fparser is still the main bottleneck when executing PSyclone, it takes 70%-80% of its execution time. There are some low code-impact modifications that can be explored without changing the main structure of the code. I have implemented some of those in a local branch and show promising results. Some of the ideas are:

• (towards 312) Improve performance by adding caching to the tokenizer #336 Use caching/memoisation in the tokenizer (because the same string is tokenized multipe times when recursing down a match chain)
• (towards 312) Remove imports and function declaration from hotpath (__new__ and match) #337 Remove the imports from the hot path (__new__ and match) . These are there because they are cyclical imports but other solutions can be found.
• Currently each object is instantiated twice: one with the reader object as a parameter and inside it another time with the string from the reader.get() as a parameter. This doubles the number of instantiations and methods calls in the hotpath and provably can be optimized significantly if merged.

Current results: