQWxyZWFkeSAmJiBs YXN0UmVhZHlTdGF0ZU9ic2VydmVkID09IDEpIHsKKyAgICAgICAgICAgICAgICB4aHIub3Blbigi R0VUIiwgInJlY3Vyc2UuaHRtbCIsIGZhbHNlKTsKKyAgICAgICAgICAgICAgICB4aHIuc2VuZChu dWxsKTsKKyAgICAgICAgICAgICAgICBoYXNDb21wZW5zYXRlZEFscmVhZHkgPSB0cnVlOworICAg ICAgICAgICAgfQogICAgICAgICB9CiAgICAgfTsKICAgICB4aHIub3BlbigiR0VUIiwgInJlY3Vy c2UuaHRtbCIsIGZhbHNlKTsK