Don't have to rerun every build when the only logfile parsing is messed up (as currently the case for ~150 GHC builds).
Somewhat better means for examining broken builds. Although in that case, feed-gipeda output should also provide enough information.
Cons:
Space! E.g. each GHC build produces 4-5 MB of log data, has >7000 relevant builds, so that's 35GB of log data. That's a lot, considering that the rackspace machine at perf-service.haskell.org only has 50GB of SSD.
In cases where this matters, it might be worthwhile to think of an alternative mechanism in the benchmark script. Like uploading log files to git repo and check that or something.
Another abstraction breaking log2csv option in the gipeda.yaml.
Benefits:
feed-gipedaoutput should also provide enough information.Cons:
log2csvoption in thegipeda.yaml.