Brett Okken

> we don't know whether this is an array of bytes or a string There is some ambiguity there, but /assuming/ an array of `bytea` seems reasonable enough in this...

> As an example, one can have `createArray(String typeName, Collection items, Function mapper)` and `U` is erased so it's not possible to construct typed array. That is not part of...

While it won't get called if the level is enabled, the lambda still needs to be created. So there is still expense for a lambda which capture context.

@rbygrave I /think/ it might be important that the lock objects are final. (It is certainly [important for Atomic*FIeldUpdaters](https://shipilev.net/blog/2015/faster-atomic-fu/#_final_fields_and_holders_identity), I think some of the jvm optimizations apply here also, but...

This is a blast from the past. Changes look good to me.

Asserts are disabled by default. What is the purpose of this change?

As I read [this](https://github.com/obriensp/iWorkFileFormat/blob/master/Docs/index.md#snappy-compression) description of iwa format, it is loosely the snappy framing format, but does not include required portions of the format. If it actually adhered to the...

Adjusting the frame size is about the only way to impact efficiency. Disabling the verification of checksums while decompressing would likely make it faster, but at the loss of being...

`SnappyFramedInputStream` is actually based on a `ReadableByteChannel`. A `FileInputStream` is converted to a `FileChannel`, which should be quite efficient. A `BufferedInputStream` would be wrapped with a `ReadableByteChannel`, which would effectively...

Did you compare going directly to a byte[] with the ascii bytes rather than using a StringBuilder?