Run lift time before and after #1128, Linux (digest check)
Commits
Plan (gpt-6.1-terra)
Compare Linux lift time and generated C using two lift-cache builds per revision
In plain words
On Linux, two lift-cache builds per revision will measure and compare lift-stage time, with no game run. The timed builds take about 80 minutes in total. The candidate fails if its lifted C differs byte for byte from the baseline or its median lift time is more than 15% slower.
Asked for
Time the lift on both sides with the lift cache, 2 timed builds each, and check the lifted C is the same as the baseline's. No game run is needed: this only tests the build.
Source: AI plan
Raw plan Download plan.json
{
"pass": {
"crash": false,
"same_lift": true,
"build_lift_s": {
"max_delta_pct": 15
}
},
"probe": [
{
"kind": "crash"
}
],
"matrix": {},
"schema": 5,
"target": [
{
"name": "linux",
"tags": [
"linux"
],
"min_reports": 1
}
],
"timing": {
"runs": 2,
"caches": [
"lift"
]
},
"summary": "Compare Linux lift time and generated C using two lift-cache builds per revision",
"baseline": {
"ref": "main"
},
"scenario": []
}
Targets
| Target | Reports | Status |
|---|---|---|
| linux | none yet | 0/1 pass, 1 running, 0 queued |
Reports
No reports yet. Each tester posts one when its job ends: the candidate's numbers against the baseline's.
Jobs
| Job | Target | State | Tester | Attempts | Note |
|---|---|---|---|---|---|
jb_2486d3b5c1af34d5 log | linux | runningbuild-base10% | jabbaxipil/Blaine | 2 | stopped by the tester: the app was closed |