Simplified evaluation model

[Bubobubobubobubo]

Topos is evaluating too much code, too often. It is easy to imagine an evaluation model that doesn't repeat the costly operation of interpreting various files super often but just one big file. Here are some tracks to follow:

• merge everything into one big file with bits evaluated conditionnally
• mark a part of that file for evaluation if a script is called. Unmark that part once the evaluation is done.

It should be possible to greatly limit the number of evaluations, thus reducing the performance cost of simply running code.