86813 2012-05-17 23:16:46 -0700 WebKit spends ~20% of time in HTMLTextAreaElement::defaultValue() when opening a review page 2013-08-09 16:52:47 -0700 1 1 1 Unclassified WebKit Forms 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED P2 Normal --- 86994 1 rniwa rniwa abarth adele aestes ap cmarcelo darin dbates eric fpizlo kling koivisto mifenton ojan rafael.lobo tkent tonikitoo webkit.review.bot oldest_to_newest 627376 0 rniwa 2012-05-17 23:16:46 -0700 Reproduction steps: 1. Open https://bugs.webkit.org/attachment.cgi?id=142634 (Caution: 5MB!) 2. Instrument the Web Process while WebKit opens it Observe that we spend 19.5% of time in HTMLTextAreaElement::defaultValue() inside HTMLTextAreaElement::childrenChanged, which in turn is called by ContainerNode::parserAddChild (18.5%). 627379 1 rniwa 2012-05-17 23:19:35 -0700 Obviously, we should be using StringBuilder in this function. But even then we still have O(n^2) algorithm here because parserAddChild is going to call this function whenever new child is added :( 373 String HTMLTextAreaElement::defaultValue() const 374 { 375 String value = ""; 376 377 // Since there may be comments, ignore nodes other than text nodes. 378 for (Node* n = firstChild(); n; n = n->nextSibling()) { 379 if (n->isTextNode()) 380 value += toText(n)->data(); 381 } 382 383 return value; 384 } Perhaps we can lazily update the default value when the parser is adding children? 627386 2 ojan 2012-05-17 23:28:54 -0700 (In reply to comment #0) > Reproduction steps: > 1. Open https://bugs.webkit.org/attachment.cgi?id=142634 (Caution: 5MB!) > 2. Instrument the Web Process while WebKit opens it > > Observe that we spend 19.5% of time in HTMLTextAreaElement::defaultValue() inside HTMLTextAreaElement::childrenChanged, which in turn is called by ContainerNode::parserAddChild (18.5%). Did you link to the wrong page? There's no textareas on that page, just plain text. 627391 3 rniwa 2012-05-17 23:32:37 -0700 (In reply to comment #2) > Did you link to the wrong page? There's no textareas on that page, just plain text. Oh sorry, I meant https://bugs.webkit.org/attachment.cgi?id=142634&action=review. 627398 4 eric 2012-05-17 23:43:50 -0700 We should be able to tell that we're still during a parse and do it during the finishedParsingChildren method, or whatever it's named these days. the "we just parsed the end tag" method. 627400 5 rniwa 2012-05-17 23:45:44 -0700 (In reply to comment #4) > We should be able to tell that we're still during a parse and do it during the finishedParsingChildren method, or whatever it's named these days. the "we just parsed the end tag" method. Can scripts run while parsing textarea content? 