Remove OrderedDictionary usage from ReflectTypeDescriptionProvider #47473
Conversation
This is showing up a bit on ASP.NET startup paths, and we can be more efficient in how we handle allocations and work.
stephentoub
added
area-System.ComponentModel
tenet-performance
|
Tagging subscribers to this area: @safern Issue DetailsThis is showing up a bit on ASP.NET startup paths, and we can be more efficient (and simpler) in how we handle allocations and work. As a microbenchmark, I tweaked TypeDescriptor.GetAttributes in both the original and PR bits to not cache attributes, so that the benchmark would simulate the impact on the first call per type. Then used it on types with no attributes, one attribute, and five identical attributes:
So, not a huge win, but the code gets simpler as well, and we remove one of the remaining dependencies on OrderedDictionary.
|
|
Oops, it seems like there was no CI ran on this PR (I have no idea why), just noticed now by looking at the merge summary. Let's keep an eye on the rolling build to see if something breaks. @MattGal @markwilkie who can we have take a look at why this happened? |
Only Azure Devops can, and reading through their documents they would like you to report stuff like this via their public-facing developer community portal at https://developercommunity.visualstudio.com . I filed https://developercommunity.visualstudio.com/content/problem/1326549/pr-checks-never-started-on-github-pr.html on your behalf here. |
|
Thank you @MattGal, appreciate it 😄 |
ghost
locked as resolved and limited conversation to collaborators
3 participants
This is showing up a bit on ASP.NET startup paths, and we can be more efficient (and simpler) in how we handle allocations and work.
As a microbenchmark, I tweaked TypeDescriptor.GetAttributes in both the original and PR bits to not cache attributes, so that the benchmark would simulate the impact on the first call per type. Then used it on types with no attributes, one attribute, and five identical attributes:
So, not a huge win, but the code gets simpler as well, and we remove one of the remaining dependencies on OrderedDictionary.