manager/allocator/cnmallocator: add temporary adaptor for constructor

[thaJeztah]

• possible alternative to libnet: add idm.NewWithNoStore moby#45843
• relates to / supersedes / closes Change idm.New() signature #3130
• supersedes / closes [draft] manager/allocator/cnmallocator: remove unused nType argument #3015

Add an adaptor to help with a signature change in the Idm constructor.

- What I did

- How I did it

- How to test it

- Description for the changelog

[neersighted]

Looks like this has legitimate test failures hitting the "invalid constructor" branch -- but I like this idea in principle, if we can make it work...

[thaJeztah]

Yes, looks like I messed it up, or at least didn't handle everything, but I can give it another try tomorrow

[thaJeztah]

🤦 that took me some time to understand "why it doesn't match".... got bitten by Go's strong typing;

This will only match a function that's an actual "one", not "any function with signature func(string)

type one func(string)

See below (notice that both one and three are allowed in the switch .. because, well, they're not the same:
code below will match "two", not "one": https://go.dev/play/p/L_J_To1xRhp

package main

import "fmt"

type one func(string)
type two = func(string)
type three func(string)

func myFunc(string) {}

func assertType(fn any) {
	switch f := fn.(type) {
	case one:
		fmt.Printf("one: %T == %T", f, fn)
	case two:
		fmt.Printf("two: %T == %T", f, fn)
	case three:
		fmt.Printf("three: %T == %T", f, fn)
	default:
		fmt.Printf("unknown: %T != %T", f, fn)
	}
}

func main() {
	assertType(myFunc)
}

[codecov-commenter]

Codecov Report

Merging #3139 (e21d93d) into master (93fe90a) will decrease coverage by 0.08%.
The diff coverage is 56.25%.

@@            Coverage Diff             @@
##           master    #3139      +/-   ##
==========================================
- Coverage   61.76%   61.69%   -0.08%     
==========================================
  Files         154      154              
  Lines       31106    31120      +14     
==========================================
- Hits        19214    19198      -16     
- Misses      10346    10383      +37     
+ Partials     1546     1539       -7     

[thaJeztah]

Whoop; this is green, and my test-PR in moby with this change is also green;

• [WIP] testing libnet: remove package bitseq with swarmkit changes moby#45876

@dperny @neersighted PTAL

[thaJeztah]

I think I can combine this PR with #3015, which now does something very similar for another signature change; let me update this one.

[corhere]

You're damn right there is no way to pass a PluginGetter when constructing the non-deprecated drvregistry types. That's because they don't need one. See also moby/moby@28edc8e

[corhere]

Suggested change

	idmConstructor = func(id string, start, end uint64) (*idm.Idm, error)
	idmConstructor = func(start, end uint64) (*idm.Idm, error)

Without the datastore, id would also be unused.

[corhere]

The only consumers of libnetwork/idm are this file and libnetwork's ovmanager driver. Notably, ovmanager initializes its idm with start=0, which means the idm transparently delegates everything to the underlying bitmap and can therefore be trivially replaced with a bitmap.Sequence. That leaves this file as the only consumer of idm which actually needs it. Therefore I propose moving the idm package into swarmkit, which would make the adapters unnecessary.

[corhere]

This PR has been superseded by #3143