Go Language¶

Domain gopkg.in has a different V2 convention

Typically, a v2 go module path looks like github.com/xieyuschen/yaml/v3, however the gopkg.in has its own convention which is gopkg.in/yaml.v3.

Since gopkg.in shares the same version suffix methodology as Go modules, the Go command accepts the .v2 in gopkg.in/yaml.v2 as a valid major version suffix. This is a special case for compatibility with gopkg.in: modules hosted at other domains need a slash suffix like /v2.

Mapping Values by a Combination of Two Keys

The topic was found during optimization. When the combination of 2 string type keys maps to a value, how do you maintain the mapping from keys to the value? Generally we can come up with 3 approaches listed below, which one is better?

type Result string

// 1d map, you need to concatenate 2 strings
var d1M map[string]Result

// 1d map and use struct as key
type Key struct {
    S1, S2 string
}
var d1MS map[Key]Result

// 2d map
var d2M map[string]map[string]Result

How Go Finds Std Libs

Usually, the go standard library is located under $GOROOT and it's go compiler's scope to find the std libs, so most of the users needn't care about how the standard library is found by compiler.

When we want to combine 2 services into a single process, we need to solve the collision of global states between 2 services in the standard library. By duplicating it with another name(and all underlying dependent packages) under GOROOT, we can segregate the global states well even though they refer to the same package before combining. However, this requires to investigate how go compiler finds the std lib to ensure the approach won't break.

Is Mutex+Chan Version of Once Better Than sync.Once?

In my previous blog packages.Load jitters, I said the jitters are caused by too many go routines are spawned so synchronization takes a lot of times.

However, at the beginning I thought the Lock in sync.Once costs a lot, so I tried to replace sync.Once with Mutex+Chan. The result is the sync.Once is still better.

golang/tools: reduce jitter of packages.Load

In CL 614975, I have migrated deprecated loader from ssa package. However, miller has reported the tests became extremely slows. In an operation-system with a slow file system, such as plan9, the test slowed by more than 50x.

Oops, a un-intentional change:( I have submitted a CL to load 300+ packages by packages.Load once instead of loading them many. Alan thought it's a bravo change:)

Moreover, when I involved to investigate why the packages.Load slows so much, I have learned a lot. This blog records how I investigate them.