246471 2022-10-13 13:58:54 -0700 [GTK] Accessibility API test fails with AT-SPI failing to read sockets not exposed to the SDK 2022-10-14 04:50:30 -0700 1 1 1 Unclassified WebKit Accessibility WebKit Nightly Build Unspecified Unspecified NEW InRadar P2 Normal --- 1 lmoura webkit-unassigned andresg_22 bugs-noreply cgarcia mcatanzaro webkit-bug-importer oldest_to_newest 1905420 0 lmoura 2022-10-13 13:58:54 -0700 Tests affected: /WebKitGTK/TestWebKitAccessibility /webkit/WebKitAccessibility/accessible/basic-hierarchy /webkit/WebKitAccessibility/accessible/children-changed /webkit/WebKitAccessibility/accessible/attributes Example log from EWS: https://ews-build.webkit.org/#/builders/34/builds/71271/steps/26/logs/stdio Sample output: TEST: /app/webkit/Tools/glib/../../WebKitBuild/Release/bin/TestWebKitAPI/WebKitGTK/TestWebKitAccessibility... /webkit/WebKitAccessibility/accessible/basic-hierarchy: FAIL GLib-GIO-DEBUG: _g_io_module_get_default: Found default implementation local (GLocalVfs) for ‘gio-vfs’ dbind-FATAL-WARNING: AT-SPI: Unable to open bus connection: Failed to connect to socket /run/user/104/at-spi2-LB05T1/socket: No such file or directory When finding the `AtSpiAccessible` for the test application[1], fatal warnings are ignored[2] while getting the child count. This works mostly fine to avoid issues with AT-SPI sockets not exposed to the SDK. But in some cases, the socket check is delayed and a fatal warning is issued as described above. Sometimes, happening even well after returning from the function, inside the actual test code. [1] https://github.com/WebKit/WebKit/blob/48f8da125ce496a5dbe81cf7b7e06476565b7128/Tools/TestWebKitAPI/Tests/WebKitGtk/TestWebKitAccessibility.cpp#L51 [2] https://github.com/WebKit/WebKit/blob/48f8da125ce496a5dbe81cf7b7e06476565b7128/Tools/TestWebKitAPI/Tests/WebKitGtk/TestWebKitAccessibility.cpp#L57 1905421 1 webkit-bug-importer 2022-10-13 13:59:07 -0700 <rdar://problem/101142072> 1905474 2 lmoura 2022-10-13 16:35:59 -0700 Gardened in https://commits.webkit.org/255511@main This seems to have flew under the radar due to infrastructure issues in the post-commit bot, which is currently not running this test. 1905585 3 mcatanzaro 2022-10-14 04:50:30 -0700 Hmm, it's similar to https://github.com/WebKit/WebKit/pull/5136, but the sandbox is not enabled for this test, so why doesn't it have full access to the host system?