119016 2013-07-23 10:18:36 -0700 REGRESSION (r151957): .gif images play at 10x speed for a few loops after loading 2013-07-30 14:03:44 -0700 1 1 1 Unclassified WebKit Images 528+ (Nightly build) Mac (Intel) OS X 10.8 RESOLVED FIXED InRadar, Regression P1 Minor --- 116041 1 cococohen1122 ossy allan.jensen ap ggaren mrowe ossy thorton zan oldest_to_newest 911075 0 cococohen1122 2013-07-23 10:18:36 -0700 Any time I load a .gif image in its own page, it loops through at 10x speed 911278 1 zan 2013-07-24 05:39:01 -0700 Can't reproduce on the GTK port, WebKit1 or WebKit2. 911332 2 ap 2013-07-24 09:39:44 -0700 I can reproduce on Mac using <http://ru.fishki.net/picsw/072013/23/post/gif/05_dem.gif>. Safari/WebKit 6.0.5: It plays very slowly for the first round while loading, then loops normally. With nightly WebKit r152754: It doesn't play at all while loading, then plays crazy fast for some time, then loops normally. This is not limited to standalone images - ones in a page like this one <http://fishki.net/comment.php?id=140726> also exhibit the same behavior. 911333 3 ap 2013-07-24 09:40:12 -0700 <rdar://problem/14534370> 911514 4 thorton 2013-07-24 22:41:51 -0700 Ossy, can you take a look? It looks like http://trac.webkit.org/changeset/151957 caused this. 911628 5 ossy 2013-07-25 08:50:07 -0700 (In reply to comment #4) > Ossy, can you take a look? It looks like http://trac.webkit.org/changeset/151957 caused this. Will check tomorrow. 911968 6 ossy 2013-07-26 07:19:06 -0700 I can't reproduce it on EFL WK1/WK2, Qt WK1/WK2 (r153370) and Nix (WK2 only port) (392d108b1ae5d42be00b7d96c7bd47689e742789, r153083 - last trunk merge) But http://fishki.net/comment.php?id=140726 crashes for me with EFL/Qt/Nix WK2 browsers. I'll investigate it. If you are sure that r151957 is the culprit, I'll find a Mac near here on monday and try to debug what happened. 911969 7 ossy 2013-07-26 07:27:53 -0700 Maybe there is some issue in platform/graphics/cg/ImageSourceCG.cpp. After r151957 frameDurationAtIndex doesn't trigger decoding the frame, maybe ImageSourceCG.cpp uses some invalid duration info. 911972 8 ossy 2013-07-26 07:40:53 -0700 I think this change doesn't fit somehow to the CG implementation: http://trac.webkit.org/changeset/151957/trunk/Source/WebCore/platform/graphics/BitmapImage.cpp Reverting the BitmapImage.cpp change with a PLATFORM(MAC) or something CG guard could be a workaround until we find the proper fix. Could you try if it fixes the issue, please? And I'll try to find the proper fix on monday. 912492 9 ossy 2013-07-29 07:28:25 -0700 I managed to reproduce this bug on Mac. For now I propose reverting r151957 only on Mac platform as a workaround. And then fix ImageSourceCG.cpp in a separated bug: https://bugs.webkit.org/show_bug.cgi?id=119218 912494 10 207656 ossy 2013-07-29 07:29:50 -0700 Created attachment 207656 Patch 912524 11 ossy 2013-07-29 09:50:05 -0700 I started debugging this quick animation bug and got the root of the problem, but I'm not sure how to fix it propely, but I'm on it. The problem in nutshell: The animation is started at the beginning and m_desiredFrameStartTime is initialized to the current time. But BitMapImage::frameIsCompleteAtIndex(index=1) is false until the last frame is loaded for some reason, which blocks the original very slow animation. I think it is _the_ bug somewhere. After the last frame is loaded, BitMapImage::frameIsCompleteAtIndex(index=1) becomes true and the animation is really started. Because the load time took ~5 seconds, m_desiredFrameStartTime is in the past, and the timer of the next frame immediately expired and so on. ... bool BitmapImage::frameIsCompleteAtIndex(size_t index) { if (index < m_frames.size() && m_frames[index].m_haveMetadata && m_frames[index].m_isComplete) return true; return m_source.frameIsCompleteAtIndex(index); <---- this one returns false! } ... 912533 12 ossy 2013-07-29 10:14:05 -0700 strange, the status returned by CGImageSourceGetStatusAtIndex() until the last frame is loaded is kCGImageStatusUnknownType = -3. 912752 13 207713 ossy 2013-07-30 03:39:41 -0700 Created attachment 207713 Patch Use USE(CG) guards instead of PLATFORM(MAC), because Windows port uses this code path too. 912754 14 ossy 2013-07-30 03:43:10 -0700 cc-ing Geoffrey and Mark as the authors of ImageSourceCG::frameIsCompleteAtIndex() - https://trac.webkit.org/changeset/55199 And I think https://trac.webkit.org/changeset/55199 is related to this bug too. 912763 15 207713 mrowe 2013-07-30 04:11:03 -0700 Comment on attachment 207713 Patch I really don't see the point in doing this rather than just fixing the problem. 912768 16 ossy 2013-07-30 04:18:26 -0700 (In reply to comment #15) > (From update of attachment 207713 [details]) > I really don't see the point in doing this rather than just fixing the problem. Because I have no idea how to fix it properly. So if you have any idea, it would be very welcome. 912770 17 ossy 2013-07-30 04:27:42 -0700 just a note: I can't fix CGImageSourceGetStatusAtIndex ( rdar://problem/7679174 ), because I have no access for Appple's internal rdar and code base. And I can't acces rdar://problem/14534370, so I don't know if anybody started working on it or if there is any discussion about it. 912777 18 mrowe 2013-07-30 04:42:30 -0700 It sounds like the changes made in <http://trac.webkit.org/changeset/151957> made assumptions about the behavior of ImageSource that aren't valid for all implementations. We should either understand what needs to be changed in the ImageSourceCG implementation to accommodate the changes made in r151957, or r151957 should be reverted. Adding #if's just heads down the path of divergent behavior which will make life difficult for everyone going forward. <rdar://problem/7679174> is unlikely to be relevant to whatever is happening here since it was fixed in Lion. Fixing CGImageSourceGetStatusAtIndex isn't relevant here either since that's OS X API and not part of WebKit. 912807 19 ossy 2013-07-30 06:08:21 -0700 The root of the problem is that http://trac.webkit.org/changeset/113486 "made a bunch of incorrect changes to BitmapImage.cpp" - https://bugs.webkit.org/show_bug.cgi?id=116041#c12 After r113486 BitmapImage::frameIsCompleteAtIndex, BitmapImage::frameHasAlphaAtIndex, ... trigger image decoding. Which is incorrect and caused flakey crashes everywhere. Chrome fixed it in Blink and r151957 fixed it in WebKit ... but unfortunately this fix doesn't fit to the ImageSourceCG implementation. Reverting r151957 is the worst choice, because it would paper over the problem and only push the bug from CG platforms to other platforms. My proposed workaround would let the CG use the previous and working version (which wastes resources with useless image decodings, but at least works) and let all other ports keep the fix of the buggy r113486 and not to have flakey crashes. If you want to revert r151957, we should revert r113486 and maybe many other patches on the top of it to be consistent and fix all ports. But let's try to collaborate a little bit instead of reverting everything. r20070 added ImageSource::frameIsCompleteAtIndex(): + bool frameIsCompleteAtIndex(size_t); // Whether or not the frame is completely decoded. ... diff --git a/WebCore/platform/graphics/BitmapImage.cpp b/WebCore/platform/graphics/BitmapImage.cpp index 136bc26..b4f66d2 100644 --- a/WebCore/platform/graphics/BitmapImage.cpp +++ b/WebCore/platform/graphics/BitmapImage.cpp @@ -209,6 +209,10 @@ void BitmapImage::startAnimation() if (m_frameTimer || !shouldAnimate() || frameCount() <= 1) return; + // Don't advance the animation until the current frame has completely loaded. + if (!m_source.frameIsCompleteAtIndex(m_currentFrame)) + return; + ... These comments are obvious and expects that ImageSource::frameIsCompleteAtIndex() returns with true if the given frames is "completely loaded" "Whether or not the frame is completely decoded.". But unfortunately ImageSourceCG implementation of ImageSource::frameIsCompleteAtIndex() doesn't pass this requirement. r151957 only revealed this bug and didn't caused. The problem is that we can't ask CGImageSource if the frame is loaded or not. And I can't fix CGImageSource, only Apple can do it. 912863 20 207713 ggaren 2013-07-30 09:05:37 -0700 Comment on attachment 207713 Patch A random walk that inserts some bugs while fixing others, peppering the code with #ifdefs as we go, is not how WebKit development works. 