#general
- b
bored-art-40741
09/08/2015, 7:30 PMThis is without name hashing - b
bored-art-40741
09/08/2015, 7:30 PMisolated only - w
witty-crayon-22786
09/08/2015, 7:30 PMwe haven't seen ooms with a 6G heap and 1200+ target graphs - b
bored-art-40741
09/08/2015, 7:31 PMThis is a target with a relatively large number of dependencies - b
bored-art-40741
09/08/2015, 7:31 PMBut not insane - w
witty-crayon-22786
09/08/2015, 7:31 PMand analysis cache = 1000 - w
witty-crayon-22786
09/08/2015, 7:31 PMhm. - b
bored-art-40741
09/08/2015, 7:31 PMI also have analysis cache set to 1k and the other two not set, letting zinc default them - w
witty-crayon-22786
09/08/2015, 7:31 PMditto. - w
witty-crayon-22786
09/08/2015, 7:32 PMhm, that's extremely interesting. - w
witty-crayon-22786
09/08/2015, 7:32 PMbecause i would expect the total amount memory usage to be strictly less than global - b
bored-art-40741
09/08/2015, 7:32 PMMaybe not though - b
bored-art-40741
09/08/2015, 7:33 PMSince we're doing things in 4x parallel - b
bored-art-40741
09/08/2015, 7:33 PMThat could theoretically mean 4x memory usage in pathological situations - w
witty-crayon-22786
09/08/2015, 7:33 PMhm. i think the working set in each compilation is the issue - w
witty-crayon-22786
09/08/2015, 7:33 PMbut yea, maybe not. - w
witty-crayon-22786
09/08/2015, 7:34 PM(if you assume global chunks usually included more than 4 targets) - b
bored-art-40741
09/08/2015, 7:34 PMBut a chunk isn't necessarily itself compiled in parallel - w
witty-crayon-22786
09/08/2015, 7:34 PMit's held in memory - w
witty-crayon-22786
09/08/2015, 7:34 PMat once - b
bored-art-40741
09/08/2015, 7:34 PMhm - w
witty-crayon-22786
09/08/2015, 7:34 PMand then the compiler runs all of the scalac phases sequentially - w
witty-crayon-22786
09/08/2015, 7:34 PMbut yea. not clearcut i guess. - b
bored-art-40741
09/08/2015, 7:35 PMAt this point we're running scalac phases in parallel though right? - b
bored-art-40741
09/08/2015, 7:35 PMSo the analysis isn't all held in memory, but whatever insanity scalac does is - w
witty-crayon-22786
09/08/2015, 7:35 PMyes - b
bored-art-40741
09/08/2015, 7:35 PMIncluding perhaps classloading - b
bored-art-40741
09/08/2015, 7:35 PMSpeaking strictly from human observation, isolated is doing a MUCH better job of burning my CPUs - b
bored-art-40741
09/08/2015, 7:36 PMSo it would not surprise me if it's just strictly more parallel than global - b
bored-art-40741
09/08/2015, 7:36 PMAnd we're seeing memory fallout from that too