Closed jerboaa closed 1 year ago
Copying from: https://github.com/Karm/mandrel-integration-tests/issues/102#issuecomment-1538167651
The regression was introduced with https://github.com/oracle/graal/pull/6504 and more specifically by https://github.com/oracle/graal/commit/b70c6402d6b37e74e4f5c9df4f83ecfafc84c82f
It's also worth noting that https://github.com/oracle/graal/commit/00eaa8ea09a001b61346cf6c0418b62dde08d29c makes -H:-OmitInlinedMethodDebugLineInfo
the default.
The above changes result in making the breakpoint appear in a different line (even with -H:+OmitInlinedMethodDebugLineInfo
). Further investigation is required to see if this is an upstream issue.
The regression is due to defaulting to not using SourceMapping
s when building the CompilationResultTree
. This choice was made in order to improve build times, but as expected it comes at the cost of less accurate generated code to source mapping.
Building the native image with -H:+DebugCodeInfoUseSourceMappings
brings back the old behavior.
This doesn't look like an upstream bug, but we probably want:
@Karm @jerboaa thoughts?
Building the native image with
-H:+DebugCodeInfoUseSourceMappings
brings back the old behavior.This doesn't look like an upstream bug, but we probably want:
1. to add this flag to Quarkus when creating debug builds.
That seems fine. We'll have to add to the debugtest for the non-quarkus app as well, though.
2. test both with and without the flag
Like test it as two variations of the debug info tests? Not sure if it's worth it.
Re-opening this until the quarkus fix is in.
debugSymbolsSmokeGDB fails with (both with Java 17 and Java 20):
Latest passing run was https://github.com/graalvm/mandrel/actions/runs/4878221407/jobs/8703969691 (4 days ago). The github image runner in both cases is Image: ubuntu-22.04 Version: 20230426.1 indicating this is not due to a change in the github runners (e.g. gdb version)