94761 2012-08-22 17:31:19 -0700 ThreadRestrictionVerifier should be opt-in, not opt-out 2012-08-24 15:56:04 -0700 1 1 1 Unclassified WebKit New Bugs 528+ (Nightly build) Unspecified Unspecified RESOLVED FIXED P2 Normal --- 1 ggaren ggaren ap benjamin levin levin+threading webkit.review.bot oldest_to_newest 702549 0 ggaren 2012-08-22 17:31:19 -0700 ThreadRestrictionVerifier should be opt-in, not opt-out 702561 1 160047 ggaren 2012-08-22 17:38:52 -0700 Created attachment 160047 Patch 702570 2 160047 mhahnenberg 2012-08-22 17:47:04 -0700 Comment on attachment 160047 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=160047&action=review r=me! > Source/ThirdParty/ANGLE/ChangeLog:8 > + Additional information of the change such as approach, rationale. Please add per-function descriptions below (OOPS!). Where did this ChangeLog come from? 702573 3 160047 levin 2012-08-22 17:51:20 -0700 Comment on attachment 160047 Patch View in context: https://bugs.webkit.org/attachment.cgi?id=160047&action=review This change appears to make it easier for JSC code to turn this off which is good. How does the rest of the code continue to get the benefit from this automatically as it has in the past? (There have been several issues uncovered by this which would have resulted in hard to figure out memory corruptions.) > Source/WTF/ChangeLog:23 > + to maintain with default off, so I removed them. Actually they aren't. They simply assert that the method calls were only made when the class was in its initial state. It is only the change to make m_mode NoVerificationMode that makes them fire. 702583 4 levin 2012-08-22 18:03:32 -0700 It sounds like the concern is that jsc is the primary place where threading is done so this is the place bitten by this assert most often. Honestly, I'm not very connected lately with WebKit so if folks actively working on the project feel this is best. Please go ahead! Thanks for taking the time to response to me in irc. 702593 5 160047 mhahnenberg 2012-08-22 18:11:29 -0700 Comment on attachment 160047 Patch It looks like David is ok with this, so r=me again! 702612 6 ggaren 2012-08-22 18:32:53 -0700 Committed r126379: <http://trac.webkit.org/changeset/126379> 703088 7 ap 2012-08-23 09:51:33 -0700 I strongly object to this change. ThreadRestrictionVerifier has been finding bugs in places we would never think of oping in for verification. JavaScriptCore is certainly not the only part of WebKit that does threading. 703243 8 ggaren 2012-08-23 12:17:54 -0700 > I strongly object to this change. OK, I hear your objection. Do you have a proposal for how to make our tests stop ASSERTing for any client that uses threads? 703363 9 ap 2012-08-23 14:21:33 -0700 Do we know when ASSERTing started, and what specific parts of JSC cause it? DumpRenderTree runs code at startup that uses threads, so one can definitely use JSC API from a thread - otherwise, no regression tests would run. 703530 10 ggaren 2012-08-23 16:39:33 -0700 > Do we know when ASSERTing started, and what specific parts of JSC cause it? This mechanism has caused ASSERTs to fire in JSC since its first introduction. It was designed with the assumption of a single-threaded VM (v8). The ASSERTs fire for anything RefPtr in Source/WTF or Source/JavaScriptCore. > DumpRenderTree runs code at startup that uses threads, so one can definitely use JSC API from a thread - otherwise, no regression tests would run. DumpRenderTree doesn't use the feature of using an object on multiple threads. 703537 11 levin 2012-08-23 16:46:55 -0700 (In reply to comment #10) > > Do we know when ASSERTing started, and what specific parts of JSC cause it? > > This mechanism has caused ASSERTs to fire in JSC since its first introduction. It was designed with the assumption of a single-threaded VM (v8). fwiw, I didn't design it for single-thread vm. v8 doesn't even use wtf as far as I know. I did it for WebCore (workers, icon db, index db, which use multiple threads, etc), etc. I tried to quiet JSC things that I hit. (I had intended to add something JSC specific that could check for locks rather than just turning it off but it still would need to go in each ref counted class in JSC -- I just never got around to it before I mostly disappeared). > > The ASSERTs fire for anything RefPtr in Source/WTF or Source/JavaScriptCore. > > > DumpRenderTree runs code at startup that uses threads, so one can definitely use JSC API from a thread - otherwise, no regression tests would run. > > DumpRenderTree doesn't use the feature of using an object on multiple threads. 703580 12 ggaren 2012-08-23 17:23:07 -0700 > > This mechanism has caused ASSERTs to fire in JSC since its first introduction. It was designed with the assumption of a single-threaded VM (v8). > > fwiw, I didn't design it for single-thread vm. v8 doesn't even use wtf as far as I know. OK, it was designed with the assumption that WebCore is the only client of WTF (which is only true in Chromium). 703721 13 levin 2012-08-23 21:31:37 -0700 (In reply to comment #12) > > > This mechanism has caused ASSERTs to fire in JSC since its first introduction. It was designed with the assumption of a single-threaded VM (v8). > > > > fwiw, I didn't design it for single-thread vm. v8 doesn't even use wtf as far as I know. > > OK, it was designed with the assumption that WebCore is the only client of WTF (which is only true in Chromium). I appreciate that you're trying to be generous, and I'm sorry that it didn't accomodate JSC better. Rather than wax poetically about my intentions, here's the data I see: JavaScriptCore, 12 classes which use RefCounted as a base class. http://code.google.com/p/chromium/source/search?q=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit%2FSource%2FJavaScriptCore&origq=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit%2FSource%2FJavaScriptCore&btnG=Search+Trunk WTF, 5 classes which use RefCounted as a base classes. http://code.google.com/p/chromium/source/search?q=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit%2FSource%2FWTF&origq=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit%2FSource%2FWTF&btnG=Search+Trunk Other directories, ~500 classes which use RefCounted as a base class. http://code.google.com/p/chromium/source/search?q=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit&origq=c%5C+RefCounted%5C%3C+file%3Athird_party%2FWebKit&btnG=Search+Trunk The default was set for those 500 classes in WebKit (not the other 17). The thread checker class had escape hatches for the special cases (which I put in places as I encountered them and I tried JSC in DRT and my debug WebKit build in Safari -- though any of my WebKit builds always hung due to some plugin thing in the Safari side of the code). As to this patch, folks working on WebKit should decide what is the best code to run in WebKit (that isn't me at present). 703745 14 ggaren 2012-08-23 22:15:17 -0700 > JavaScriptCore, 12 classes which use RefCounted as a base class. > WTF, 5 classes which use RefCounted as a base classes. OK, does anybody have a proposal for what to do with these 17 classes? 703747 15 levin 2012-08-23 22:25:28 -0700 (In reply to comment #14) > > JavaScriptCore, 12 classes which use RefCounted as a base class. > > WTF, 5 classes which use RefCounted as a base classes. > > OK, does anybody have a proposal for what to do with these 17 classes? Call "turnOffVerifier();" in their constructors? (Or just leave in this patch.) I didn't do this by default in the past because my approach with this was only to add that when I hit an assert (or someone else hit one, etc.) because I felt like this was the cautious path (verifying that each class I added it to was indeed an exception). However, I now realize that was over cautious for these classes. Ideally this wouldn't be in the 5 wtf classes unless they were used by JSC. However, if they are used in JSC, turning off the checking for them seems like a fine compromise. 704286 16 ggaren 2012-08-24 11:41:29 -0700 > Call "turnOffVerifier();" in their constructors? I would probably use a super-type of RefCounted (UnverifiedRefCounted? FreedomRefCounted?), to make the find-and-replace simpler, and leave a clearer message to future coders about the design rule. But... > Ideally this wouldn't be in the 5 wtf classes unless they were used by JSC. JSC uses fundamental WTF types, like StringImpl. So, a solution to turn things on by default in WebCore will still leave the verifier off for many of the classes in which it has caught bugs before. My understanding is that those WTF classes gave us most of the value before, which is why I didn't see much difference between a targeted solution and just turning it all off: either way, to get the value out of the verifier, you have to pick specific strings, etc., in WebCore and opt them into verification. Alexey, David, do you like the solution of "off by default in WTF and JSC types, on by default in WebCore types"? 704376 17 levin 2012-08-24 13:16:02 -0700 (In reply to comment #16) > > Call "turnOffVerifier();" in their constructors? > > I would probably use a super-type of RefCounted (UnverifiedRefCounted? FreedomRefCounted?), to make the find-and-replace simpler, and leave a clearer message to future coders about the design rule. > > But... > > > Ideally this wouldn't be in the 5 wtf classes unless they were used by JSC. > > JSC uses fundamental WTF types, like StringImpl. So, a solution to turn things on by default in WebCore will still leave the verifier off for many of the classes in which it has caught bugs before. > > My understanding is that those WTF classes gave us most of the value before, which is why I didn't see much difference between a targeted solution and just turning it all off: either way, to get the value out of the verifier, you have to pick specific strings, etc., in WebCore and opt them into verification. > > Alexey, David, do you like the solution of "off by default in WTF and JSC types, on by default in WebCore types"? fwiw, imo, that is a fine compromise. I don't remember any instances in which those found issues anyway. More context: The WebCore types (and higher) are actually the places where I have seen usefulness. fwiw, StringImpl doesn't use RefCounted so unfortunately strings never get this check. There are likely issue with strings. Do I need to say that it was on mental list to get to it (but then I did a DB Cooper -- w/o the extortion and ransom)? :) 704479 18 ap 2012-08-24 15:56:04 -0700 > Alexey, David, do you like the solution of "off by default in WTF and JSC types, on by default in WebCore types"? Makes sense to me, too. 160047 2012-08-22 17:38:52 -0700 2012-08-22 18:11:29 -0700 Patch bug-94761-20120822173849.patch text/plain 5689 ggaren SW5kZXg6IFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9DaGFuZ2VMb2cKPT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291 cmNlL0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkocmV2aXNpb24gMTI2MzY4KQorKysgU291cmNl L0phdmFTY3JpcHRDb3JlL0NoYW5nZUxvZwkod29ya2luZyBjb3B5KQpAQCAtMSwzICsxLDE4IEBA CisyMDEyLTA4LTIyICBHZW9mZnJleSBHYXJlbiAgPGdnYXJlbkBhcHBsZS5jb20+CisKKyAgICAg ICAgVGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllciBzaG91bGQgYmUgb3B0LWluLCBub3Qgb3B0LW91 dAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9OTQ3NjEK KworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAgICAgICBSZW1vdmVk IGV4cGxpY2l0IGNhbGxzIHRvIGRpc2FibGUgdGhlIHZlcmlmaWVyLCBzaW5jZSBpdCdzIG9mZiBi eSBkZWZhdWx0IG5vdy4KKworICAgICAgICAqIHBhcnNlci9Tb3VyY2VQcm92aWRlci5oOgorICAg ICAgICAoSlNDOjpTb3VyY2VQcm92aWRlcjo6U291cmNlUHJvdmlkZXIpOgorICAgICAgICAoU291 cmNlUHJvdmlkZXIpOgorICAgICAgICAqIHJ1bnRpbWUvU3ltYm9sVGFibGUuaDoKKyAgICAgICAg KEpTQzo6U2hhcmVkU3ltYm9sVGFibGU6OlNoYXJlZFN5bWJvbFRhYmxlKToKKwogMjAxMi0wOC0y MiAgTWFyayBIYWhuZW5iZXJnICA8bWhhaG5lbmJlcmdAYXBwbGUuY29tPgogCiAgICAgICAgIFNl cGFyYXRlIE1hcmtTdGFja1RocmVhZFNoYXJlZERhdGEgZnJvbSBNYXJrU3RhY2sKSW5kZXg6IFNv dXJjZS9KYXZhU2NyaXB0Q29yZS9wYXJzZXIvU291cmNlUHJvdmlkZXIuaAo9PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0t LSBTb3VyY2UvSmF2YVNjcmlwdENvcmUvcGFyc2VyL1NvdXJjZVByb3ZpZGVyLmgJKHJldmlzaW9u IDEyNjM2OCkKKysrIFNvdXJjZS9KYXZhU2NyaXB0Q29yZS9wYXJzZXIvU291cmNlUHJvdmlkZXIu aAkod29ya2luZyBjb3B5KQpAQCAtNDksOCArNDksOCBAQCBuYW1lc3BhY2UgSlNDIHsKICAgICAg ICAgICAgICwgbV9jYWNoZShjYWNoZSA/IGNhY2hlIDogbmV3IFNvdXJjZVByb3ZpZGVyQ2FjaGUp CiAgICAgICAgICAgICAsIG1fY2FjaGVPd25lZCghY2FjaGUpCiAgICAgICAgIHsKLSAgICAgICAg ICAgIHR1cm5PZmZWZXJpZmllcigpOwogICAgICAgICB9CisKICAgICAgICAgdmlydHVhbCB+U291 cmNlUHJvdmlkZXIoKQogICAgICAgICB7CiAgICAgICAgICAgICBpZiAobV9jYWNoZU93bmVkKQpJ bmRleDogU291cmNlL0phdmFTY3JpcHRDb3JlL3J1bnRpbWUvU3ltYm9sVGFibGUuaAo9PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09Ci0tLSBTb3VyY2UvSmF2YVNjcmlwdENvcmUvcnVudGltZS9TeW1ib2xUYWJsZS5oCShyZXZp c2lvbiAxMjYzNjgpCisrKyBTb3VyY2UvSmF2YVNjcmlwdENvcmUvcnVudGltZS9TeW1ib2xUYWJs ZS5oCSh3b3JraW5nIGNvcHkpCkBAIC0zMzAsNyArMzMwLDcgQEAgbmFtZXNwYWNlIEpTQyB7CiAg ICAgcHVibGljOgogICAgICAgICBzdGF0aWMgUGFzc1JlZlB0cjxTaGFyZWRTeW1ib2xUYWJsZT4g Y3JlYXRlKCkgeyByZXR1cm4gYWRvcHRSZWYobmV3IFNoYXJlZFN5bWJvbFRhYmxlKTsgfQogICAg IHByaXZhdGU6Ci0gICAgICAgIFNoYXJlZFN5bWJvbFRhYmxlKCkgeyB0dXJuT2ZmVmVyaWZpZXIo KTsgfQorICAgICAgICBTaGFyZWRTeW1ib2xUYWJsZSgpIHsgfQogICAgIH07CiAgICAgCiB9IC8v IG5hbWVzcGFjZSBKU0MKSW5kZXg6IFNvdXJjZS9UaGlyZFBhcnR5L0FOR0xFL0NoYW5nZUxvZwo9 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09Ci0tLSBTb3VyY2UvVGhpcmRQYXJ0eS9BTkdMRS9DaGFuZ2VMb2cJKHJldmlzaW9u IDEyNjM2OCkKKysrIFNvdXJjZS9UaGlyZFBhcnR5L0FOR0xFL0NoYW5nZUxvZwkod29ya2luZyBj b3B5KQpAQCAtMSwzICsxLDE0IEBACisyMDEyLTA4LTIyICBHZW9mZnJleSBHYXJlbiAgPGdnYXJl bkBhcHBsZS5jb20+CisKKyAgICAgICAgVGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllciBzaG91bGQg YmUgb3B0LWluLCBub3Qgb3B0LW91dAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9z aG93X2J1Zy5jZ2k/aWQ9OTQ3NjEKKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMh KS4KKworICAgICAgICBBZGRpdGlvbmFsIGluZm9ybWF0aW9uIG9mIHRoZSBjaGFuZ2Ugc3VjaCBh cyBhcHByb2FjaCwgcmF0aW9uYWxlLiBQbGVhc2UgYWRkIHBlci1mdW5jdGlvbiBkZXNjcmlwdGlv bnMgYmVsb3cgKE9PUFMhKS4KKworICAgICAgICAqIEFOR0xFLnhjb2RlcHJvai9wcm9qZWN0LnBi eHByb2o6CisKIDIwMTItMDctMTggIEtyaXN0w7NmIEtvc3p0ecOzICA8a2tyaXN0b2ZAaW5mLnUt c3plZ2VkLmh1PgogCiAgICAgICAgIFtRdF0gQnVpbGRmaXggYWZ0ZXIgcjEyMjg3MC4KSW5kZXg6 IFNvdXJjZS9XVEYvQ2hhbmdlTG9nCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFNvdXJjZS9XVEYvQ2hhbmdlTG9n CShyZXZpc2lvbiAxMjYzNjgpCisrKyBTb3VyY2UvV1RGL0NoYW5nZUxvZwkod29ya2luZyBjb3B5 KQpAQCAtMSwzICsxLDI3IEBACisyMDEyLTA4LTIyICBHZW9mZnJleSBHYXJlbiAgPGdnYXJlbkBh cHBsZS5jb20+CisKKyAgICAgICAgVGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllciBzaG91bGQgYmUg b3B0LWluLCBub3Qgb3B0LW91dAorICAgICAgICBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9zaG93 X2J1Zy5jZ2k/aWQ9OTQ3NjEKKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4K KworICAgICAgICBXZWJLaXQncyBKYXZhU2NyaXB0IGVuZ2luZSBzdXBwb3J0cyB1c2Ugb24gbXVs dGlwbGUgdGhyZWFkcywgc28gZGVmYXVsdC1vbgorICAgICAgICBpcyBub3QgYXBwcm9wcmlhdGUg Zm9yIG1vc3Qgb2Ygb3VyIG9iamVjdHMsIGFuZCBpdCBjYXVzZXMgbG90cyBvZiBzdXByaW91cwor ICAgICAgICBhc3NlcnRpb25zLgorCisgICAgICAgICogd3RmL01ldGFBbGxvY2F0b3IuY3BwOgor ICAgICAgICAoV1RGOjpNZXRhQWxsb2NhdG9ySGFuZGxlOjpNZXRhQWxsb2NhdG9ySGFuZGxlKTog Tm8gbmVlZCB0byB0dXJuIG9mZgorICAgICAgICBleHBsaWNpdGx5LCBzaW5jZSBpdCdzIG9mZiBi eSBkZWZhdWx0IG5vdy4KKworICAgICAgICAqIHd0Zi9UaHJlYWRSZXN0cmljdGlvblZlcmlmaWVy Lmg6CisgICAgICAgIChXVEY6OlRocmVhZFJlc3RyaWN0aW9uVmVyaWZpZXI6OlRocmVhZFJlc3Ry aWN0aW9uVmVyaWZpZXIpOiBUdXJuIG9mZiBieSBkZWZhdWx0LgorCisgICAgICAgIChXVEY6OlRo cmVhZFJlc3RyaWN0aW9uVmVyaWZpZXI6OnNldE11dGV4TW9kZSk6CisgICAgICAgIChXVEY6OlRo cmVhZFJlc3RyaWN0aW9uVmVyaWZpZXI6OnNldERpc3BhdGNoUXVldWVNb2RlKToKKyAgICAgICAg KFdURjo6VGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllcjo6dHVybk9mZlZlcmlmaWNhdGlvbik6IFRo ZXNlIGFzc2VydGlvbnMKKyAgICAgICAgYWJvdXQgc3RhdGUgdHJhbnNpdGlvbnMgd2VyZSBpbmNv bnNpc3RlbnQgd2l0aCBlYWNoIG90aGVyLCBhbmQgaW1wb3NzaWJsZQorICAgICAgICB0byBtYWlu dGFpbiB3aXRoIGRlZmF1bHQgb2ZmLCBzbyBJIHJlbW92ZWQgdGhlbS4KKwogMjAxMi0wOC0yMiAg QWxsYW4gU2FuZGZlbGQgSmVuc2VuICA8YWxsYW4uamVuc2VuQG5va2lhLmNvbT4KIAogICAgICAg ICBbUXRdIE9wdGlvbmFsbHkgc3VwcG9ydCBzbW9vdGgtc2Nyb2xsaW5nIG9uIGFsbCBwbGF0Zm9y bXMKSW5kZXg6IFNvdXJjZS9XVEYvd3RmL01ldGFBbGxvY2F0b3IuY3BwCj09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0t IFNvdXJjZS9XVEYvd3RmL01ldGFBbGxvY2F0b3IuY3BwCShyZXZpc2lvbiAxMjYzNjgpCisrKyBT b3VyY2UvV1RGL3d0Zi9NZXRhQWxsb2NhdG9yLmNwcAkod29ya2luZyBjb3B5KQpAQCAtNzksNyAr NzksNiBAQCBNZXRhQWxsb2NhdG9ySGFuZGxlOjpNZXRhQWxsb2NhdG9ySGFuZGxlCiAgICAgQVNT RVJUKGFsbG9jYXRvcik7CiAgICAgQVNTRVJUKHN0YXJ0KTsKICAgICBBU1NFUlQoc2l6ZUluQnl0 ZXMpOwotICAgIHR1cm5PZmZWZXJpZmllcigpOwogfQogCiBNZXRhQWxsb2NhdG9ySGFuZGxlOjp+ TWV0YUFsbG9jYXRvckhhbmRsZSgpCkluZGV4OiBTb3VyY2UvV1RGL3d0Zi9UaHJlYWRSZXN0cmlj dGlvblZlcmlmaWVyLmgKPT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0gU291cmNlL1dURi93dGYvVGhyZWFkUmVzdHJp Y3Rpb25WZXJpZmllci5oCShyZXZpc2lvbiAxMjYzNjgpCisrKyBTb3VyY2UvV1RGL3d0Zi9UaHJl YWRSZXN0cmljdGlvblZlcmlmaWVyLmgJKHdvcmtpbmcgY29weSkKQEAgLTUwLDcgKzUwLDExIEBA IG5hbWVzcGFjZSBXVEYgewogY2xhc3MgVGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllciB7CiBwdWJs aWM6CiAgICAgVGhyZWFkUmVzdHJpY3Rpb25WZXJpZmllcigpCisjaWYgVVNFKEpTQykKKyAgICAg ICAgOiBtX21vZGUoTm9WZXJpZmljYXRpb25Nb2RlKQorI2Vsc2UKICAgICAgICAgOiBtX21vZGUo U2luZ2xlVGhyZWFkVmVyaWZpY2F0aW9uTW9kZSkKKyNlbmRpZgogICAgICAgICAsIG1fc2hhcmVk KGZhbHNlKQogICAgICAgICAsIG1fb3duaW5nVGhyZWFkKDApCiAgICAgICAgICwgbV9tdXRleCgw KQpAQCAtNzAsNyArNzQsNiBAQCBwdWJsaWM6CiAKICAgICB2b2lkIHNldE11dGV4TW9kZShNdXRl eCYgbXV0ZXgpCiAgICAgewotICAgICAgICBBU1NFUlQobV9tb2RlID09IFNpbmdsZVRocmVhZFZl cmlmaWNhdGlvbk1vZGUgfHwgKG1fbW9kZSA9PSBNdXRleFZlcmlmaWNhdGlvbk1vZGUgJiYgJm11 dGV4ID09IG1fbXV0ZXgpKTsKICAgICAgICAgbV9tb2RlID0gTXV0ZXhWZXJpZmljYXRpb25Nb2Rl OwogICAgICAgICBtX211dGV4ID0gJm11dGV4OwogICAgIH0KQEAgLTc4LDcgKzgxLDYgQEAgcHVi bGljOgogI2lmIEhBVkUoRElTUEFUQ0hfSCkKICAgICB2b2lkIHNldERpc3BhdGNoUXVldWVNb2Rl KGRpc3BhdGNoX3F1ZXVlX3QgcXVldWUpCiAgICAgewotICAgICAgICBBU1NFUlQobV9tb2RlID09 IFNpbmdsZVRocmVhZFZlcmlmaWNhdGlvbk1vZGUpOwogICAgICAgICBtX21vZGUgPSBTaW5nbGVE aXNwYXRjaFF1ZXVlVmVyaWZpY2F0aW9uTW9kZTsKICAgICAgICAgbV9vd25pbmdRdWV1ZSA9IHF1 ZXVlOwogICAgICAgICBkaXNwYXRjaF9yZXRhaW4obV9vd25pbmdRdWV1ZSk7CkBAIC04Nyw3ICs4 OSw2IEBAIHB1YmxpYzoKIAogICAgIHZvaWQgdHVybk9mZlZlcmlmaWNhdGlvbigpCiAgICAgewot ICAgICAgICBBU1NFUlQobV9tb2RlID09IFNpbmdsZVRocmVhZFZlcmlmaWNhdGlvbk1vZGUpOwog ICAgICAgICBtX21vZGUgPSBOb1ZlcmlmaWNhdGlvbk1vZGU7CiAgICAgfQogCg==