Orson Peters

@JulianCologne I think you misunderstand. `length=4` for integers centered around 30 would indeed be `[28, 29, 30, 31]`, but `length=5` would be `[28, 29, 30, 31, 32]` with `closed="left"` as...

@kszlim As long as you create a column with a timestamp associated for each value as per your clock, `rolling_interval` should do that.

@kszlim Your example would be computed by `pl.col("a").sum().rolling_interval("b", 3)`. `rolling_interval` would not require anything time-based, just a weakly monotonically increasing column of a type that you can add/subtract.

The implicit panics of indexing `v[i]` are fine in my opinion. Feel free to send a PR that fixes the comments that incorrectly state they check if `idx > self.len()`...

I'm not completely convinced that returning -1 when the match is not found is superior to returning null. The distinguishing argument is kind of moot since you can just use...

@alexander-beedie Another argument in favour of null is if you try to compute any statistics on the found index, e.g. 'average text length before the first link' with `pl.col(c).str.find("https?://").mean()`. Including...

Even more disastrous is the minimum, e.g. trying to find the webpage with the shortest header: `pl.col("html").str.find("").arg_min()`.

I think it's easier/faster if we modify `split_to_struct` instead of every single caller of it. That way we also don't have to do a filter over every single character in...

@haocheng6 What you should do is when `by == ""` you should use `s.splitn(n + 1, "").skip(1)` instead of `op` which should automatically handle all cases.

@haocheng6 Thanks for the PR!