| [#1019] | project: compiler | priority: medium | category: bug | |
|---|---|---|---|---|
| submitter | assigned to | status | date submitted | |
| Martin | Martin | fixed | 2007-03-24 17:52:29.0 | |
| subject | possible inefficiency in typer | |||
| code |
[locker] java.lang.StackOverflowError [locker] at java.util.LinkedHashMap.get(LinkedHashMap.java:273) [locker] at java.io.ExpiringCache.entryFor(ExpiringCache.java:83) [locker] at java.io.ExpiringCache.get(ExpiringCache.java:58) [locker] at java.io.Win32FileSystem.canonicalize(Win32FileSystem.java:376) [locker] at java.io.File.getCanonicalPath(File.java:531) [locker] at scala.tools.nsc.io.PlainFile.hashCode(PlainFile.scala:47) [locker] at scala.collection.mutable.FlatHashTable$class.elemHashCode(FlatHashTable.scala:137) [locker] at scala.collection.mutable.HashSet.elemHashCode(HashSet.scala:31) [locker] at scala.collection.mutable.FlatHashTable$class.addEntry(FlatHashTable.scala:61) [locker] at scala.collection.mutable.HashSet.addEntry(HashSet.scala:31) [locker] at scala.collection.mutable.HashSet.$plus$eq(HashSet.scala:35) [locker] at scala.tools.nsc.typechecker.Infer$Inferencer.checkAccessible(Infer.scala:343) |
|||
| what happened | When looking at the top of the stacktrace for the pattern matcher, I saw the code given above. checkAccessible is a hotspot in the compiler; it is run twice for every identifier use (once in typer another time in erasure). Now, it seems that every such call constructs the canonical path of a sourcefile to compute a hashcode to check \ whether the source file is in the depends set. Looks like a lot of work that's repeated all over again. I am not sure whether this is a performance bottleneck \ or not, but prudence dictates that we should optimize this! Probably a simple hash-consing for sourcefiles would\ be enough. |
|||
| what expected | ||||
| [back to overview] | ||||
| Martin edited on 2007-04-17 16:30:41.0 |