You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Based on investigation by @stefanmb the issue is because the v8 scripts rely on the valid address range being less than the full 64 bits. Since AIX supports the full 64 bits and some must be in the upper range the script does not handle the addresses correctly, and ends up running out of memory.
@mhdawson I can't actually view the failure since I don't have permissions in Jenkins. However, assuming it's the tick-processor out-of-memory timeout, credit actually goes to @mtbrandy for figuring out the address range problem. I've done a proof of concept of V8 tooling with Big Integer support here: https://github.com/stefanmb/v8/tree/v8-tooling-bigint
I'll clean it up tomorrow and we can decide what to do with it.
@mhdawson I've updated my Big Integer branch linked above, at the moment it's an ugly hack but using it the tooling successfully parses the AIX PPC logs, and the unit tests from V8 ('make x64.check') pass as well. I'll wait on some more feedback from others to see if this is the direction we want to take in V8 itself.
It may not be possible, but anything that can be done to make the test more deterministic would be great. It's been the source of a fair bit of flakiness.
Activity
mhdawson commentedon Feb 4, 2016
Related to #2957 but I believe in a different test.
Will use this issue to track excluding for AIX until 2957 is resolved.
mhdawson commentedon Feb 4, 2016
Based on investigation by @stefanmb the issue is because the v8 scripts rely on the valid address range being less than the full 64 bits. Since AIX supports the full 64 bits and some must be in the upper range the script does not handle the addresses correctly, and ends up running out of memory.
stefanmb commentedon Feb 4, 2016
@mhdawson I can't actually view the failure since I don't have permissions in Jenkins. However, assuming it's the tick-processor out-of-memory timeout, credit actually goes to @mtbrandy for figuring out the address range problem. I've done a proof of concept of V8 tooling with Big Integer support here: https://github.com/stefanmb/v8/tree/v8-tooling-bigint
I'll clean it up tomorrow and we can decide what to do with it.
mhdawson commentedon Feb 4, 2016
Jenkins is temporarily locked down to collaborators you will be able to see it later on.
Trott commentedon Feb 4, 2016
/cc @matthewloring
mhdawson commentedon Feb 4, 2016
PR to temporarily disable on AIX until we get an v8 fix upstreamed so we can get CI green
#5093
stefanmb commentedon Feb 5, 2016
@mhdawson I've updated my Big Integer branch linked above, at the moment it's an ugly hack but using it the tooling successfully parses the AIX PPC logs, and the unit tests from V8 ('make x64.check') pass as well. I'll wait on some more feedback from others to see if this is the direction we want to take in V8 itself.
Trott commentedon Feb 6, 2016
It may not be possible, but anything that can be done to make the test more deterministic would be great. It's been the source of a fair bit of flakiness.
mhdawson commentedon Feb 10, 2016
excluded for now under #5093 and added ref to that PR in #2957 so that we remember to disable. Closing this issue.