Optimization for name value pair attribute sets #7676
Labels
feature
Feature request or proposal
Comments
|
What's extra here is having multiple implementations of the attrset interface, which has its own cost. Some thoughts:
That's all I have for now, but I'll conclude that it's worth trying. If this shows that a " |
|
Oh yeah, also #4090 would benefit from an abstracted attribute representation. I didn't have the time for it, but that's what I wanted to do next for that PR. Specifically I think that should be an explicit |
|
I was curious what the impact of such an optimization would be and measured a nixpkgs eval.
{
"cpuTime": 712.0946655273438,
"envs": {
"bytes": 17685813200,
"elements": 923836208,
"number": 643445221
},
"gc": {
"heapSize": 49061007360,
"totalBytes": 104299666128
},
"list": {
"bytes": 2495572504,
"concats": 54666602,
"elements": 311946563
},
"nrAvoided": 784922249,
"nrExprs": 3814091,
"nrFunctionCalls": 599839037,
"nrLookups": 315198579,
"nrOpUpdateValuesCopied": 1507273801,
"nrOpUpdates": 70052121,
"nrPrimOpCalls": 296204095,
"nrThunks": 849099332,
"sets": {
"bytes": 33392546032,
"elements": 1966117714,
"number": 120916413
},
"sizes": {
"Attr": 16,
"Bindings": 16,
"Env": 16,
"Value": 24
},
"symbols": {
"bytes": 2095785,
"number": 158751
},
"values": {
"bytes": 26435170488,
"number": 1101465437
}
}Number of key value pairs created by lib.nameValuePair550616 Attrset size distributionI think these numbers point to that there are some performance outliers to find in nixpkgs. List very long and disruptive, click to expand.Rendered as right-aligned because it better points out the outliers. |
3 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Is your feature request related to a problem? Please describe.
Name-value pairs like
{ name = "..."; value = "..."; }are frequently used, especially when converting between attribute sets and lists usingbuiltins.listToAttrsandlib.mapAttrsToList lib.nameValuePair. It would be beneficial for these operations to be optimized.Describe the solution you'd like
It should be possible to implement an optimization for such values by storing the
nameandvaluedirectly in theValueunion, just like how lists of sizes <= 2 are.Ping @roberth
Priorities
Add 👍 to issues you find important.
The text was updated successfully, but these errors were encountered: