205069 2019-12-10 09:57:04 -0800 There should be no user-noticeable delay when closing a tab 2019-12-11 20:55:33 -0800 1 1 1 Unclassified WebKit WebKit2 WebKit Nightly Build Unspecified Unspecified RESOLVED FIXED InRadar P2 Normal --- 1 cdumez cdumez commit-queue ggaren mitz rniwa webkit-bug-importer oldest_to_newest 1597225 0 cdumez 2019-12-10 09:57:04 -0800 There should be no user-noticeable delay when closing a tab. To achieve this, use a timeout of 50ms for doing the IPC handshake with the WebContent process, instead of 500ms. 1597226 1 webkit-bug-importer 2019-12-10 09:57:50 -0800 <rdar://problem/57797494> 1597231 2 385276 cdumez 2019-12-10 10:00:50 -0800 Created attachment 385276 Patch 1597629 3 cdumez 2019-12-11 08:30:25 -0800 ping review? 1597781 4 385276 commit-queue 2019-12-11 14:05:28 -0800 Comment on attachment 385276 Patch Clearing flags on attachment: 385276 Committed r253395: <https://trac.webkit.org/changeset/253395> 1597782 5 commit-queue 2019-12-11 14:05:30 -0800 All reviewed patches have been landed. Closing bug. 1597909 6 mitz 2019-12-11 18:36:11 -0800 What’s the user-visible symptom when the UI process misses the 50ms deadline? Can you share any field data on how often it would happen? 1597912 7 cdumez 2019-12-11 19:06:26 -0800 (In reply to mitz from comment #6) > What’s the user-visible symptom when the UI process misses the 50ms > deadline? Can you share any field data on how often it would happen? It would not be the UIProcess missing the deadline but rather the WebProcess missing the deadline. From the UIProcess point of view, after 50ms we close the tab no matter what, even if the page’s script is still handling the beforeunload event. The idea is that we give the page’s script a chance to tell us if we should display an unload confirm prompt but not at the cost of a delaying the tab closing for more than 50ms. The script has 50ms to tell us to show the prompt or we close the tab without showing such prompts. I do not have data on how common this will happen in practice. However, there is error logging in release builds if it happens. Also, from my own personal experience unload confirm prompt are not very common on the Web. 1597926 8 mitz 2019-12-11 20:55:33 -0800 (In reply to Chris Dumez from comment #7) Thanks for the detailed response, Chris! > (In reply to mitz from comment #6) > > What’s the user-visible symptom when the UI process misses the 50ms > > deadline? Can you share any field data on how often it would happen? > > It would not be the UIProcess missing the deadline but rather the WebProcess > missing the deadline. Sorry, my silly mistake. > From the UIProcess point of view, after 50ms we close > the tab no matter what, even if the page’s script is still handling the > beforeunload event. > > The idea is that we give the page’s script a chance to tell us if we should > display an unload confirm prompt but not at the cost of a delaying the tab > closing for more than 50ms. The script has 50ms to tell us to show the > prompt or we close the tab without showing such prompts. Makes sense. > I do not have data on how common this will happen in practice. However, > there is error logging in release builds if it happens. Also, from my own > personal experience unload confirm prompt are not very common on the Web. I also rarely get prompted on close (even with the current 500ms delay). 385276 2019-12-10 10:00:50 -0800 2019-12-11 14:05:28 -0800 Patch bug-205069-20191210100050.patch text/plain 1608 cdumez U3VidmVyc2lvbiBSZXZpc2lvbjogMjUzMzI4CmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L0No YW5nZUxvZyBiL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCmluZGV4IDk4YmIwZDRmYzhjY2ZiZDIw Y2I3MmY2MWUxZmI0ZjUzYmZhZmQ3OWEuLmY2MWZiMTlkMmRjODA3NDY1ZjUxOWM2NWY1MDhjNzdi YTQ0MGJhMTEgMTAwNjQ0Ci0tLSBhL1NvdXJjZS9XZWJLaXQvQ2hhbmdlTG9nCisrKyBiL1NvdXJj ZS9XZWJLaXQvQ2hhbmdlTG9nCkBAIC0xLDMgKzEsMTYgQEAKKzIwMTktMTItMTAgIENocmlzIER1 bWV6ICA8Y2R1bWV6QGFwcGxlLmNvbT4KKworICAgICAgICBUaGVyZSBzaG91bGQgYmUgbm8gdXNl ci1ub3RpY2VhYmxlIGRlbGF5IHdoZW4gY2xvc2luZyBhIHRhYgorICAgICAgICBodHRwczovL2J1 Z3Mud2Via2l0Lm9yZy9zaG93X2J1Zy5jZ2k/aWQ9MjA1MDY5CisgICAgICAgIDxyZGFyOi8vcHJv YmxlbS81Nzc5NzQ5ND4KKworICAgICAgICBSZXZpZXdlZCBieSBOT0JPRFkgKE9PUFMhKS4KKwor ICAgICAgICBUaGVyZSBzaG91bGQgYmUgbm8gdXNlci1ub3RpY2VhYmxlIGRlbGF5IHdoZW4gY2xv c2luZyBhIHRhYi4gVG8gYWNoaWV2ZSB0aGlzLCB1c2UgYSB0aW1lb3V0IG9mIDUwbXMgZm9yIGRv aW5nIHRoZQorICAgICAgICBJUEMgaGFuZHNoYWtlIHdpdGggdGhlIFdlYkNvbnRlbnQgcHJvY2Vz cywgaW5zdGVhZCBvZiA1MDBtcy4KKworICAgICAgICAqIFVJUHJvY2Vzcy9XZWJQYWdlUHJveHku Y3BwOgorCiAyMDE5LTEyLTEwICBDaHJpcyBEdW1leiAgPGNkdW1lekBhcHBsZS5jb20+CiAKICAg ICAgICAgUmVncmVzc2lvbihyMjUzMjc3KSBBU1NFUlRJT04gRkFJTEVEOiAhcGFyYW1ldGVycy5y ZXF1ZXN0Lmh0dHBCb2R5KCkgaW4gTmV0d29ya0Nvbm5lY3Rpb25Ub1dlYlByb2Nlc3M6OnByZWNv bm5lY3RUbygpCmRpZmYgLS1naXQgYS9Tb3VyY2UvV2ViS2l0L1VJUHJvY2Vzcy9XZWJQYWdlUHJv eHkuY3BwIGIvU291cmNlL1dlYktpdC9VSVByb2Nlc3MvV2ViUGFnZVByb3h5LmNwcAppbmRleCA5 NjM4YTAzNzIyZDA5MjliZDFlM2JjNzFjNWI2MzdmNTNhMTZhMjhjLi5iYzE5YjE2ZjRkMDEzNWRh MTQ3YTFhOTA4OTUzMzEzODg4YTMwMDgzIDEwMDY0NAotLS0gYS9Tb3VyY2UvV2ViS2l0L1VJUHJv Y2Vzcy9XZWJQYWdlUHJveHkuY3BwCisrKyBiL1NvdXJjZS9XZWJLaXQvVUlQcm9jZXNzL1dlYlBh Z2VQcm94eS5jcHAKQEAgLTI3Niw3ICsyNzYsNyBAQCBzdGF0aWMgY29uc3QgdW5zaWduZWQgd2hl ZWxFdmVudFF1ZXVlU2l6ZVRocmVzaG9sZCA9IDEwOwogc3RhdGljIGNvbnN0IFNlY29uZHMgcmVz ZXRSZWNlbnRDcmFzaENvdW50RGVsYXkgPSAzMF9zOwogc3RhdGljIHVuc2lnbmVkIG1heGltdW1X ZWJQcm9jZXNzUmVsYXVuY2hBdHRlbXB0cyA9IDE7CiBzdGF0aWMgY29uc3QgU2Vjb25kcyBhdWRp YmxlQWN0aXZpdHlDbGVhckRlbGF5ID0gMTBfczsKLXN0YXRpYyBjb25zdCBTZWNvbmRzIHRyeUNs b3NlVGltZW91dERlbGF5ID0gNTAwX21zOworc3RhdGljIGNvbnN0IFNlY29uZHMgdHJ5Q2xvc2VU aW1lb3V0RGVsYXkgPSA1MF9tczsKIAogbmFtZXNwYWNlIFdlYktpdCB7CiB1c2luZyBuYW1lc3Bh Y2UgV2ViQ29yZTsK