fix: Remove abandoned im crate in favor of Arc::make_mut
#305
Conversation
|
NOTE: Please don’t merge yet. There were concerns about the performance, and I would like to measure the impact first. There are some benchmarks in #306 |
The benchmarks so far test the scope with tags and breadcrumbs in a few situations: * inactive client, * active client, only scope manipulation * active client, scope and capturing messages The Benchmarks also allow measuring the "minimal API" mode, which basically removes the complete implementation at compile time. Measurements on my system are: Minimal API (compile-time noop): ~2.6ns Tags: * inactive: 66ns * active: 26us * capture: 53us Breadcrumbs: * inactive: 2us * active: 52us * capture: 88us
|
Okay, I re-ran the benchmarks and get these results: Tags:
Breadcrumbs:
So at least with the testcase (relatively few [10-30] tags in the hashmap), we even get a nice speedup. Interesting that the hashmap testcase is sped up, but it kind of makes sense, since immutable hashmaps use buckets (I think 32 elements maybe?) so with a small number of tags they don’t win us anything. I think the number of k/v pairs should be fairly small usually, but I can also create another test with a higher number. For the breadcrumbs, the testcase runs into the default |
With a populated outer scope, do a few nested `with_scope` calls and add tags / breadcrumbs in those. Perf numbers are similar to before: -8% for tags (speedup) and +2% for breadcrumbs, so quite negligible.
This incurs a lot more cloning when using lots of scopes and breadcrumbs for example.
fixes #258