Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add memoise primop #10280

Draft
wants to merge 1 commit into
base: master
Choose a base branch
from
Draft

Add memoise primop #10280

wants to merge 1 commit into from

Conversation

roberth
Copy link
Member

@roberth roberth commented Mar 21, 2024

Motivation

Written by @edolstra, but PR opened by @roberth: This change comes up every now and then. I'm opening a draft PR for it, to increase visibility and so that a discussion can be had.


'builtins.memoise f x' is equal to 'f x', but uses a cache to speed up repeated invocations of 'f' with the same 'x'. A typical application is memoising evaluations of Nixpkgs in NixOps network specifications. For example, with the patch below, the time to evaluate a 10-machine NixOps network is reduced from 17.1s to 9.6s, while memory consumption goes from 4486 MiB to 3089 MiB. (This is with GC_INITIAL_HEAP_SIZE=16G.)

Nixpkgs patch:

diff --git a/pkgs/top-level/impure.nix b/pkgs/top-level/impure.nix index a9f21e45aed..f641067e022 100644
--- a/pkgs/top-level/impure.nix
+++ b/pkgs/top-level/impure.nix
@@ -79,7 +79,7 @@ in
 # not be passed.
 assert args ? localSystem -> !(args ? system || args ? platform);

-import ./. (builtins.removeAttrs args [ "system" "platform" ] // { 
+builtins.memoise or (x: x) (import ./.) (builtins.removeAttrs args [ "system" "platform" ] // {
   inherit config overlays crossSystem;
   # Fallback: Assume we are building packages on the current (build, in GNU
   # Autotools parlance) system.

Context

Priorities and Process

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

'builtins.memoise f x' is equal to 'f x', but uses a cache to speed up
repeated invocations of 'f' with the same 'x'. A typical application
is memoising evaluations of Nixpkgs in NixOps network
specifications. For example, with the patch below, the time to
evaluate a 10-machine NixOps network is reduced from 17.1s to 9.6s,
while memory consumption goes from 4486 MiB to 3089 MiB. (This is with
GC_INITIAL_HEAP_SIZE=16G.)

Nixpkgs patch:

diff --git a/pkgs/top-level/impure.nix b/pkgs/top-level/impure.nix
index a9f21e45aed..f641067e022 100644
--- a/pkgs/top-level/impure.nix
+++ b/pkgs/top-level/impure.nix
@@ -79,7 +79,7 @@ in
 # not be passed.
 assert args ? localSystem -> !(args ? system || args ? platform);

-import ./. (builtins.removeAttrs args [ "system" "platform" ] // {
+builtins.memoise or (x: x) (import ./.) (builtins.removeAttrs args [ "system" "platform" ] // {
   inherit config overlays crossSystem;
   # Fallback: Assume we are building packages on the current (build, in GNU
   # Autotools parlance) system.
@roberth roberth mentioned this pull request Mar 21, 2024
Copy link
Member Author

@roberth roberth left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Optimizations, and reduced strictness.

(and needs rebase)

if ((*this)(v2->listElems()[n], v1->listElems()[n])) return false;
}

return n == v1->listSize() && n < v2->listSize();
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The ordering established by this comparator doesn't have to be exactly lexicographical.
What if we compare the size first, and only compare lexicographically if the sizes are equal?

if ((*this)(i->value, j->value)) return true;
if ((*this)(j->value, i->value)) return false;
}
return i == v1->attrs->end() && j != v2->attrs->end();
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Like the list ordering, the size could be compared first.

Also, the set of names should be compared before evaluating anything extra.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants