During the externalization process of the lambda expressions, at the end of the executable model generation, we do a pretty print of the class materializing the lambda, only to make an hashing of the whole source code.

This mechanism has been introduced with this commit and the corresponding pull request contains a detailed explanation of why we did so.

This solution works, but making a pretty print of the source code only to extract its hashing looks wasteful and in our benchmarks takes around 20% of the whole executable model generation time. It is required to investigate if we could achieve the same with a less expensive implementation.