Skip to content

Tidy up benchmarks #263

@dkharms

Description

@dkharms

There are several things we need to fix.

Since go 1.24 pereferable way to write benchmarks is to use b.Loop. We have a lot of benchmark functions which use b.N and some of them rely on b.N value which is incorrect.

Some benchmarks are just useless (take a look at node.BenchmarkCopy). We need to get rid of such benchmarks. Not just because they are useless but also some of them take a lot of time to execute (they are too fast and it takes time for testing package to figure out number of iterations to reach duration of 1 second).

Some of benchmarks are incorrect. For example, node.BenchmarkOr. Node size depends on b.N and therefore (on some laptops) such benchmarks never converge. Here is article by Dave Cheney (https://dave.cheney.net/2013/06/30/how-to-write-benchmarks-in-go) - take a look at Traps for young players section. Also for some benchmarks (again node.BenchmarkCopy) compiler optimizes away all the code.

  • use b.Loop everywhere;
  • remove useless benchmarks;
  • do not rely on b.N;

Metadata

Metadata

Assignees

Labels

benchmarksFeatures or improvements over our benchmark processes

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions