summaryrefslogtreecommitdiff
path: root/compiler/optimizing/instruction_builder.h
diff options
context:
space:
mode:
author Jiakai Zhang <jiakaiz@google.com> 2024-01-08 16:57:47 +0000
committer Treehugger Robot <android-test-infra-autosubmit@system.gserviceaccount.com> 2024-01-10 11:58:12 +0000
commite685ea811d95f54ccf6168397b840e125076d834 (patch)
tree0f2e7836b37d52bbeef836d860ec19773ed11b59 /compiler/optimizing/instruction_builder.h
parentcb5d0c9ab7fdd514173d71066e3e0ae106d01cfc (diff)
Avoid replacing framework.jar in odsign_e2e_tests.
In a test case, in order to simulate an OTA update and test odrefresh's behavior on the OTA, we do a hack: we replace a bootclasspath jar /system/framework/framework.jar by bind-mounting a test jar at this path, and then we umount in tearDown. Normally, this should not be disruptive because Zygote loads framework.jar during boot and keeps the FD, and apps fork from Zygote. However, there are commands such as `svc power reboot` that brings up a separate runtime, not forked from Zygote, and therefore needs to read framework.jar from disk. Such commands will crash if they hit the fake framework.jar. If the umount fails, the fake framework.jar will stay, and break all subsequent commands such as `svc power reboot`. In this CL, I'm changing the test case to replace framework-graphics.jar instead, to make it less disruptive when the umount fails. Bug: 318706268 Change-Id: I1ba76412484da07e3c6521a2f6bd40bc45c70a7e Test: ABTD (http://go/abtd-run/L98800030001127223)
Diffstat (limited to 'compiler/optimizing/instruction_builder.h')
0 files changed, 0 insertions, 0 deletions