627402 6 fpizlo 2012-05-17 23:47:50 -0700 (In reply to comment #5) > (In reply to comment #4) > > We should be able to tell that we're still during a parse and do it during the finishedParsingChildren method, or whatever it's named these days. the "we just parsed the end tag" method. > > Can scripts run while parsing textarea content? Even if they can't currently, it feels safer to assume that they could. Imagine the horror if someone made it so that scripts could run while parsing textarea content and we had put in an optimization that relied on them not running. 627403 7 eric 2012-05-17 23:48:10 -0700 We could be in the middle of a <textArea> when the parser pauses and scripts run (from a load or whatever), sure. 627410 8 rniwa 2012-05-17 23:54:16 -0700 (In reply to comment #7) > We could be in the middle of a <textArea> when the parser pauses and scripts run (from a load or whatever), sure. I see. That makes things a little tricker I guess because some Web-facing API relies on the fact we update values whenever children changes. So there's some book keeping to do. 627411 9 eric 2012-05-17 23:56:15 -0700 Lots of web-facing apis rely on the fact that we layout/resolve-style before we return them values. :) Similarly we would need to keep a dirty bit here and know to refresh our cache whenever those are called. 627412 10 ojan 2012-05-17 23:56:56 -0700 I can't even load the page on my laptop. :( If I could, I'd try to understand what textareas are so slow. Only think I can think of is that we load the old review page in an iframe that contains the full diff as text in the textarea. I could imagine that being slow. It's silly through really. We don't need to load the diff in that iframe at all anymore. Noone uses the old review page anymore. We should fix the inefficiency on the C++ side first of course, but after that, we should just change the review page to not be dumb. 629209 11 koivisto 2012-05-21 09:51:12 -0700 Can anyone think of a valid use for String::operator+=()? If not we should just remove it and make all current clients use StringBuilder instead. 629295 12 rniwa 2012-05-21 11:26:37 -0700 It appears that just using StringBuilder instead of operator+= will fix this perf. hit. Now we're spending roughly 8% of time in treeScope() :( Someone needs to make treeScope O(1). It's showing up every time I profile some DOM operation. 629306 13 rniwa 2012-05-21 11:33:23 -0700 Filed https://bugs.webkit.org/show_bug.cgi?id=87034 to track treeScope() slowness per Dimitri's request. 629338 14 rniwa 2012-05-21 11:55:27 -0700 We could also lazy construct string at finishParsingChildren (paired with beginParsingChildren) but maybe we can do something more general like not replacing all children in the shadow DOM when some nodes match. 629354 15 rniwa 2012-05-21 12:05:47 -0700 (In reply to comment #14) > We could also lazy construct string at finishParsingChildren (paired with beginParsingChildren) but maybe we can do something more general like not replacing all children in the shadow DOM when some nodes match. Turns out that this requires a considerable amount of refactoring. Let's just replace String by StringBuilder for now. 629497 16 143097 rniwa 2012-05-21 14:30:20 -0700 Created attachment 143097 Fixes the bug 629516 17 143097 eric 2012-05-21 14:38:28 -0700 Comment on attachment 143097 Fixes the bug View in context: https://bugs.webkit.org/attachment.cgi?id=143097&action=review LGTM. > Source/WebCore/ChangeLog:13 > + Also avoid the redundant call to setValue in childrenChaned when the value is dirty. childrenChaned! 629527 18 143101 rniwa 2012-05-21 14:47:49 -0700 Created attachment 143101 Patch for landing 629529 19 rniwa 2012-05-21 14:50:31 -0700 I'm going to hold off on landing this patch until the bug 86994 is fixed (hopefully later today). 631389 20 143101 webkit.review.bot 2012-05-23 01:49:26 -0700 Comment on attachment 143101 Patch for landing Clearing flags on attachment: 143101 Committed r118149: <http://trac.webkit.org/changeset/118149> 631390 21 webkit.review.bot 2012-05-23 01:49:33 -0700 All reviewed patches have been landed. Closing bug. 632350 22 rniwa 2012-05-23 19:58:43 -0700 This patch improved the score of tests added by the bug 86994 by 1000-3000%: http://webkit-perf.appspot.com/graph.html#tests=[[2831883,2001,3001],[2831883,2001,173262],[2831883,2001,963028],[2831883,2001,957069],[2831883,2001,2389050],[2965621,2001,32196],[2965621,2001,957069],[2965621,2001,173262],[2965621,2001,2389050],[2965621,2001,3001],[2965621,2001,963028],[2831883,2001,32196]]&sel=1337223431149,1337828231149&displayrange=7&datatype=running 632386 23 tkent 2012-05-23 20:53:04 -0700 (In reply to comment #22) > This patch improved the score of tests added by the bug 86994 by 1000-3000%: > http://webkit-perf.appspot.com/graph.html#tests=[[2831883,2001,3001],[2831883,2001,173262],[2831883,2001,963028],[2831883,2001,957069],[2831883,2001,2389050],[2965621,2001,32196],[2965621,2001,957069],[2965621,2001,173262],[2965621,2001,2389050],[2965621,2001,3001],[2965621,2001,963028],[2831883,2001,32196]]&sel=1337223431149,1337828231149&displayrange=7&datatype=running Cool! textarea-dom.html: Huge improvement textarea-parsing.html: 4% improvement textarea-edit.html: Not affected. 632407 24 rniwa 2012-05-23 21:39:16 -0700 (In reply to comment #23) > Cool! > textarea-dom.html: Huge improvement > textarea-parsing.html: 4% improvement > textarea-edit.html: Not affected. Do you know where we're spending time in the latter two tests? 636987 25 tkent 2012-05-30 00:57:46 -0700 (In reply to comment #24) > (In reply to comment #23) > > Cool! > > textarea-dom.html: Huge improvement > > textarea-parsing.html: 4% improvement > > textarea-edit.html: Not affected. > > Do you know where we're spending time in the latter two tests? I haven't. Try replacing the textarea in textarea-edit.html with a <div contentEditable=true>. It's slightly fastter than the <textarea>, but it's still very slow. 143097 2012-05-21 14:30:20 -0700 2012-05-21 14:47:41 -0700 Fixes the bug bug-86813-20120521143356.patch text/plain 2780 rniwa SW5kZXg6IFNvdXJjZS9XZWJDb3JlL0NoYW5nZUxvZwo9PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBTb3VyY2UvV2Vi Q29yZS9DaGFuZ2VMb2cJKHJldmlzaW9uIDExNzgxNSkKKysrIFNvdXJjZS9XZWJDb3JlL0NoYW5n ZUxvZwkod29ya2luZyBjb3B5KQpAQCAtMSwzICsxLDIyIEBACisyMDEyLTA1LTIxICBSeW9zdWtl IE5pd2EgIDxybml3YUB3ZWJraXQub3JnPgorCisgICAgICAgIFdlYktpdCBzcGVuZHMgfjIwJSBv ZiB0aW1lIGluIEhUTUxUZXh0QXJlYUVsZW1lbnQ6OmRlZmF1bHRWYWx1ZSgpIHdoZW4gb3Blbmlu ZyBhIHJldmlldyBwYWdlCisgICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVn LmNnaT9pZD04NjgxMworCisgICAgICAgIFJldmlld2VkIGJ5IE5PQk9EWSAoT09QUyEpLgorCisg ICAgICAgIFJlcGxhY2UgU3RyaW5nIGFuZCBjYWxscyB0byBTdHJpbmc6Om9wZXJhdG9yKz0gYnkg U3RyaW5nQnVpbGRlciBhbmQgY2FsbHMgdG8gYXBwZW5kIHJlc3BlY3RpdmVseS4KKyAgICAgICAg V2hpbGUgdGhlcmUgYXJlIG90aGVyIG9wdGltaXphdGlvbnMgd2UgY2FuIGltcGxlbWVudCBzdWNo IGFzIG5vdCByZXBsYWNpbmcgYWxsIG5vZGVzIGluIHRoZSBzaGFkb3cgRE9NLAorICAgICAgICBv ciBub3Qgc2VyaWFsaXppbmcgdGhlIHZhbHVlIHdoZW4gZGV0ZWN0aW5nIHdoZXRoZXIgdGhlIHZh bHVlIGhhZCBjaGFuZ2VkLCB0aGV5IHJlcXVpcmUgY29uc2lkZXJhYmxlCisgICAgICAgIGFtb3Vu dCBvZiByZWZhY3RvcmluZyBhbmQgdGhpcyBjaGFuZ2UgYWxvbmUgc2VlbXMgdG8gYnV5IHVzIGEg bG90IHNvIGxldCB1cyBkZWZlciB0aG9zZSAndGlsIGxhdGVyIHRpbWUuCisKKyAgICAgICAgQWxz byBhdm9pZCB0aGUgcmVkdW5kYW50IGNhbGwgdG8gc2V0VmFsdWUgaW4gY2hpbGRyZW5DaGFuZWQg d2hlbiB0aGUgdmFsdWUgaXMgZGlydHkuCisgICAgICAgIFRoaXMgc2hvdWxkIGhhbHZlIHRoZSB0 aW1lIHdlIHNwZW50IGluIHNldHRpbmcgdGhlIHZhbHVlLgorCisgICAgICAgICogaHRtbC9IVE1M VGV4dEFyZWFFbGVtZW50LmNwcDoKKyAgICAgICAgKFdlYkNvcmU6OkhUTUxUZXh0QXJlYUVsZW1l bnQ6OmNoaWxkcmVuQ2hhbmdlZCk6CisgICAgICAgIChXZWJDb3JlOjpIVE1MVGV4dEFyZWFFbGVt ZW50OjpkZWZhdWx0VmFsdWUpOgorCiAyMDEyLTA1LTIxICBaaGVueWFvIE1vICA8em1vQGdvb2ds ZS5jb20+CiAKICAgICAgICAgUmVzdG9yZSBXZWJHTCdzIGZyYW1lYnVmZmVyIGJpbmRpbmcgYWZ0 ZXIgRHJhd2luZ0J1ZmZlcidzIHByZXBhcmVCYWNrQnVmZmVyKCkKSW5kZXg6IFNvdXJjZS9XZWJD b3JlL2h0bWwvSFRNTFRleHRBcmVhRWxlbWVudC5jcHAKPT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291cmNlL1dl YkNvcmUvaHRtbC9IVE1MVGV4dEFyZWFFbGVtZW50LmNwcAkocmV2aXNpb24gMTE3Nzk4KQorKysg U291cmNlL1dlYkNvcmUvaHRtbC9IVE1MVGV4dEFyZWFFbGVtZW50LmNwcAkod29ya2luZyBjb3B5 KQpAQCAtNDQsNiArNDQsNyBAQAogI2luY2x1ZGUgIlRleHRDb250cm9sSW5uZXJFbGVtZW50cy5o IgogI2luY2x1ZGUgIlRleHRJdGVyYXRvci5oIgogI2luY2x1ZGUgPHd0Zi9TdGRMaWJFeHRyYXMu aD4KKyNpbmNsdWRlIDx3dGYvdGV4dC9TdHJpbmdCdWlsZGVyLmg+CiAKIG5hbWVzcGFjZSBXZWJD b3JlIHsKIApAQCAtMTE1LDkgKzExNiwxMCBAQCB2b2lkIEhUTUxUZXh0QXJlYUVsZW1lbnQ6OmNo aWxkcmVuQ2hhbmdlCiB7CiAgICAgSFRNTEVsZW1lbnQ6OmNoaWxkcmVuQ2hhbmdlZChjaGFuZ2Vk QnlQYXJzZXIsIGJlZm9yZUNoYW5nZSwgYWZ0ZXJDaGFuZ2UsIGNoaWxkQ291bnREZWx0YSk7CiAg ICAgc2V0TGFzdENoYW5nZVdhc05vdFVzZXJFZGl0KCk7Ci0gICAgaWYgKCFtX2lzRGlydHkpCisg ICAgaWYgKG1faXNEaXJ0eSkKKyAgICAgICAgc2V0SW5uZXJUZXh0VmFsdWUodmFsdWUoKSk7Cisg ICAgZWxzZQogICAgICAgICBzZXROb25EaXJ0eVZhbHVlKGRlZmF1bHRWYWx1ZSgpKTsKLSAgICBz ZXRJbm5lclRleHRWYWx1ZSh2YWx1ZSgpKTsKIH0KIAogYm9vbCBIVE1MVGV4dEFyZWFFbGVtZW50 Ojppc1ByZXNlbnRhdGlvbkF0dHJpYnV0ZShjb25zdCBRdWFsaWZpZWROYW1lJiBuYW1lKSBjb25z dApAQCAtMzcyLDE1ICszNzQsMTUgQEAgdm9pZCBIVE1MVGV4dEFyZWFFbGVtZW50OjpzZXRWYWx1 ZUNvbW1vbgogCiBTdHJpbmcgSFRNTFRleHRBcmVhRWxlbWVudDo6ZGVmYXVsdFZhbHVlKCkgY29u c3QKIHsKLSAgICBTdHJpbmcgdmFsdWUgPSAiIjsKKyAgICBTdHJpbmdCdWlsZGVyIHZhbHVlOwog CiAgICAgLy8gU2luY2UgdGhlcmUgbWF5IGJlIGNvbW1lbnRzLCBpZ25vcmUgbm9kZXMgb3RoZXIg dGhhbiB0ZXh0IG5vZGVzLgogICAgIGZvciAoTm9kZSogbiA9IGZpcnN0Q2hpbGQoKTsgbjsgbiA9 IG4tPm5leHRTaWJsaW5nKCkpIHsKICAgICAgICAgaWYgKG4tPmlzVGV4dE5vZGUoKSkKLSAgICAg ICAgICAgIHZhbHVlICs9IHRvVGV4dChuKS0+ZGF0YSgpOworICAgICAgICAgICAgdmFsdWUuYXBw ZW5kKHRvVGV4dChuKS0+ZGF0YSgpKTsKICAgICB9CiAKLSAgICByZXR1cm4gdmFsdWU7CisgICAg cmV0dXJuIHZhbHVlLnRvU3RyaW5nKCk7CiB9CiAKIHZvaWQgSFRNTFRleHRBcmVhRWxlbWVudDo6 c2V0RGVmYXVsdFZhbHVlKGNvbnN0IFN0cmluZyYgZGVmYXVsdFZhbHVlKQo= 143101 2012-05-21 14:47:49 -0700 2012-05-23 01:49:26 -0700 Patch for landing bug-86813-20120521145125.patch text/plain 2778 rniwa SW5kZXg6IFNvdXJjZS9XZWJDb3JlL0NoYW5nZUxvZwo9PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBTb3VyY2UvV2Vi Q29yZS9DaGFuZ2VMb2cJKHJldmlzaW9uIDExNzgxNSkKKysrIFNvdXJjZS9XZWJDb3JlL0NoYW5n ZUxvZwkod29ya2luZyBjb3B5KQpAQCAtMSwzICsxLDIyIEBACisyMDEyLTA1LTIxICBSeW9zdWtl IE5pd2EgIDxybml3YUB3ZWJraXQub3JnPgorCisgICAgICAgIFdlYktpdCBzcGVuZHMgfjIwJSBv ZiB0aW1lIGluIEhUTUxUZXh0QXJlYUVsZW1lbnQ6OmRlZmF1bHRWYWx1ZSgpIHdoZW4gb3Blbmlu ZyBhIHJldmlldyBwYWdlCisgICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVn LmNnaT9pZD04NjgxMworCisgICAgICAgIFJldmlld2VkIGJ5IEVyaWMgU2VpZGVsLgorCisgICAg ICAgIFJlcGxhY2UgU3RyaW5nIGFuZCBjYWxscyB0byBTdHJpbmc6Om9wZXJhdG9yKz0gYnkgU3Ry aW5nQnVpbGRlciBhbmQgY2FsbHMgdG8gYXBwZW5kIHJlc3BlY3RpdmVseS4KKyAgICAgICAgV2hp bGUgdGhlcmUgYXJlIG90aGVyIG9wdGltaXphdGlvbnMgd2UgY2FuIGltcGxlbWVudCBzdWNoIGFz IG5vdCByZXBsYWNpbmcgYWxsIG5vZGVzIGluIHRoZSBzaGFkb3cgRE9NLAorICAgICAgICBvciBu b3Qgc2VyaWFsaXppbmcgdGhlIHZhbHVlIHdoZW4gZGV0ZWN0aW5nIHdoZXRoZXIgdGhlIHZhbHVl IGhhZCBjaGFuZ2VkLCB0aGV5IHJlcXVpcmUgY29uc2lkZXJhYmxlCisgICAgICAgIGFtb3VudCBv ZiByZWZhY3RvcmluZyBhbmQgdGhpcyBjaGFuZ2UgYWxvbmUgc2VlbXMgdG8gYnV5IHVzIGEgbG90 IHNvIGxldCB1cyBkZWZlciB0aG9zZSAndGlsIGxhdGVyIHRpbWUuCisKKyAgICAgICAgQWxzbyBh dm9pZCB0aGUgcmVkdW5kYW50IGNhbGwgdG8gc2V0VmFsdWUgaW4gY2hpbGRyZW5DaGFuZ2VkIHdo ZW4gdGhlIHZhbHVlIGlzIGRpcnR5LgorICAgICAgICBUaGlzIHNob3VsZCBoYWx2ZSB0aGUgdGlt ZSB3ZSBzcGVuZCBpbiBzZXR0aW5nIHRoZSB2YWx1ZS4KKworICAgICAgICAqIGh0bWwvSFRNTFRl eHRBcmVhRWxlbWVudC5jcHA6CisgICAgICAgIChXZWJDb3JlOjpIVE1MVGV4dEFyZWFFbGVtZW50 OjpjaGlsZHJlbkNoYW5nZWQpOgorICAgICAgICAoV2ViQ29yZTo6SFRNTFRleHRBcmVhRWxlbWVu dDo6ZGVmYXVsdFZhbHVlKToKKwogMjAxMi0wNS0yMSAgWmhlbnlhbyBNbyAgPHptb0Bnb29nbGUu Y29tPgogCiAgICAgICAgIFJlc3RvcmUgV2ViR0wncyBmcmFtZWJ1ZmZlciBiaW5kaW5nIGFmdGVy IERyYXdpbmdCdWZmZXIncyBwcmVwYXJlQmFja0J1ZmZlcigpCkluZGV4OiBTb3VyY2UvV2ViQ29y ZS9odG1sL0hUTUxUZXh0QXJlYUVsZW1lbnQuY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFNvdXJjZS9XZWJD b3JlL2h0bWwvSFRNTFRleHRBcmVhRWxlbWVudC5jcHAJKHJldmlzaW9uIDExNzc5OCkKKysrIFNv dXJjZS9XZWJDb3JlL2h0bWwvSFRNTFRleHRBcmVhRWxlbWVudC5jcHAJKHdvcmtpbmcgY29weSkK QEAgLTQ0LDYgKzQ0LDcgQEAKICNpbmNsdWRlICJUZXh0Q29udHJvbElubmVyRWxlbWVudHMuaCIK ICNpbmNsdWRlICJUZXh0SXRlcmF0b3IuaCIKICNpbmNsdWRlIDx3dGYvU3RkTGliRXh0cmFzLmg+ CisjaW5jbHVkZSA8d3RmL3RleHQvU3RyaW5nQnVpbGRlci5oPgogCiBuYW1lc3BhY2UgV2ViQ29y ZSB7CiAKQEAgLTExNSw5ICsxMTYsMTAgQEAgdm9pZCBIVE1MVGV4dEFyZWFFbGVtZW50OjpjaGls ZHJlbkNoYW5nZQogewogICAgIEhUTUxFbGVtZW50OjpjaGlsZHJlbkNoYW5nZWQoY2hhbmdlZEJ5 UGFyc2VyLCBiZWZvcmVDaGFuZ2UsIGFmdGVyQ2hhbmdlLCBjaGlsZENvdW50RGVsdGEpOwogICAg IHNldExhc3RDaGFuZ2VXYXNOb3RVc2VyRWRpdCgpOwotICAgIGlmICghbV9pc0RpcnR5KQorICAg IGlmIChtX2lzRGlydHkpCisgICAgICAgIHNldElubmVyVGV4dFZhbHVlKHZhbHVlKCkpOworICAg IGVsc2UKICAgICAgICAgc2V0Tm9uRGlydHlWYWx1ZShkZWZhdWx0VmFsdWUoKSk7Ci0gICAgc2V0 SW5uZXJUZXh0VmFsdWUodmFsdWUoKSk7CiB9CiAKIGJvb2wgSFRNTFRleHRBcmVhRWxlbWVudDo6 aXNQcmVzZW50YXRpb25BdHRyaWJ1dGUoY29uc3QgUXVhbGlmaWVkTmFtZSYgbmFtZSkgY29uc3QK QEAgLTM3MiwxNSArMzc0LDE1IEBAIHZvaWQgSFRNTFRleHRBcmVhRWxlbWVudDo6c2V0VmFsdWVD b21tb24KIAogU3RyaW5nIEhUTUxUZXh0QXJlYUVsZW1lbnQ6OmRlZmF1bHRWYWx1ZSgpIGNvbnN0 CiB7Ci0gICAgU3RyaW5nIHZhbHVlID0gIiI7CisgICAgU3RyaW5nQnVpbGRlciB2YWx1ZTsKIAog ICAgIC8vIFNpbmNlIHRoZXJlIG1heSBiZSBjb21tZW50cywgaWdub3JlIG5vZGVzIG90aGVyIHRo YW4gdGV4dCBub2Rlcy4KICAgICBmb3IgKE5vZGUqIG4gPSBmaXJzdENoaWxkKCk7IG47IG4gPSBu LT5uZXh0U2libGluZygpKSB7CiAgICAgICAgIGlmIChuLT5pc1RleHROb2RlKCkpCi0gICAgICAg ICAgICB2YWx1ZSArPSB0b1RleHQobiktPmRhdGEoKTsKKyAgICAgICAgICAgIHZhbHVlLmFwcGVu ZCh0b1RleHQobiktPmRhdGEoKSk7CiAgICAgfQogCi0gICAgcmV0dXJuIHZhbHVlOworICAgIHJl dHVybiB2YWx1ZS50b1N0cmluZygpOwogfQogCiB2b2lkIEhUTUxUZXh0QXJlYUVsZW1lbnQ6OnNl dERlZmF1bHRWYWx1ZShjb25zdCBTdHJpbmcmIGRlZmF1bHRWYWx1ZSkK