#general
- e
enough-analyst-54434
08/21/2015, 9:27 PMThanks for all the help on the release reviews and patience with my lack of RB-promptness as a reviewer these past 2 Medium weeks - release is live: https://pypi.python.org/pypi/pantsbuild.pants/0.0.44 - r
rough-minister-58256
08/21/2015, 9:27 PM🎉 - r
rough-minister-58256
08/21/2015, 9:27 PM0.0.44 (8/21/2025) - r
rough-minister-58256
08/21/2015, 9:27 PMwe are living in the future - r
rough-minister-58256
08/21/2015, 9:28 PMseems at 0.0.41 we jumped from 2015 -> 2025 for the year in the release notes - a
ancient-stone-50795
08/21/2015, 9:28 PM+1 - f
fast-megabyte-33579
08/21/2015, 11:36 PMhaha - b
bored-art-40741
08/22/2015, 7:27 PMSo we just upgraded to 0.0.43 internally and we're seeing some major regressions in artifact cache behavior - b
bored-art-40741
08/22/2015, 7:28 PMHere's one example (sorry I don't have a smaller repro) where the number of targets passed to
varies from run to run for no apparent reasonjava-execute - b
bored-art-40741
08/22/2015, 7:29 PM-.txt - b
bored-art-40741
08/22/2015, 7:29 PMNote the cache hits forjava-execute - b
bored-art-40741
08/22/2015, 7:30 PMI can't think of any way in which that kind of non-determinism is correct, or even how it's generated - b
bored-art-40741
08/22/2015, 7:30 PMIt also seems like our cache keys are getting scrambled (CI cached artifacts are not being hit by local compiles) - b
bored-art-40741
08/22/2015, 7:31 PMMy suspicion is that our use of
is a culprit here, but once again, I don't know how that would explain the non-determinismjava_sources - b
bored-art-40741
08/22/2015, 7:34 PMAlso worth noting we're upgrading from 0.0.34 - b
bored-art-40741
08/22/2015, 7:35 PMSo this could have been introduced anywhere between 0.0.34 and 0.0.43 - b
bored-art-40741
08/22/2015, 7:36 PMAha! - b
bored-art-40741
08/22/2015, 7:36 PMThe order of
andzinc-prepare
is also different in those two runsjava-prepare - b
bored-art-40741
08/22/2015, 7:36 PMI would not be surprised if that's related - b
bored-art-40741
08/22/2015, 7:44 PMDo we have docs or an example of how to switch away from jmake to use zinc for scala and java compilation? - b
bored-art-40741
08/22/2015, 7:45 PMI suspect that might bypass this problem, though this bug should be fixed as well - b
bored-art-40741
08/22/2015, 9:28 PMI have a much smaller repro now - b
bored-art-40741
08/22/2015, 9:28 PMMy best guess here is that
is non-deterministicTaskIdentityFingerprintStrategy - b
bored-art-40741
08/22/2015, 9:29 PMIt's certainly generating hashes from iterators that have no documented guarantee of order - b
bored-art-40741
08/22/2015, 10:07 PMAh, I was chasing my tail believing that this might be the result of different machine environments - b
bored-art-40741
08/22/2015, 10:07 PMIt looks like fingerprinting is also broken on subsequent runs on the same machine - b
bored-art-40741
08/22/2015, 10:33 PMAha! - b
bored-art-40741
08/22/2015, 10:33 PMIt is the fatality of missing-deps - b
bored-art-40741
08/22/2015, 10:34 PMWhich I would argue should not befingerprint=True - b
bored-art-40741
08/22/2015, 10:56 PMThat was indeed the problem