- Until now the source code was being parsed during every fitness computation. Now the parsed program is reused and only the optimisation steps are applied each time.
- Giving the function ownership of an object and then receiving back another object is better than letting it implicitly modify data passed in an argument.
- I need some sample .yul files for testing but I see that existing tests generally have source code hard-coded in them rather than in standalone .yul files. There are lots of .yul files but they seem to be automatically processed by a special test case rather loaded ad-hoc by manually created tests.
- Program and Population required a file name until now. I'm making them accept loaded source code to be able to give them data hard-coded in a test.
- This now displays internal step names rather than human-readable ones but the internal ones are readable enough and it's not something worth creating another map.
- Options in the banner are now aligned in columns and thus easier to read.
- Abbreviations match those used in yulopti.
- I considered using boost::bimap but I think it would be an overkill. Two simple maps are good enough in a situation where data is constant and there isn't much of it.
- I could also use InvertibleMap from libsolutil but I don't even need any of its methods since my map is a constant. I also don't need the inverted map to store sets because my values are unique.
- The reverseMap() is generic enough to be moved to some global file with utilities but I don't sure if it's going to actually be useful to others in practice.
- unique_ptr is more flexible and generally recommended for factory methods. It gets automatically converted to shared_ptr if necessary. Returning shared_ptr, on the other hand, forces the caller to use shared_ptr because a conversion to unique_ptr is not possible.
1) `Popen.communicate()` expects `bytes` (a raw, binary string) if `stdout`/`stderr` are open in binary mode but is given output from `json.loads()` which is str (an abstract unicode string). Encoding the `str` object into `bytes` using UTF-8 encoding fixes that.
2) `REPORT_FILE` gets opened in binary mode which means that functions like `write()` expect `bytes`. We're giving them `str` which results in an error. Changed mode to text solves the problem.
cameel