Clear resolveFrameFlags after each frame walk #2569
Conversation
Clear resolveFrameFlags in the interpreter stack walker after reporting the frame. This is consistent with the JIT stack walker and avoids leaving invalid flags set on entry to the JIT walker in very rare cases. Also fix a small error in the resumeable stack walker, which has no effect on the only current consumer of the feature. Fixes: eclipse-openj9#2472 Signed-off-by: Graham Chapman <[email protected]>
| resumeInterpreterWalk: | ||
| walkState->previousFrameFlags = walkState->frameFlags; | ||
| walkState->resolveFrameFlags = 0; |
There was a problem hiding this comment.
Is the intention that the resolveFrameFlags are only set if the current frame is a resolveFrame? And otherwise cleared?
Overall, I have to wonder why we bother with a separate resolveFrameFlags in that case.
There was a problem hiding this comment.
Agreed - a lot of this is historical cruft, but fixing that is for a different day.
resolveFrameFlags is set when the previously-walked frame is a resolve frame. Generally only applicable if the current frame is a JIT frame, but this seemed the safest and most "correct" way to fix it, since there clearly is one case where the resolve frame is not on top of a JIT frame.
There was a problem hiding this comment.
Thanks for confirming that I understood the fix :)
|
Jenkins test extended xlinux jdk10 |
Forgot the change in the original PR. Signed-off-by: Graham Chapman <[email protected]>
Clear resolveFrameFlags in the interpreter stack walker after reporting
the frame. This is consistent with the JIT stack walker and avoids
leaving invalid flags set on entry to the JIT walker in very rare cases.
Also fix a small error in the resumeable stack walker, which has no
effect on the only current consumer of the feature.
Fixes: #2472
Signed-off-by: Graham Chapman [email protected]