[Trimming message, adding laszlo-dev] The lzo compile path will not have any of your recent "chunking" improvements, right? And we know that when running on a memory-constrained (and virtual) machine making the JVM heap bigger buys you nothing because you just trade GC for swap.
If it is easy, maybe you could move the "chunking" writer stuff 'up the tree' of writer classes so that the lzo-writing path gets the same benefit? I expect right now that the lzo path is still the 'one big string' issue. lzo's should not be affected by debug/backtrace except that when you turn on debug the code is not obfuscated or compressed and the #file/line info is preserved in the lzo. [... Long email thread about excessive lzo compile times with confidential information removed ...]
