postpone timezone regex evaluation until first use - shaves off time from package import

[beda42]

This MR is related to issue #533 which is caused mainly by a time intensive parsing of regular expressions for timezone matching. This MR introduces a global object TzRegexCache and moves preparation of the regexps into it. The regexps are no longer parsed on startup, but during first use instead. On my machine it shaves off about 200 ms from the import time of dateparser thus reducing it to less than 20 % of the original import time.

[tobymao]

is this library still actively maintained? i ran into this issue and would like to fix this. my idea is instead to cache the regexes so that overall time gets faster. this solution will still incur regex parsing time, although it does speed up parsing.

but i only want to do a PR if it actually has a shot at making it in.

[Gallaecio]

We are reviewing PRs, and may even have a release “soon” (see the most recent PRs).

[tobymao]

We are reviewing PRs, and may even have a release “soon” (see the most recent PRs).

amazing! would you accept my idea? is there somewhere to discuss with maintainers this approach before i start to make iterations faster?

[Gallaecio]

You can either start a draft PR with little code and start the discussion there, or discuss it here.

[tobymao]

i've taken another stab at this @beda42 #1250

instead of lazy evaluation, i do caching so that we only pay the regex compile time cost once. it will break the cache if there's a version bump.