Perf increase

[freedmand]

Patch in changes for increasing the performance of the FEC file parser.

• refactored date parsing to not actually localize and parse the times (always in same YYYYMMDD format) (⚡⚡)
• refactored the way cache keys are computed to be faster (⚡)
• initialized the character chr(0x1c) outside of the function to save a tiny slice of time (🙂)
• refactored out slow startsWith calls with string indexing and character comparisons (⚡⚡)
• found a completely redundant function call that was taking time for each line (⚡⚡)
• refactored nones to be a set rather than a list for faster membership checking (⚡)

Ensured the unit tests still pass. To enable faster date parsing, include {'simple_conversion': True} in the options to bypass int conversion and convert datetimes to strings that look like YYYY-MM-DD (from YYYYMMDD), which speeds up conversion significantly.

[freedmand]

Not quite ready yet — closing for now

[esonderegger]

Hi @freedmand ! Most of these performance changes look like great ideas - thanks for working on this!

I'm totally happy to leave this open while it's still in development. Please let me know when it's ready for me to take a look.

I'm intrigued by the idea of there being a middle ground between the as_string option, which does no type conversion, and converting everything to native python types (which I think should remain the default). I'm curious about your use case - I think a discussion about this could yield improvements to how options are handled.