Go 1.27 was released on August 19, and it closes out one of the longest-running gaps in the language’s generics implementation.

Generic methods

A method declaration may now declare its own type parameters.

When Go added generics in 1.18, methods were left out. Type parameters could only appear on functions and on type declarations, which meant a generic function could not live in the namespace of a particular data type. In practice you wrote a package-level function taking the receiver as its first argument and accepted that it read worse than it should.

That restriction is gone. The design took years partly because generic methods interact badly with Go’s interface satisfaction rules if you are not careful, and getting a version that does not break the type system took real work.

encoding/json/v2

The standard library gains encoding/json/v2 and encoding/json/jsontext. The former is a substantial revision of the existing JSON package; the latter provides lower-level syntactic JSON processing.

Critically, the existing encoding/json package is now backed by the v2 implementation, so unmarshaling gets faster without any code changes and without breaking backward compatibility. That is the interesting engineering decision here: rather than asking everyone to migrate, the Go team rebuilt the engine underneath the existing API.

If you have JSON-heavy services, this is a free performance win on upgrade.

Goroutine leak profiling

The goroutineleak profile is now generally available. It identifies goroutines permanently blocked on concurrency primitives: channels, mutexes, and condition variables.

Goroutine leaks are the classic Go production problem. Each leaked goroutine holds its stack and everything the stack references, so a slow leak looks like a slow memory leak with no obvious allocation site. Diagnosing it previously meant dumping all goroutine stacks and reading them, which works but does not scale to a service with tens of thousands of them.

A dedicated profile that reports only the permanently blocked ones is a considerably better tool.

Smaller but welcome

Size-specialized memory allocation routines cut the cost of some small allocations, under 80 bytes, by up to 30%. Small allocations dominate in most real programs, so this is not a microbenchmark curiosity.

crypto/mldsa implements ML-DSA, the post-quantum signature scheme standardized as FIPS 204. Go continues adding post-quantum primitives to the standard library ahead of any urgent need, which is the correct order to do it in.

Installing

# Check current version
go version

# Arch
sudo pacman -Syu go

# Fedora
sudo dnf upgrade golang

# Or install directly from go.dev

Debian and Ubuntu ship older Go versions in their repositories; the upstream tarball or a toolchain manager is the usual route there.