Masataro Asai

I wish there is a test case. (then I can reuse the same test for Trivia)

[From his recent comment](https://github.com/m2ym/optima/issues/110#issuecomment-77733403), I guess m2ym just does not have enough time to handle the merge requests (it even contains conflicts). [Trivia already have those patterns](https://github.com/guicho271828/trivia/wiki/Type-Based-Destructuring-Patterns#vector-vector-patterns), FYI.

@jorams I see. If I'm getting right, it requires creating a displaced array every time the matcher is invoked. It requires consing, and I don't really like it... Making a...

This is addressed in Trivia. https://github.com/guicho271828/trivia/blob/master/level2/arrays.lisp#L150

https://github.com/guicho271828/trivia/wiki/Benchmarking-Results

Optima implements its failure handling (move to the next clause if it exists, or signal an error) with `%fail` symbol macro and `(symbol-macrolet (%fail ...)) ...)` which shadows the global...

For any significant speedup I believe the dataset and the model should be large.

Not entirely unrelated, but when i tried AVM it complained that coerce was not found.

how about subclassing cffi-grovel::grovel-file (e.g. `cuda-header`) and make it have the `:around` method that handles the error? well, I'm not concerned with cl-cuda, but responding just on curiosity.

Adding /usr/local/cuda-8.0/targets/x86_64-linux/include to CPATH environment variable solves this.