See bug #641025 comment 124:

We also need to mark whenever we look up a filename in the hashtable, since it's effectively a weak pointer.

(In reply to Bill McCloskey (:billm) from comment #1) > We also need to mark whenever we look up a filename in the hashtable, since > it's effectively a weak pointer. My plan is to mark newly incremental filename entries under the incremental GC. The rest of cases should be covered by normal marking as the filename is never overwritten. But eventually I want to either remove them or at least turn them into atoms. My initial measurements shows that filename strings are atoms in about 10-15% cases so turning mostly ascii string into UTF16 char array + JSString header should not bloat memory match...

Comment on attachment 600089 [details] [diff] [review] v1 Thanks.