38557 2010-05-04 18:03:42 -0700 r58526 introduced a ~30% regression on Dromaeo JS lib 2010-05-14 08:59:24 -0700 1 1 1 Unclassified WebKit New Bugs 528+ (Nightly build) PC OS X 10.5 RESOLVED FIXED P2 Normal --- 1 barraclough sam antonm darin eric oliver sam vitalyr oldest_to_newest 220914 0 barraclough 2010-05-04 18:03:42 -0700 This change -> http://trac.webkit.org/changeset/58526 Regressed Dromaeo JS lib tests by ~30% on my machine. That said, it was about an equally huge win on Dromaeo DOM core (awesome!), but given our no-regressions policy I'd imagine we'll need to find a way to avoid the performance hit if we are to keep this in the tree. Might be best to revert & re-land a new patch, once this issue has been resolved? The hit seems to mostly fall on the CSS style & DOM traversal tests. http://dromaeo.com/?id=102238 http://dromaeo.com/?id=102247 Darin, cc'ing you as reviewer of the original patch. 221093 1 antonm 2010-05-05 07:12:33 -0700 Gavin, sorry for introducing that. What is interesting there is nothing like that for Chromium: http://build.chromium.org/buildbot/perf/xp-release-single-core/dromaeo_jslib/report.html?history=350&header=&thumbnail=undefined&rev=-1&graph=jslib_traverse_prototype The only difference between Safari and Chromium that immediately come into my mind is change to GC policy---I haven't yet implemented retention of cached node lists in v8's bindings (will implement this week). Gavin, if you have some spare cycles, may I ask you to comment out GC thing (this is the only change in JSNodeCustom.cpp: http://trac.webkit.org/changeset/58526/trunk/WebCore/bindings/js/JSNodeCustom.cpp) and check if it solves the problem? If you don't have any cycles, just let me know and I'll investigate. If possible, I'd ask not to revert it for, say, a week---I'll start working on it immediately. And just for your information: there is a pretty convenient way to compare several Dromaeo runs---just separate run ids with comma: http://dromaeo.com/?id=102238,102247 (In reply to comment #0) > This change -> http://trac.webkit.org/changeset/58526 > > Regressed Dromaeo JS lib tests by ~30% on my machine. That said, it was about > an equally huge win on Dromaeo DOM core (awesome!), but given our > no-regressions policy I'd imagine we'll need to find a way to avoid the > performance hit if we are to keep this in the tree. Might be best to revert & > re-land a new patch, once this issue has been resolved? > > The hit seems to mostly fall on the CSS style & DOM traversal tests. > > http://dromaeo.com/?id=102238 > http://dromaeo.com/?id=102247 > > Darin, cc'ing you as reviewer of the original patch. 221152 2 oliver 2010-05-05 09:50:22 -0700 My thoughts would be to see whether the cost is in GC overhead (that being the obvious culprit if we're now caching GC objects) and failing that i would look at what is causing us to miss the cache (which would be the second most obvious culprit) 221512 3 barraclough 2010-05-05 22:40:27 -0700 Hi Anton, I'm fairly snowed under myself this week, so Sam Weinig has kindly offered to step in for me and try to help. I think he plans to start with the basics, take some shark profiles & see what shows up. I'm sure he'll let you know once he has any info. many thanks, G. 221745 4 antonm 2010-05-06 09:45:21 -0700 (In reply to comment #3) > Hi Anton, > > I'm fairly snowed under myself this week, so Sam Weinig has kindly offered to > step in for me and try to help. I think he plans to start with the basics, > take some shark profiles & see what shows up. I'm sure he'll let you know once > he has any info. > > many thanks, > G. Sure, Gavin. So the main suspect is GC for now: http://dromaeo.com/?id=102562,102570,102580 . First one is head of WebKit, the second one is head of WebKit with my change backed out, the last one is head of WebKit with the following change: $ git diff diff --git a/WebCore/bindings/js/JSNodeCustom.cpp b/WebCore/bindings/js/JSNodeCu index 6d61037..c2ff970 100644 --- a/WebCore/bindings/js/JSNodeCustom.cpp +++ b/WebCore/bindings/js/JSNodeCustom.cpp @@ -179,7 +179,7 @@ void JSNode::markChildren(MarkStack& markStack) Node* node = m_impl.get(); node->markJSEventListeners(markStack); - node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); + // node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); // Nodes in the document are kept alive by JSDocument::mark, so, if we're i // the document, we need to mark the document, but we don't need to explici That's on windows box, but I hope to repro it on MBP as well. Looking further. 221748 5 antonm 2010-05-06 09:46:30 -0700 And I assign it to myself. But if anyone else wishes to grab it, that's fine, I'll still keep investigating. 221752 6 darin 2010-05-06 09:48:17 -0700 (In reply to comment #4) > - node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); > + // node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); To get maximum speed on this line of code, we need to make sure we do the minimum amount of work on nodes without outstanding node lists. Finding the heap for "this" so we can get the global data pointer is one example of unwanted work. Also, we need to inline much of the checking for node lists, including at least the rare data bit check and possibly also the rare data hash table lookup and then the subsequent check for the cached node lists pointer inside the rare data structure. At the moment, I believe we do have the rare data bit check inlined, so the issue may be the overhead in getting the global data pointer. If worse comes to worst we might have to devote a bit in the Node to indicating whether we have any cached node lists. 221755 7 antonm 2010-05-06 09:51:41 -0700 (In reply to comment #6) > (In reply to comment #4) > > - node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); > > + // node->markCachedNodeLists(markStack, *Heap::heap(this)->globalData()); > > To get maximum speed on this line of code, we need to make sure we do the > minimum amount of work on nodes without outstanding node lists. Finding the > heap for "this" so we can get the global data pointer is one example of > unwanted work. Also, we need to inline much of the checking for node lists, > including at least the rare data bit check and possibly also the rare data hash > table lookup and then the subsequent check for the cached node lists pointer > inside the rare data structure. > > At the moment, I believe we do have the rare data bit check inlined, so the > issue may be the overhead in getting the global data pointer. > > If worse comes to worst we might have to devote a bit in the Node to indicating > whether we have any cached node lists. Agree, and thanks a lot for your suggestions. Will investigate various approaches. I inlined by your suggestion a fast check for presence of rare data, but will start with inlining the check for presence of cached node lists. Let's hope it's not a node with tons on node lists cached on it. 221758 8 darin 2010-05-06 09:54:21 -0700 (In reply to comment #7) > I inlined by your suggestion a fast check for presence of rare > data, but will start with inlining the check for presence of cached node lists. > Let's hope it's not a node with tons on node lists cached on it. I think that instead, the right first step is to restructure to eliminate the call to *Heap::heap(this)->globalData() when rare data is not present. 221760 9 antonm 2010-05-06 09:55:49 -0700 (In reply to comment #8) > (In reply to comment #7) > > I inlined by your suggestion a fast check for presence of rare > > data, but will start with inlining the check for presence of cached node lists. > > Let's hope it's not a node with tons on node lists cached on it. > > I think that instead, the right first step is to restructure to eliminate the > call to *Heap::heap(this)->globalData() when rare data is not present. Thanks a lot, Darin, going to give it a try right now. 221763 10 darin 2010-05-06 09:58:03 -0700 The easiest way is probably to remove the JSGlobalData* argument and add a JSObject* argument instead. 221849 11 55276 antonm 2010-05-06 11:36:58 -0700 Created attachment 55276 Late Heap lookup Alas, this doesn't solve the problem. Any obvious mistakes? 221887 12 antonm 2010-05-06 12:40:09 -0700 [+ Vitaly] So we carried out couple of experiments. Current results: 1) speeding up fast checks doesn't help: see patch above, I ran another patch which uses bit flag when cached node lists are present---regression still persists; 2) heap profiler (in Chromium) hints that Dromaeo tests retain documents (one per suite like "DOM Attributes (jQuery)" and those apparently keep node lists; 3) if one runs 'regressed suite' separately, omitting suites before, there is no regression: http://dromaeo.com/?id=102605,102610 So the best hypothesis for now (which I will check tomorrow) is we've got 'leaked' cached node lists which sit and make subsequent GCs slower. If that's true, ideally Dromaeo should have cleaned up its garbage (and I'll experiment with it nulling out globals in suites). But anyway, esp. as JSC heap is not segregated by DOMWindow yet, we probably need to collect node lists if they don't have custom properties set. Thoughts? 221929 13 antonm 2010-05-06 13:36:03 -0700 BTW, Darin, do you remember why it was decided that we need to retain cached node lists as long as node is alive? I remember it was due to failing layout tests, but is it indeed desired semantics? I'll check tomorrow HTML5 wording, but any insights are most appreciated. 221996 14 darin 2010-05-06 15:13:17 -0700 (In reply to comment #13) > BTW, Darin, do you remember why it was decided that we need to retain cached > node lists as long as node is alive? I remember it was due to failing layout > tests, but is it indeed desired semantics? > > I'll check tomorrow HTML5 wording, but any insights are most appreciated. The general principle is that garbage collection must not be detectable. So if someone adds custom properties to a node list it’s important we don’t lose those. That’s the main concept. The rest is just about performance tuning, I suppose. 222044 15 sam 2010-05-06 16:31:42 -0700 I did the experiment of running the whole Dromaeo JavaScript Library Test and sharking just "DOM Traversal (Prototype)" (that one was particularly slow. It seems an exorbitant amount of time is being spent creating the emptyValue QualifiedName for the TagNodeListCache in markNodeLists. The emptyValue Qualified name can be greatly simplified in one of two ways, 1) use a 0 value for QualifiedName::m_impl or 2) use a static global. I am trying #1 now and will report back the results. In addition, I am going to investigate adding a call to hasCustomProperties() in markDOMObjectWrapper (or something similar that actually works, like adding a new version of markDOMObjectWrapper that does this) to avoid keeping alive the wrappers when there are no observable properties. 222169 16 antonm 2010-05-06 22:25:12 -0700 (In reply to comment #14) > (In reply to comment #13) > > BTW, Darin, do you remember why it was decided that we need to retain cached > > node lists as long as node is alive? I remember it was due to failing layout > > tests, but is it indeed desired semantics? > > > > I'll check tomorrow HTML5 wording, but any insights are most appreciated. > > The general principle is that garbage collection must not be detectable. So if > someone adds custom properties to a node list it’s important we don’t lose > those. That’s the main concept. The rest is just about performance tuning, I > suppose. Darin, I'll respond somewhat later---I need to carry on couple of experiments before. 222170 17 antonm 2010-05-06 22:26:56 -0700 (In reply to comment #15) > I did the experiment of running the whole Dromaeo JavaScript Library Test and > sharking just "DOM Traversal (Prototype)" (that one was particularly slow. It > seems an exorbitant amount of time is being spent creating the emptyValue > QualifiedName for the TagNodeListCache in markNodeLists. The emptyValue > Qualified name can be greatly simplified in one of two ways, 1) use a 0 value > for QualifiedName::m_impl or 2) use a static global. I am trying #1 now and > will report back the results. > > In addition, I am going to investigate adding a call to hasCustomProperties() > in markDOMObjectWrapper (or something similar that actually works, like adding > a new version of markDOMObjectWrapper that does this) to avoid keeping alive > the wrappers when there are no observable properties. Sam, thanks a lot. Any news about your experiments? I am asking as in ~3 hrs I'll start working on it and I'd rather not duplicate the work you're already doing. 222532 18 55404 antonm 2010-05-07 12:15:30 -0700 Created attachment 55404 Next patch: faster emptyValue 222539 19 antonm 2010-05-07 12:17:54 -0700 (In reply to comment #18) > Created an attachment (id=55404) [details] > Next patch: faster emptyValue This patch restores notable amount of score for me---this is all due to faster emptyValue (thanks to Sam's investigation). Would appreciate if anyone else checks if it helps on their box. So the main suspect for now would be iteration over cached node lists. I am going to use NodeListsNodeData::m_listsWithCaches to see if it'll help. 222559 20 55408 sam 2010-05-07 12:37:37 -0700 Created attachment 55408 My emptyValue patch 222711 21 55404 darin 2010-05-07 17:05:43 -0700 Comment on attachment 55404 Next patch: faster emptyValue This is another copy of your JSGlobalData patch, not an emptyValue one. I like Sam's approach to the emptyValue issue -- I suggest we use his. 222730 22 55442 sam 2010-05-07 17:47:36 -0700 Created attachment 55442 Fix 222731 23 55442 darin 2010-05-07 17:48:42 -0700 Comment on attachment 55442 Fix > + - Only mark cached NodeLists on Documents instead of all Nodes. Why is this OK? 222971 24 eric 2010-05-08 23:15:51 -0700 Attachment 55442 was posted by a committer and has review+, assigning to Sam Weinig for commit. 223060 25 sam 2010-05-09 13:34:58 -0700 Landed in r59058. 223920 26 antonm 2010-05-11 07:35:31 -0700 (In reply to comment #21) > (From update of attachment 55404 [details]) > This is another copy of your JSGlobalData patch, not an emptyValue one. > > I like Sam's approach to the emptyValue issue -- I suggest we use his. Sorry, wrong patch. 223921 27 antonm 2010-05-11 07:36:16 -0700 (In reply to comment #23) > (From update of attachment 55442 [details]) > > + - Only mark cached NodeLists on Documents instead of all Nodes. > > Why is this OK? I'd be curious to know as well. And thanks a lot for fixing the problem, Sam. 223998 28 darin 2010-05-11 10:34:34 -0700 (In reply to comment #27) > (In reply to comment #23) > > (From update of attachment 55442 [details] [details]) > > > + - Only mark cached NodeLists on Documents instead of all Nodes. > > > > Why is this OK? > > I'd be curious to know as well. And thanks a lot for fixing the problem, Sam. Sam answered my question in person. Sam, when you get a chance can you comment here in writing? 225748 29 antonm 2010-05-14 04:15:14 -0700 (In reply to comment #28) > (In reply to comment #27) > > (In reply to comment #23) > > > (From update of attachment 55442 [details] [details] [details]) > > > > + - Only mark cached NodeLists on Documents instead of all Nodes. > > > > > > Why is this OK? > > > > I'd be curious to know as well. And thanks a lot for fixing the problem, Sam. > > Sam answered my question in person. Sam, when you get a chance can you comment here in writing? Sam, Darin, sorry for being pushy, but could you clarify the story with Document? I'd like to adjust v8's GC as well, but would like to have it as similar to JSC as possible. 225859 30 darin 2010-05-14 08:59:24 -0700 Sam should answer the question. Here's what I remember: Part of the answer is that almost all use of these functions is on Document. Another part is that collection of the JavaScript wrapper only affects the lifetime of custom properties on the node list in cases where no one is retaining a pointer to that list. 55276 2010-05-06 11:36:58 -0700 2010-05-06 11:36:58 -0700 Late Heap lookup patch.diff text/plain 2527 antonm SW5kZXg6IFdlYkNvcmUvZG9tL05vZGUuaAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBXZWJDb3JlL2RvbS9Ob2Rl LmgJKHJldmlzaW9uIDU4ODI2KQorKysgV2ViQ29yZS9kb20vTm9kZS5oCSh3b3JraW5nIGNvcHkp CkBAIC0zNiw2ICszNiw3IEBACiBuYW1lc3BhY2UgSlNDIHsKIAogICAgIGNsYXNzIEpTR2xvYmFs RGF0YTsKKyAgICBjbGFzcyBKU09iamVjdDsKICAgICBjbGFzcyBNYXJrU3RhY2s7CiAKIH0KQEAg LTU4MCwxMyArNTgxLDEzIEBACiAgICAgdmlydHVhbCBFdmVudFRhcmdldERhdGEqIGVuc3VyZUV2 ZW50VGFyZ2V0RGF0YSgpOwogCiAjaWYgVVNFKEpTQykKLSAgICB2b2lkIG1hcmtDYWNoZWROb2Rl TGlzdHMoSlNDOjpNYXJrU3RhY2smIG1hcmtTdGFjaywgSlNDOjpKU0dsb2JhbERhdGEmIGdsb2Jh bERhdGEpCisgICAgdm9pZCBtYXJrQ2FjaGVkTm9kZUxpc3RzKEpTQzo6TWFya1N0YWNrJiBtYXJr U3RhY2ssIEpTQzo6SlNPYmplY3QqIHdyYXBwZXIpCiAgICAgewogICAgICAgICAvLyBOb2RlTGlz dHMgbWF5IGJlIHByZXNlbnQuICBJZiBzbywgdGhleSBuZWVkIHRvIGJlIG1hcmtlZC4KICAgICAg ICAgaWYgKCFoYXNSYXJlRGF0YSgpKQogICAgICAgICAgICAgcmV0dXJuOwogCi0gICAgICAgIG1h cmtDYWNoZWROb2RlTGlzdHNTbG93KG1hcmtTdGFjaywgZ2xvYmFsRGF0YSk7CisgICAgICAgIG1h cmtDYWNoZWROb2RlTGlzdHNTbG93KG1hcmtTdGFjaywgd3JhcHBlcik7CiAgICAgfQogI2VuZGlm CiAKQEAgLTYxMiw3ICs2MTMsNyBAQAogCiBwcml2YXRlOgogI2lmIFVTRShKU0MpCi0gICAgdm9p ZCBtYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtTdGFjayYsIEpTQzo6SlNHbG9iYWxE YXRhJik7CisgICAgdm9pZCBtYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtTdGFjayYs IEpTQzo6SlNPYmplY3QqKTsKICNlbmRpZgogCiAgICAgc3RhdGljIGJvb2wgaW5pdGlhbFJlZkNv dW50KENvbnN0cnVjdGlvblR5cGUpOwpJbmRleDogV2ViQ29yZS9kb20vTm9kZS5jcHAKPT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PQotLS0gV2ViQ29yZS9kb20vTm9kZS5jcHAJKHJldmlzaW9uIDU4ODI2KQorKysgV2ViQ29y ZS9kb20vTm9kZS5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTI1NjQsMTIgKzI1NjQsMTMgQEAKICAg ICAgICAgbWFya0RPTU9iamVjdFdyYXBwZXIobWFya1N0YWNrLCBnbG9iYWxEYXRhLCBpdC0+c2Vj b25kKTsKIH0KIAotdm9pZCBOb2RlOjptYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtT dGFjayYgbWFya1N0YWNrLCBKU0M6OkpTR2xvYmFsRGF0YSYgZ2xvYmFsRGF0YSkKK3ZvaWQgTm9k ZTo6bWFya0NhY2hlZE5vZGVMaXN0c1Nsb3coSlNDOjpNYXJrU3RhY2smIG1hcmtTdGFjaywgSlND OjpKU09iamVjdCogd3JhcHBlcikKIHsKICAgICBOb2RlTGlzdHNOb2RlRGF0YSogbm9kZUxpc3Rz ID0gcmFyZURhdGEoKS0+bm9kZUxpc3RzKCk7CiAgICAgaWYgKCFub2RlTGlzdHMpCiAgICAgICAg IHJldHVybjsKIAorICAgIEpTQzo6SlNHbG9iYWxEYXRhJiBnbG9iYWxEYXRhID0gKkpTQzo6SGVh cDo6aGVhcCh3cmFwcGVyKS0+Z2xvYmFsRGF0YSgpOwogICAgIG1hcmtOb2RlTGlzdHMobm9kZUxp c3RzLT5tX2NsYXNzTm9kZUxpc3RDYWNoZSwgbWFya1N0YWNrLCBnbG9iYWxEYXRhKTsKICAgICBt YXJrTm9kZUxpc3RzKG5vZGVMaXN0cy0+bV9uYW1lTm9kZUxpc3RDYWNoZSwgbWFya1N0YWNrLCBn bG9iYWxEYXRhKTsKICAgICBtYXJrTm9kZUxpc3RzKG5vZGVMaXN0cy0+bV90YWdOb2RlTGlzdENh Y2hlLCBtYXJrU3RhY2ssIGdsb2JhbERhdGEpOwpJbmRleDogV2ViQ29yZS9iaW5kaW5ncy9qcy9K U05vZGVDdXN0b20uY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFdlYkNvcmUvYmluZGluZ3MvanMvSlNOb2Rl Q3VzdG9tLmNwcAkocmV2aXNpb24gNTg4MjYpCisrKyBXZWJDb3JlL2JpbmRpbmdzL2pzL0pTTm9k ZUN1c3RvbS5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTE3OSw3ICsxNzksNyBAQAogCiAgICAgTm9k ZSogbm9kZSA9IG1faW1wbC5nZXQoKTsKICAgICBub2RlLT5tYXJrSlNFdmVudExpc3RlbmVycyht YXJrU3RhY2spOwotICAgIG5vZGUtPm1hcmtDYWNoZWROb2RlTGlzdHMobWFya1N0YWNrLCAqSGVh cDo6aGVhcCh0aGlzKS0+Z2xvYmFsRGF0YSgpKTsKKyAgICBub2RlLT5tYXJrQ2FjaGVkTm9kZUxp c3RzKG1hcmtTdGFjaywgdGhpcyk7CiAKICAgICAvLyBOb2RlcyBpbiB0aGUgZG9jdW1lbnQgYXJl IGtlcHQgYWxpdmUgYnkgSlNEb2N1bWVudDo6bWFyaywgc28sIGlmIHdlJ3JlIGluCiAgICAgLy8g dGhlIGRvY3VtZW50LCB3ZSBuZWVkIHRvIG1hcmsgdGhlIGRvY3VtZW50LCBidXQgd2UgZG9uJ3Qg bmVlZCB0byBleHBsaWNpdGx5Cg== 55404 2010-05-07 12:15:30 -0700 2010-05-07 17:05:43 -0700 Next patch: faster emptyValue patch.diff text/plain 2527 antonm SW5kZXg6IFdlYkNvcmUvZG9tL05vZGUuaAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBXZWJDb3JlL2RvbS9Ob2Rl LmgJKHJldmlzaW9uIDU4ODI2KQorKysgV2ViQ29yZS9kb20vTm9kZS5oCSh3b3JraW5nIGNvcHkp CkBAIC0zNiw2ICszNiw3IEBACiBuYW1lc3BhY2UgSlNDIHsKIAogICAgIGNsYXNzIEpTR2xvYmFs RGF0YTsKKyAgICBjbGFzcyBKU09iamVjdDsKICAgICBjbGFzcyBNYXJrU3RhY2s7CiAKIH0KQEAg LTU4MCwxMyArNTgxLDEzIEBACiAgICAgdmlydHVhbCBFdmVudFRhcmdldERhdGEqIGVuc3VyZUV2 ZW50VGFyZ2V0RGF0YSgpOwogCiAjaWYgVVNFKEpTQykKLSAgICB2b2lkIG1hcmtDYWNoZWROb2Rl TGlzdHMoSlNDOjpNYXJrU3RhY2smIG1hcmtTdGFjaywgSlNDOjpKU0dsb2JhbERhdGEmIGdsb2Jh bERhdGEpCisgICAgdm9pZCBtYXJrQ2FjaGVkTm9kZUxpc3RzKEpTQzo6TWFya1N0YWNrJiBtYXJr U3RhY2ssIEpTQzo6SlNPYmplY3QqIHdyYXBwZXIpCiAgICAgewogICAgICAgICAvLyBOb2RlTGlz dHMgbWF5IGJlIHByZXNlbnQuICBJZiBzbywgdGhleSBuZWVkIHRvIGJlIG1hcmtlZC4KICAgICAg ICAgaWYgKCFoYXNSYXJlRGF0YSgpKQogICAgICAgICAgICAgcmV0dXJuOwogCi0gICAgICAgIG1h cmtDYWNoZWROb2RlTGlzdHNTbG93KG1hcmtTdGFjaywgZ2xvYmFsRGF0YSk7CisgICAgICAgIG1h cmtDYWNoZWROb2RlTGlzdHNTbG93KG1hcmtTdGFjaywgd3JhcHBlcik7CiAgICAgfQogI2VuZGlm CiAKQEAgLTYxMiw3ICs2MTMsNyBAQAogCiBwcml2YXRlOgogI2lmIFVTRShKU0MpCi0gICAgdm9p ZCBtYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtTdGFjayYsIEpTQzo6SlNHbG9iYWxE YXRhJik7CisgICAgdm9pZCBtYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtTdGFjayYs IEpTQzo6SlNPYmplY3QqKTsKICNlbmRpZgogCiAgICAgc3RhdGljIGJvb2wgaW5pdGlhbFJlZkNv dW50KENvbnN0cnVjdGlvblR5cGUpOwpJbmRleDogV2ViQ29yZS9kb20vTm9kZS5jcHAKPT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PQotLS0gV2ViQ29yZS9kb20vTm9kZS5jcHAJKHJldmlzaW9uIDU4ODI2KQorKysgV2ViQ29y ZS9kb20vTm9kZS5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTI1NjQsMTIgKzI1NjQsMTMgQEAKICAg ICAgICAgbWFya0RPTU9iamVjdFdyYXBwZXIobWFya1N0YWNrLCBnbG9iYWxEYXRhLCBpdC0+c2Vj b25kKTsKIH0KIAotdm9pZCBOb2RlOjptYXJrQ2FjaGVkTm9kZUxpc3RzU2xvdyhKU0M6Ok1hcmtT dGFjayYgbWFya1N0YWNrLCBKU0M6OkpTR2xvYmFsRGF0YSYgZ2xvYmFsRGF0YSkKK3ZvaWQgTm9k ZTo6bWFya0NhY2hlZE5vZGVMaXN0c1Nsb3coSlNDOjpNYXJrU3RhY2smIG1hcmtTdGFjaywgSlND OjpKU09iamVjdCogd3JhcHBlcikKIHsKICAgICBOb2RlTGlzdHNOb2RlRGF0YSogbm9kZUxpc3Rz ID0gcmFyZURhdGEoKS0+bm9kZUxpc3RzKCk7CiAgICAgaWYgKCFub2RlTGlzdHMpCiAgICAgICAg IHJldHVybjsKIAorICAgIEpTQzo6SlNHbG9iYWxEYXRhJiBnbG9iYWxEYXRhID0gKkpTQzo6SGVh cDo6aGVhcCh3cmFwcGVyKS0+Z2xvYmFsRGF0YSgpOwogICAgIG1hcmtOb2RlTGlzdHMobm9kZUxp c3RzLT5tX2NsYXNzTm9kZUxpc3RDYWNoZSwgbWFya1N0YWNrLCBnbG9iYWxEYXRhKTsKICAgICBt YXJrTm9kZUxpc3RzKG5vZGVMaXN0cy0+bV9uYW1lTm9kZUxpc3RDYWNoZSwgbWFya1N0YWNrLCBn bG9iYWxEYXRhKTsKICAgICBtYXJrTm9kZUxpc3RzKG5vZGVMaXN0cy0+bV90YWdOb2RlTGlzdENh Y2hlLCBtYXJrU3RhY2ssIGdsb2JhbERhdGEpOwpJbmRleDogV2ViQ29yZS9iaW5kaW5ncy9qcy9K U05vZGVDdXN0b20uY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFdlYkNvcmUvYmluZGluZ3MvanMvSlNOb2Rl Q3VzdG9tLmNwcAkocmV2aXNpb24gNTg4MjYpCisrKyBXZWJDb3JlL2JpbmRpbmdzL2pzL0pTTm9k ZUN1c3RvbS5jcHAJKHdvcmtpbmcgY29weSkKQEAgLTE3OSw3ICsxNzksNyBAQAogCiAgICAgTm9k ZSogbm9kZSA9IG1faW1wbC5nZXQoKTsKICAgICBub2RlLT5tYXJrSlNFdmVudExpc3RlbmVycyht YXJrU3RhY2spOwotICAgIG5vZGUtPm1hcmtDYWNoZWROb2RlTGlzdHMobWFya1N0YWNrLCAqSGVh cDo6aGVhcCh0aGlzKS0+Z2xvYmFsRGF0YSgpKTsKKyAgICBub2RlLT5tYXJrQ2FjaGVkTm9kZUxp c3RzKG1hcmtTdGFjaywgdGhpcyk7CiAKICAgICAvLyBOb2RlcyBpbiB0aGUgZG9jdW1lbnQgYXJl IGtlcHQgYWxpdmUgYnkgSlNEb2N1bWVudDo6bWFyaywgc28sIGlmIHdlJ3JlIGluCiAgICAgLy8g dGhlIGRvY3VtZW50LCB3ZSBuZWVkIHRvIG1hcmsgdGhlIGRvY3VtZW50LCBidXQgd2UgZG9uJ3Qg bmVlZCB0byBleHBsaWNpdGx5Cg== 55408 2010-05-07 12:37:37 -0700 2010-05-07 12:37:37 -0700 My emptyValue patch patchFast.diff text/plain 1844 sam SW5kZXg6IFdlYkNvcmUvZG9tL05vZGUuY3BwCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFdlYkNvcmUvZG9tL05v ZGUuY3BwCShyZXZpc2lvbiA1ODkwNCkKKysrIFdlYkNvcmUvZG9tL05vZGUuY3BwCSh3b3JraW5n IGNvcHkpCkBAIC0xMDEwLDggKzEwMTAsOCBAQCB2b2lkIE5vZGU6OnJlbW92ZUNhY2hlZFRhZ05v ZGVMaXN0KFRhZ05vCiAgICAgQVNTRVJUX1VOVVNFRChsaXN0LCBsaXN0LT5oYXNPd25DYWNoZXMo KSk7CiAKICAgICBOb2RlTGlzdHNOb2RlRGF0YSogZGF0YSA9IHJhcmVEYXRhKCktPm5vZGVMaXN0 cygpOwotICAgIEFTU0VSVF9VTlVTRUQobGlzdCwgbGlzdCA9PSBkYXRhLT5tX3RhZ05vZGVMaXN0 Q2FjaGUuZ2V0KG5hbWUpKTsKLSAgICBkYXRhLT5tX3RhZ05vZGVMaXN0Q2FjaGUucmVtb3ZlKG5h bWUpOworICAgIEFTU0VSVF9VTlVTRUQobGlzdCwgbGlzdCA9PSBkYXRhLT5tX3RhZ05vZGVMaXN0 Q2FjaGUuZ2V0KG5hbWUuaW1wbCgpKSk7CisgICAgZGF0YS0+bV90YWdOb2RlTGlzdENhY2hlLnJl bW92ZShuYW1lLmltcGwoKSk7CiB9CiAKIE5vZGUgKk5vZGU6OnRyYXZlcnNlTmV4dE5vZGUoY29u c3QgTm9kZSAqc3RheVdpdGhpbikgY29uc3QKQEAgLTE2NDQsNyArMTY0NCw3IEBAIFBhc3NSZWZQ dHI8Tm9kZUxpc3Q+IE5vZGU6OmdldEVsZW1lbnRzQnkKICAgICAKICAgICBBdG9taWNTdHJpbmcg bG9jYWxOYW1lQXRvbSA9IG5hbWU7CiAgICAgICAgIAotICAgIHBhaXI8Tm9kZUxpc3RzTm9kZURh dGE6OlRhZ05vZGVMaXN0Q2FjaGU6Oml0ZXJhdG9yLCBib29sPiByZXN1bHQgPSBkYXRhLT5ub2Rl TGlzdHMoKS0+bV90YWdOb2RlTGlzdENhY2hlLmFkZChRdWFsaWZpZWROYW1lKG51bGxBdG9tLCBs b2NhbE5hbWVBdG9tLCBuYW1lc3BhY2VVUkkpLCAwKTsKKyAgICBwYWlyPE5vZGVMaXN0c05vZGVE YXRhOjpUYWdOb2RlTGlzdENhY2hlOjppdGVyYXRvciwgYm9vbD4gcmVzdWx0ID0gZGF0YS0+bm9k ZUxpc3RzKCktPm1fdGFnTm9kZUxpc3RDYWNoZS5hZGQoUXVhbGlmaWVkTmFtZShudWxsQXRvbSwg bG9jYWxOYW1lQXRvbSwgbmFtZXNwYWNlVVJJKS5pbXBsKCksIDApOwogICAgIGlmICghcmVzdWx0 LnNlY29uZCkKICAgICAgICAgcmV0dXJuIFBhc3NSZWZQdHI8VGFnTm9kZUxpc3Q+KHJlc3VsdC5m aXJzdC0+c2Vjb25kKTsKICAgICAKSW5kZXg6IFdlYkNvcmUvZG9tL05vZGVSYXJlRGF0YS5oCj09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT0KLS0tIFdlYkNvcmUvZG9tL05vZGVSYXJlRGF0YS5oCShyZXZpc2lvbiA1ODkwNCkK KysrIFdlYkNvcmUvZG9tL05vZGVSYXJlRGF0YS5oCSh3b3JraW5nIGNvcHkpCkBAIC00OCw3ICs0 OCw3IEBAIHN0cnVjdCBOb2RlTGlzdHNOb2RlRGF0YSA6IE5vbmNvcHlhYmxlIHsKICAgICB0eXBl ZGVmIEhhc2hNYXA8U3RyaW5nLCBOYW1lTm9kZUxpc3QqPiBOYW1lTm9kZUxpc3RDYWNoZTsKICAg ICBOYW1lTm9kZUxpc3RDYWNoZSBtX25hbWVOb2RlTGlzdENhY2hlOwogICAgIAotICAgIHR5cGVk ZWYgSGFzaE1hcDxRdWFsaWZpZWROYW1lLCBUYWdOb2RlTGlzdCo+IFRhZ05vZGVMaXN0Q2FjaGU7 CisgICAgdHlwZWRlZiBIYXNoTWFwPFJlZlB0cjxRdWFsaWZpZWROYW1lOjpRdWFsaWZpZWROYW1l SW1wbD4sIFRhZ05vZGVMaXN0Kj4gVGFnTm9kZUxpc3RDYWNoZTsKICAgICBUYWdOb2RlTGlzdENh Y2hlIG1fdGFnTm9kZUxpc3RDYWNoZTsKIAogICAgIHN0YXRpYyBQYXNzT3duUHRyPE5vZGVMaXN0 c05vZGVEYXRhPiBjcmVhdGUoKQo= 55442 2010-05-07 17:47:36 -0700 2010-05-07 18:01:15 -0700 Fix dromaeo.diff text/plain 4221 sam SW5kZXg6IFdlYkNvcmUvQ2hhbmdlTG9nCj09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT0KLS0tIFdlYkNvcmUvQ2hhbmdlTG9n CShyZXZpc2lvbiA1ODk4NikKKysrIFdlYkNvcmUvQ2hhbmdlTG9nCSh3b3JraW5nIGNvcHkpCkBA IC0xLDMgKzEsMjUgQEAKKzIwMTAtMDUtMDcgIFNhbSBXZWluaWcgIDxzYW1Ad2Via2l0Lm9yZz4K KworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKworICAgICAgICBGaXggZm9y IGh0dHBzOi8vYnVncy53ZWJraXQub3JnL3Nob3dfYnVnLmNnaT9pZD0zODU1NworICAgICAgICBy NTg1MjYgaW50cm9kdWNlZCBhIH4zMCUgcmVncmVzc2lvbiBvbiBEcm9tYWVvIEpTIGxpYgorCisg ICAgICAgIFRoaXMgZml4IGRvZXMgdHdvIHRoaW5ncy4KKyAgICAgICAgLSBEb24ndCB1c2UgUXVh bGlmaWVkTmFtZSBhcyB0aGUga2V5IHRvIGEgSGFzaE1hcCwgdXNlIGEKKyAgICAgICAgICBSZWZQ dHI8UXVhbGlmaWVkTmFtZT4gaW5zdGVhZC4gIFdlIHNob3VsZCByZW1vdmUgdGhlIEhhc2hUcmFp dHMgZm9yCisgICAgICAgICAgUXVhbGlmaWVkTmFtZSBhbmQgdGhhdCB3aWxsIGhhcHBlbiBpbiBh IGZvbGxvdyB1cCBwYXRjaC4KKyAgICAgICAgLSBPbmx5IG1hcmsgY2FjaGVkIE5vZGVMaXN0cyBv biBEb2N1bWVudHMgaW5zdGVhZCBvZiBhbGwgTm9kZXMuCisKKyAgICAgICAgKiBiaW5kaW5ncy9q cy9KU0RvY3VtZW50Q3VzdG9tLmNwcDoKKyAgICAgICAgKFdlYkNvcmU6OkpTRG9jdW1lbnQ6Om1h cmtDaGlsZHJlbik6CisgICAgICAgICogYmluZGluZ3MvanMvSlNOb2RlQ3VzdG9tLmNwcDoKKyAg ICAgICAgKFdlYkNvcmU6OkpTTm9kZTo6bWFya0NoaWxkcmVuKToKKyAgICAgICAgKiBkb20vTm9k ZS5jcHA6CisgICAgICAgIChXZWJDb3JlOjpOb2RlOjpyZW1vdmVDYWNoZWRUYWdOb2RlTGlzdCk6 CisgICAgICAgIChXZWJDb3JlOjpOb2RlOjpnZXRFbGVtZW50c0J5VGFnTmFtZU5TKToKKyAgICAg ICAgKiBkb20vTm9kZVJhcmVEYXRhLmg6CisKIDIwMTAtMDUtMDcgIEJldGggRGFraW4gIDxiZGFr aW5AYXBwbGUuY29tPgogCiAgICAgICAgIFJldmlld2VkIGJ5IFNpbW9uIEZyYXNlci4KSW5kZXg6 IFdlYkNvcmUvYmluZGluZ3MvanMvSlNEb2N1bWVudEN1c3RvbS5jcHAKPT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQotLS0g V2ViQ29yZS9iaW5kaW5ncy9qcy9KU0RvY3VtZW50Q3VzdG9tLmNwcAkocmV2aXNpb24gNTg5MDQp CisrKyBXZWJDb3JlL2JpbmRpbmdzL2pzL0pTRG9jdW1lbnRDdXN0b20uY3BwCSh3b3JraW5nIGNv cHkpCkBAIC01Niw2ICs1Niw3IEBAIHZvaWQgSlNEb2N1bWVudDo6bWFya0NoaWxkcmVuKE1hcmtT dGFjayYKICAgICBtYXJrQWN0aXZlT2JqZWN0c0ZvckNvbnRleHQobWFya1N0YWNrLCBnbG9iYWxE YXRhLCBkb2N1bWVudCk7CiAgICAgbWFya0RPTU9iamVjdFdyYXBwZXIobWFya1N0YWNrLCBnbG9i YWxEYXRhLCBkb2N1bWVudC0+aW1wbGVtZW50YXRpb24oKSk7CiAgICAgbWFya0RPTU9iamVjdFdy YXBwZXIobWFya1N0YWNrLCBnbG9iYWxEYXRhLCBkb2N1bWVudC0+c3R5bGVTaGVldHMoKSk7Cisg ICAgZG9jdW1lbnQtPm1hcmtDYWNoZWROb2RlTGlzdHMobWFya1N0YWNrLCBnbG9iYWxEYXRhKTsK IH0KIAogSlNWYWx1ZSBKU0RvY3VtZW50Ojpsb2NhdGlvbihFeGVjU3RhdGUqIGV4ZWMpIGNvbnN0 CkluZGV4OiBXZWJDb3JlL2JpbmRpbmdzL2pzL0pTTm9kZUN1c3RvbS5jcHAKPT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PQot LS0gV2ViQ29yZS9iaW5kaW5ncy9qcy9KU05vZGVDdXN0b20uY3BwCShyZXZpc2lvbiA1ODkwNCkK KysrIFdlYkNvcmUvYmluZGluZ3MvanMvSlNOb2RlQ3VzdG9tLmNwcAkod29ya2luZyBjb3B5KQpA QCAtMTc5LDcgKzE3OSw2IEBAIHZvaWQgSlNOb2RlOjptYXJrQ2hpbGRyZW4oTWFya1N0YWNrJiBt YXIKIAogICAgIE5vZGUqIG5vZGUgPSBtX2ltcGwuZ2V0KCk7CiAgICAgbm9kZS0+bWFya0pTRXZl bnRMaXN0ZW5lcnMobWFya1N0YWNrKTsKLSAgICBub2RlLT5tYXJrQ2FjaGVkTm9kZUxpc3RzKG1h cmtTdGFjaywgKkhlYXA6OmhlYXAodGhpcyktPmdsb2JhbERhdGEoKSk7CiAKICAgICAvLyBOb2Rl cyBpbiB0aGUgZG9jdW1lbnQgYXJlIGtlcHQgYWxpdmUgYnkgSlNEb2N1bWVudDo6bWFyaywgc28s IGlmIHdlJ3JlIGluCiAgICAgLy8gdGhlIGRvY3VtZW50LCB3ZSBuZWVkIHRvIG1hcmsgdGhlIGRv Y3VtZW50LCBidXQgd2UgZG9uJ3QgbmVlZCB0byBleHBsaWNpdGx5CkluZGV4OiBXZWJDb3JlL2Rv bS9Ob2RlLmNwcAo9PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09Ci0tLSBXZWJDb3JlL2RvbS9Ob2RlLmNwcAkocmV2aXNpb24g NTg5MDQpCisrKyBXZWJDb3JlL2RvbS9Ob2RlLmNwcAkod29ya2luZyBjb3B5KQpAQCAtMTAxMCw4 ICsxMDEwLDggQEAgdm9pZCBOb2RlOjpyZW1vdmVDYWNoZWRUYWdOb2RlTGlzdChUYWdObwogICAg IEFTU0VSVF9VTlVTRUQobGlzdCwgbGlzdC0+aGFzT3duQ2FjaGVzKCkpOwogCiAgICAgTm9kZUxp c3RzTm9kZURhdGEqIGRhdGEgPSByYXJlRGF0YSgpLT5ub2RlTGlzdHMoKTsKLSAgICBBU1NFUlRf VU5VU0VEKGxpc3QsIGxpc3QgPT0gZGF0YS0+bV90YWdOb2RlTGlzdENhY2hlLmdldChuYW1lKSk7 Ci0gICAgZGF0YS0+bV90YWdOb2RlTGlzdENhY2hlLnJlbW92ZShuYW1lKTsKKyAgICBBU1NFUlRf VU5VU0VEKGxpc3QsIGxpc3QgPT0gZGF0YS0+bV90YWdOb2RlTGlzdENhY2hlLmdldChuYW1lLmlt cGwoKSkpOworICAgIGRhdGEtPm1fdGFnTm9kZUxpc3RDYWNoZS5yZW1vdmUobmFtZS5pbXBsKCkp OwogfQogCiBOb2RlICpOb2RlOjp0cmF2ZXJzZU5leHROb2RlKGNvbnN0IE5vZGUgKnN0YXlXaXRo aW4pIGNvbnN0CkBAIC0xNjQ0LDcgKzE2NDQsNyBAQCBQYXNzUmVmUHRyPE5vZGVMaXN0PiBOb2Rl OjpnZXRFbGVtZW50c0J5CiAgICAgCiAgICAgQXRvbWljU3RyaW5nIGxvY2FsTmFtZUF0b20gPSBu YW1lOwogICAgICAgICAKLSAgICBwYWlyPE5vZGVMaXN0c05vZGVEYXRhOjpUYWdOb2RlTGlzdENh Y2hlOjppdGVyYXRvciwgYm9vbD4gcmVzdWx0ID0gZGF0YS0+bm9kZUxpc3RzKCktPm1fdGFnTm9k ZUxpc3RDYWNoZS5hZGQoUXVhbGlmaWVkTmFtZShudWxsQXRvbSwgbG9jYWxOYW1lQXRvbSwgbmFt ZXNwYWNlVVJJKSwgMCk7CisgICAgcGFpcjxOb2RlTGlzdHNOb2RlRGF0YTo6VGFnTm9kZUxpc3RD YWNoZTo6aXRlcmF0b3IsIGJvb2w+IHJlc3VsdCA9IGRhdGEtPm5vZGVMaXN0cygpLT5tX3RhZ05v ZGVMaXN0Q2FjaGUuYWRkKFF1YWxpZmllZE5hbWUobnVsbEF0b20sIGxvY2FsTmFtZUF0b20sIG5h bWVzcGFjZVVSSSkuaW1wbCgpLCAwKTsKICAgICBpZiAoIXJlc3VsdC5zZWNvbmQpCiAgICAgICAg IHJldHVybiBQYXNzUmVmUHRyPFRhZ05vZGVMaXN0PihyZXN1bHQuZmlyc3QtPnNlY29uZCk7CiAg ICAgCkluZGV4OiBXZWJDb3JlL2RvbS9Ob2RlUmFyZURhdGEuaAo9PT09PT09PT09PT09PT09PT09 PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09PT09Ci0tLSBXZWJD b3JlL2RvbS9Ob2RlUmFyZURhdGEuaAkocmV2aXNpb24gNTg5MDQpCisrKyBXZWJDb3JlL2RvbS9O b2RlUmFyZURhdGEuaAkod29ya2luZyBjb3B5KQpAQCAtNDgsNyArNDgsNyBAQCBzdHJ1Y3QgTm9k ZUxpc3RzTm9kZURhdGEgOiBOb25jb3B5YWJsZSB7CiAgICAgdHlwZWRlZiBIYXNoTWFwPFN0cmlu ZywgTmFtZU5vZGVMaXN0Kj4gTmFtZU5vZGVMaXN0Q2FjaGU7CiAgICAgTmFtZU5vZGVMaXN0Q2Fj aGUgbV9uYW1lTm9kZUxpc3RDYWNoZTsKICAgICAKLSAgICB0eXBlZGVmIEhhc2hNYXA8UXVhbGlm aWVkTmFtZSwgVGFnTm9kZUxpc3QqPiBUYWdOb2RlTGlzdENhY2hlOworICAgIHR5cGVkZWYgSGFz aE1hcDxSZWZQdHI8UXVhbGlmaWVkTmFtZTo6UXVhbGlmaWVkTmFtZUltcGw+LCBUYWdOb2RlTGlz dCo+IFRhZ05vZGVMaXN0Q2FjaGU7CiAgICAgVGFnTm9kZUxpc3RDYWNoZSBtX3RhZ05vZGVMaXN0 Q2FjaGU7CiAKICAgICBzdGF0aWMgUGFzc093blB0cjxOb2RlTGlzdHNOb2RlRGF0YT4gY3JlYXRl KCkK