Improve performance of Group-Object #7409
Description
As of pwsh 6.1, preview-4, we have quadratic performance when the number of unique values -> n.
By gathering the input first, sorting it, and then only each new value to the last group, we can come much closer to n * log(n) instead of n * n.
I have a prototype with the following perf measurements:
| Count | Unique | OldImpl | newImpl | Speedup | Command |
|---|---|---|---|---|---|
| 10689 | 8220 | 00:00:06.81 | 00:00:00.23 | 29,1 | $allItemsInPowerShellSrcTree | group {[io.path]::GetFileName($_)} |
| 1690765 | 3761 | 00:02:30.34 | 00:00:22.32 | 6,7 | $u | group |
where $u is a dataset of string values out of which 3700 is unique.
The only downside I have seen is that the both the order of the output objects are different, and so is the order within the groups.
Is that part of the public contract?
Is it worth a PR?