Tld parsers optimize access

[Anvil]

Hey there.

While looking at the code, it appeared there was a giant if/elif combo in parse_tld.py, which implied that to reach the "ve" tld parser you need to test all other combinations (linear complexity)

I've moved the parsers to their specific files, and replaced the nearly 150 lines of if/elif by a getattr, which should prove efficient enough to allow access to any RegexFOO class in constant time.

Also, I've made all RegexFOO respect the same interface, so that there's no need for a specific __init__ in each class.. which allowed the removal of another 200 lines. This should be easier to maintain.

There are a few other simplifications, in the branch to reduce the memory consumption.

Hope this helps.

[pogzyb]

Wow yeah this is so much better!

The if/else block was originally a big dictionary, but that was even worse for memory usage. Can't believe I never thought to leverage getattr here.

I will try to get this rolled out in 1.1.3 sometime this weekend - hopefully I don't mess it up again lol. I will probably refactor the tests and maybe reorganize the "parse" related files into a separate dir or something too.

Again, really appreciate your contributions to this library.

[Anvil]

Happy to help.

Do you want to keep the python 3.9 compat ? If so, I'll restore the Union i've dropped (Union[t1, t2] replaced by t1 | t2).

(Doing less imports from typing speeds up loading type:

• Dict, List, Set, etc. which are deprecated since 3.9 can be replaced by dict, list, set, and so on.
• Union which is not deprecated can be by replaced | syntax since 3.10
  )

[pogzyb]

Happy to help.

Do you want to keep the python 3.9 compat ? If so, I'll restore the Union i've dropped (Union[t1, t2] replaced by t1 | t2).

(Doing less imports from typing speeds up loading type:

* Dict, List, Set, etc. which are deprecated since 3.9 can be replaced by dict, list, set, and so on.

* Union which is _not_ deprecated can be by replaced `|` syntax since 3.10
  )

Yes thank you - Python 3.9 won't be EOL until October 2025, so I'd vote that we keep the old-style typing for the near future.

Other than that some of the parser tests are failing:

• RegexNU: This is due to my comment above about tld_specific_expressions not being present in the specific parser classes for RegexPT nor RegexNU.
• RegexGQ: Even though this class is inheriting from RegexTK, it's still carrying over it's empty expression dictionary. I added a possible fix below.

class RegexGQ(RegexTK):
    ...
    # tld_specific_expressions: ExpressionDict = {}  # comment out or remove, so parent class expressions are used

After these changes, we should be good to merge!