912865 21 ggaren 2013-07-30 09:06:30 -0700 Ossy, I'd like to give you a chance to fix r151957 -- but if you're still sure you can't fix it, I'll roll it out. 912893 22 ossy 2013-07-30 09:40:59 -0700 (In reply to comment #20) > (From update of attachment 207713 [details]) > A random walk that inserts some bugs while fixing others, peppering the code with #ifdefs as we go, is not how WebKit development works. I still think that r151957 was the correct fix and ImageSourceCG.cpp does the wrong thing. I'm sorry that I couldn't convince you about it in https://bugs.webkit.org/show_bug.cgi?id=119016#c19 :-( (In reply to comment #21) > Ossy, I'd like to give you a chance to fix r151957 -- but if you're still sure you can't fix it, I'll roll it out. Thanks for the chance, but as I said, I can't fix the 3rdparty CG library and have no idea how can we do a workaround for Mac inside WebKit. You won, let's break animated gifs everywhere except Mac. I reverted r151957 and its followup patch r152531 by http://trac.webkit.org/changeset/153475 913010 23 mrowe 2013-07-30 13:47:02 -0700 (In reply to comment #22) > You won, let's break animated gifs everywhere except Mac. The change that you rolled out was an optimization. It broke Mac. Rolling it out undoes the optimization, it doesn't break anything. 913016 24 ossy 2013-07-30 14:03:44 -0700 (In reply to comment #23) > (In reply to comment #22) > > You won, let's break animated gifs everywhere except Mac. > > The change that you rolled out was an optimization. It broke Mac. Rolling it out undoes the optimization, it doesn't break anything. No, it wasn't only an optimization, it was originally to fix flakey crashes during animated gifs on all non-CG platform. Maybe the name of the original bug was a little bit misleading. But the details can be found as comments. Right now I got an idea in the original bug how to fix it properly and will try it tomorrow. I hope all platforms would be happy soon. 207656 2013-07-29 07:29:50 -0700 2013-07-30 03:39:36 -0700 Patch bug-119016-20130729163012.patch text/plain 2400 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMTUzNDIzCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggODk5Y2E3ODM1OWZhYjY2 YzJmYTk4MGUzMDI0M2IzYTc2YTZmMmU5MS4uMTFlNTk1ZTdjM2I0MGNlZWIxNDFlNGQzNDA2MjM2 ZTFjN2Q1OTU4ZCAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDE3IEBACisyMDEzLTA3LTI5ICBDc2Fi YSBPc3p0cm9nb27DoWMgIDxvc3N5QHdlYmtpdC5vcmc+CisKKyAgICAgICAgUkVHUkVTU0lPTiAo cjE1MTk1Nyk6IC5naWYgaW1hZ2VzIHBsYXkgYXQgMTB4IHNwZWVkIGZvciBhIGZldyBsb29wcyBh ZnRlciBsb2FkaW5nCisgICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNn aT9pZD0xMTkwMTYKKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAg ICAgICBSZXZlcnQgcjE1MTk1NyBmb3IgTWFjIGFzIGEgd29ya2Fyb3VuZCB1bnRpbCBwcm9wZXIg Zml4IGZvcgorICAgICAgICBJbWFnZVNvdXJjZUNHLmNwcCBpbiBodHRwczovL2J1Z3Mud2Via2l0 Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MTE5MjE4CisKKyAgICAgICAgKiBwbGF0Zm9ybS9ncmFwaGlj cy9CaXRtYXBJbWFnZS5jcHA6CisgICAgICAgIChXZWJDb3JlOjpCaXRtYXBJbWFnZTo6ZnJhbWVJ c0NvbXBsZXRlQXRJbmRleCk6CisgICAgICAgIChXZWJDb3JlOjpCaXRtYXBJbWFnZTo6ZnJhbWVE dXJhdGlvbkF0SW5kZXgpOgorCiAyMDEzLTA3LTI5ICBLd2FuZyBZdWwgU2VvICA8c2t5dWxAY29t cGFueTEwMC5uZXQ+CiAKICAgICAgICAgW0VGTF0gSW1wcm92ZSBzdHJpbmcgdXNlIGluIEVycm9y c0VmbApkaWZmIC0tZ2l0IGEvU291cmNlL1dlYkNvcmUvcGxhdGZvcm0vZ3JhcGhpY3MvQml0bWFw SW1hZ2UuY3BwIGIvU291cmNlL1dlYkNvcmUvcGxhdGZvcm0vZ3JhcGhpY3MvQml0bWFwSW1hZ2Uu Y3BwCmluZGV4IDA3NjEwZDJkODNlZTA2OGI1ZWFhZGZlNjczNmEyN2Y3YTVjYjg1MzkuLjQyNzM5 MTkxYzA4NDkzZDFhMGVlM2QxMjY0YmQ0YjU0ZDhiNjAyYzIgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9X ZWJDb3JlL3BsYXRmb3JtL2dyYXBoaWNzL0JpdG1hcEltYWdlLmNwcAorKysgYi9Tb3VyY2UvV2Vi Q29yZS9wbGF0Zm9ybS9ncmFwaGljcy9CaXRtYXBJbWFnZS5jcHAKQEAgLTMwNSwxNiArMzA1LDMy IEBAIFBhc3NOYXRpdmVJbWFnZVB0ciBCaXRtYXBJbWFnZTo6ZnJhbWVBdEluZGV4KHNpemVfdCBp bmRleCkKIAogYm9vbCBCaXRtYXBJbWFnZTo6ZnJhbWVJc0NvbXBsZXRlQXRJbmRleChzaXplX3Qg aW5kZXgpCiB7CisvLyBGSVhNRTogSW1hZ2VTb3VyY2VDRy5jcHAgc2hvdWxkbid0IGV4cGVjdCB0 aGF0IGZyYW1lSXNDb21wbGV0ZUF0SW5kZXgoKSBkZWNvZGVzIHRoZSBpbWFnZQorLy8gaHR0cHM6 Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTExOTIxOAorI2lmIFBMQVRGT1JNKE1B QykKKyAgICBpZiAoIWVuc3VyZUZyYW1lSXNDYWNoZWQoaW5kZXgpKQorICAgICAgICByZXR1cm4g ZmFsc2U7CisgICAgcmV0dXJuIG1fZnJhbWVzW2luZGV4XS5tX2lzQ29tcGxldGU7CisjZWxzZQog ICAgIGlmIChpbmRleCA8IG1fZnJhbWVzLnNpemUoKSAmJiBtX2ZyYW1lc1tpbmRleF0ubV9oYXZl TWV0YWRhdGEgJiYgbV9mcmFtZXNbaW5kZXhdLm1faXNDb21wbGV0ZSkKICAgICAgICAgcmV0dXJu IHRydWU7CiAgICAgcmV0dXJuIG1fc291cmNlLmZyYW1lSXNDb21wbGV0ZUF0SW5kZXgoaW5kZXgp OworI2VuZGlmCiB9CiAKIGZsb2F0IEJpdG1hcEltYWdlOjpmcmFtZUR1cmF0aW9uQXRJbmRleChz aXplX3QgaW5kZXgpCiB7CisvLyBGSVhNRTogSW1hZ2VTb3VyY2VDRy5jcHAgc2hvdWxkbid0IGV4 cGVjdCB0aGF0IGZyYW1lRHVyYXRpb25BdEluZGV4KCkgZGVjb2RlcyB0aGUgaW1hZ2UKKy8vIGh0 dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD0xMTkyMTgKKyNpZiBQTEFURk9S TShNQUMpCisgICAgaWYgKCFlbnN1cmVGcmFtZUlzQ2FjaGVkKGluZGV4KSkKKyAgICAgICAgcmV0 dXJuIDA7CisgICAgcmV0dXJuIG1fZnJhbWVzW2luZGV4XS5tX2R1cmF0aW9uOworI2Vsc2UKICAg ICBpZiAoaW5kZXggPCBtX2ZyYW1lcy5zaXplKCkgJiYgbV9mcmFtZXNbaW5kZXhdLm1faGF2ZU1l dGFkYXRhKQogICAgICAgICByZXR1cm4gbV9mcmFtZXNbaW5kZXhdLm1fZHVyYXRpb247CiAgICAg cmV0dXJuIG1fc291cmNlLmZyYW1lRHVyYXRpb25BdEluZGV4KGluZGV4KTsKKyNlbmRpZgogfQog CiBQYXNzTmF0aXZlSW1hZ2VQdHIgQml0bWFwSW1hZ2U6Om5hdGl2ZUltYWdlRm9yQ3VycmVudEZy YW1lKCkK 207713 2013-07-30 03:39:41 -0700 2013-07-30 09:05:37 -0700 Patch bug-119016-20130730124004.patch text/plain 2396 ossy U3VidmVyc2lvbiBSZXZpc2lvbjogMTUzNDYwCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9D aGFuZ2VMb2cgYi9Tb3VyY2UvV2ViQ29yZS9DaGFuZ2VMb2cKaW5kZXggYTQ3NmQ4Njc5NTc5ZThm ZTlkYTQ3OTRlNTdhN2E2ZDliZDlkMmMyMC4uYzA0YmYxNzEwMTkwNTZmOTIyMDM2ZWRkZGZmNGNh NmZlZWY4MWUwNSAxMDA2NDQKLS0tIGEvU291cmNlL1dlYkNvcmUvQ2hhbmdlTG9nCisrKyBiL1Nv dXJjZS9XZWJDb3JlL0NoYW5nZUxvZwpAQCAtMSwzICsxLDE3IEBACisyMDEzLTA3LTMwICBDc2Fi YSBPc3p0cm9nb27DoWMgIDxvc3N5QHdlYmtpdC5vcmc+CisKKyAgICAgICAgUkVHUkVTU0lPTiAo cjE1MTk1Nyk6IC5naWYgaW1hZ2VzIHBsYXkgYXQgMTB4IHNwZWVkIGZvciBhIGZldyBsb29wcyBh ZnRlciBsb2FkaW5nCisgICAgICAgIGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNn aT9pZD0xMTkwMTYKKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAg ICAgICBSZXZlcnQgcjE1MTk1NyBmb3IgQ0cgcGxhdGZvcm1zIGFzIGEgd29ya2Fyb3VuZCB1bnRp bCBwcm9wZXIgZml4IGZvcgorICAgICAgICBJbWFnZVNvdXJjZUNHLmNwcCBpbiBodHRwczovL2J1 Z3Mud2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MTE5MjE4CisKKyAgICAgICAgKiBwbGF0Zm9y bS9ncmFwaGljcy9CaXRtYXBJbWFnZS5jcHA6CisgICAgICAgIChXZWJDb3JlOjpCaXRtYXBJbWFn ZTo6ZnJhbWVJc0NvbXBsZXRlQXRJbmRleCk6CisgICAgICAgIChXZWJDb3JlOjpCaXRtYXBJbWFn ZTo6ZnJhbWVEdXJhdGlvbkF0SW5kZXgpOgorCiAyMDEzLTA3LTI5ICBDYXJsb3MgR2FyY2lhIENh bXBvcyAgPGNnYXJjaWFAaWdhbGlhLmNvbT4KIAogICAgICAgICBVbnJldmlld2VkLiBGaXggbWFr ZSBkaXN0Y2hlY2suCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9ncmFwaGlj cy9CaXRtYXBJbWFnZS5jcHAgYi9Tb3VyY2UvV2ViQ29yZS9wbGF0Zm9ybS9ncmFwaGljcy9CaXRt YXBJbWFnZS5jcHAKaW5kZXggMDc2MTBkMmQ4M2VlMDY4YjVlYWFkZmU2NzM2YTI3ZjdhNWNiODUz OS4uMWFlMmUyNzQxNmMyNjIzMDcwZjQ2MThhYTk5NGQ4YWEyNzg2YjFlMSAxMDA2NDQKLS0tIGEv U291cmNlL1dlYkNvcmUvcGxhdGZvcm0vZ3JhcGhpY3MvQml0bWFwSW1hZ2UuY3BwCisrKyBiL1Nv dXJjZS9XZWJDb3JlL3BsYXRmb3JtL2dyYXBoaWNzL0JpdG1hcEltYWdlLmNwcApAQCAtMzA1LDE2 ICszMDUsMzIgQEAgUGFzc05hdGl2ZUltYWdlUHRyIEJpdG1hcEltYWdlOjpmcmFtZUF0SW5kZXgo c2l6ZV90IGluZGV4KQogCiBib29sIEJpdG1hcEltYWdlOjpmcmFtZUlzQ29tcGxldGVBdEluZGV4 KHNpemVfdCBpbmRleCkKIHsKKy8vIEZJWE1FOiBJbWFnZVNvdXJjZUNHLmNwcCBzaG91bGRuJ3Qg ZXhwZWN0IHRoYXQgZnJhbWVJc0NvbXBsZXRlQXRJbmRleCgpIGRlY29kZXMgdGhlIGltYWdlCisv LyBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MTE5MjE4CisjaWYgVVNF KENHKQorICAgIGlmICghZW5zdXJlRnJhbWVJc0NhY2hlZChpbmRleCkpCisgICAgICAgIHJldHVy biBmYWxzZTsKKyAgICByZXR1cm4gbV9mcmFtZXNbaW5kZXhdLm1faXNDb21wbGV0ZTsKKyNlbHNl CiAgICAgaWYgKGluZGV4IDwgbV9mcmFtZXMuc2l6ZSgpICYmIG1fZnJhbWVzW2luZGV4XS5tX2hh dmVNZXRhZGF0YSAmJiBtX2ZyYW1lc1tpbmRleF0ubV9pc0NvbXBsZXRlKQogICAgICAgICByZXR1 cm4gdHJ1ZTsKICAgICByZXR1cm4gbV9zb3VyY2UuZnJhbWVJc0NvbXBsZXRlQXRJbmRleChpbmRl eCk7CisjZW5kaWYKIH0KIAogZmxvYXQgQml0bWFwSW1hZ2U6OmZyYW1lRHVyYXRpb25BdEluZGV4 KHNpemVfdCBpbmRleCkKIHsKKy8vIEZJWE1FOiBJbWFnZVNvdXJjZUNHLmNwcCBzaG91bGRuJ3Qg ZXhwZWN0IHRoYXQgZnJhbWVEdXJhdGlvbkF0SW5kZXgoKSBkZWNvZGVzIHRoZSBpbWFnZQorLy8g aHR0cHM6Ly9idWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTExOTIxOAorI2lmIFVTRShD RykKKyAgICBpZiAoIWVuc3VyZUZyYW1lSXNDYWNoZWQoaW5kZXgpKQorICAgICAgICByZXR1cm4g MDsKKyAgICByZXR1cm4gbV9mcmFtZXNbaW5kZXhdLm1fZHVyYXRpb247CisjZWxzZQogICAgIGlm IChpbmRleCA8IG1fZnJhbWVzLnNpemUoKSAmJiBtX2ZyYW1lc1tpbmRleF0ubV9oYXZlTWV0YWRh dGEpCiAgICAgICAgIHJldHVybiBtX2ZyYW1lc1tpbmRleF0ubV9kdXJhdGlvbjsKICAgICByZXR1 cm4gbV9zb3VyY2UuZnJhbWVEdXJhdGlvbkF0SW5kZXgoaW5kZXgpOworI2VuZGlmCiB9CiAKIFBh c3NOYXRpdmVJbWFnZVB0ciBCaXRtYXBJbWFnZTo6bmF0aXZlSW1hZ2VGb3JDdXJyZW50RnJhbWUo KQo=