> - The reduced code readability if generics are used. This slows down development pace in Go.
Speak for yourself: I prefer generics to the mess if copy-pasta or generated stuff that one had to use before.
Speak for yourself: I prefer generics to the mess if copy-pasta or generated stuff that one had to use before.
What priority would you assign a bug that migrates the existing, working code that other people are doing god knows what with downstream? What would the theoretical benefits "better code" be to your downstream, and how do they weigh against the cost of "I ran go mod tidy, and our build broke?"
I can think of a use where it would lead to better code, and that's in k8s custom resources, where each resource type also has an associated list type that people create with code generation. It'd be much neater for k8s lists to be
type List[T runtime.Object] struct {
...
Items []T
}
It is.
> Additionally, it may slow down the resulting code.
FUD.
- Go compiler may fail to inline the callback passed to the maps.DeleteFunc(). This will result into an additional overhead for callback calls per each item in the map.
- Go compiler may move some variables inside the callback from stack to a heap. This will result in an additional memory allocations comparing to a simple loop, leading to an additional load on garbage collector.
The strings.Split() implementation is non-trivial because of performance optimizations.
The bytes.Equal() is actually written in highly tuned and optimized assembly in order to achieve high performance for inputs of various lengths.
Now compare this to trivial implementations behind generic-based functions for maps. And do not forget that these implementations may hurt performance because of excess memory allocations in Keys() and Values() functions or because the compiler may fail inlining the callback passed to DeleteFunc().
Though part of why they’re optimized and in the stdlib in the first place is because they’re such common patterns. So without them people would end up writing trivial, unperformant, custom versions. So now that more routines can be moved into the stdlib, they can benefit from optimization later.
(I’m not sure how much the maps routines specifically can be optimized, but stdlib routines routines can generally be more aggressive with unsafe or asm or being coupled to the runtime and its quirks, like bytes.Clone, strings.Builder, etc.)
And there are still plenty of ubiquitous patterns that have been worth including in the stdlib even if they’re usually just simple loops that aren’t very optimizable. Like strings.Index is an easy loop to write, but it comes up so often. Or strings.Cut is basically just an if-statement. But it makes code clearer about its intentions; and optimizations to these down the road benefit everyone.
It’s also true that maps.Keys and maps.Values allocate slices, and that you could avoid this with a loop, but strings.Split, bytes.Split, regexp.FindAll, os.ReadDir return slices and are still worthwhile as opposed to specialized iterators for each one. As with any code, you’re conscious of memory allocations where it counts, and optimize as needed.
In fact, now that generics make it possible, the Go team has discussed using iterators (https://github.com/golang/go/discussions/54245), which would benefit strings.Split even further in addition to all the other slice-returning functions.
So generally you have three options for those slice-returning functions:
- Custom inline loop for some of them. More verbose, will probably be naive and not benefit from stdlib optimizations. - Return a slice and iterate over it with a for loop. Creates allocations that could probably be avoided. - Create a customized iterator for that type. Unfortunately, you can’t really use an ordinary for loop, and extra custom iterators for each type. - Use generic iterators to benefit from the optimized functions and also avoid allocation overhead.
So part of the motivation is that now with generics there’s a variety of further optimizations available even to old functions like strings.Split and regexp.FindAll, in addition to opening up common patterns and optimizations for maps/slices/etc. to be included in the stdlib.
A few remarks:
> Like strings.Index is an easy loop to write, but it comes up so often
Actually, strings.Index() is very non-trivial function partially written in assembly in order to achieve high performance [1]. This function is used in Go projects *much more frequently* than functions from the golang.org/x/exp/maps package.
> strings.Cut is basically just an if-statement
No, strings.Cut() has non-trivial code when comparing to a trivial loop for map copy or for map delete [2].
> It’s also true that maps.Keys and maps.Values allocate slices, and that you could avoid this with a loop, but strings.Split, bytes.Split, regexp.FindAll, os.ReadDir return slices and are still worthwhile as opposed to specialized iterators for each one.
The *key* difference between maps.{Key,Value} and the mentioned functions from the standard library is that it is trivial to write the `for k, v := range m` instead of maps.{Key,Value} and avoid memory allocations, while it isn't trivial to write the corresponding code without memory allocations, which substitutes strings.Split() or other mentioned functions from the standard library.
[1] https://github.com/golang/go/blob/86c4b0a6ec70b07ab49d3813a5...
[2] https://github.com/golang/go/blob/86c4b0a6ec70b07ab49d3813a5...
This optimization then could speed up other similar interface-based algorithms.
[1] https://github.com/VictoriaMetrics/VictoriaMetrics/blob/mast...
I effectively used it in this little experimental CLI library: https://github.com/cpuguy83/go-cli/blob/main/command.go
It's pretty simple, but the nice thing is it can use any flag library you want (stdlib flag package, pflag, whatever).
In fact, the generic-free implementation based on the FlagSet interface is more flexible, since it allows storing multiple different FlagSet implementations in the same Cmd.
The main point is that generics come in handy when building libraries so that you aren't forcing callers of your library into specific types or loosing some type safety.
Of course I want sync.Map to use generics instead of interface{}. How could I not? And it’s less complex-looking than type-asserting everywhere.
The container/heap is more useful, but it could benefit more from adding an optimization for inlining interface method calls when Go compiler knows the underlying implementation behind the interface.
The golang.org/x/exp/maps is useless and may be harmful [1].
The golang.org/x/exp/slices is mostly useless, except of Sort*() functions. But it would be better to apply the optimization mentioned above to standard sort.* functions instead of forcing users to switch to different Sort*() implementations in other packages.
This is exactly what generics do. With e.g. a heap.Heap[uint32] the compiler knows the implementation and there’s no interface method call overhead.
In order for the compiler to do this optimization, it has to know that you don’t e.g. pass a *heap.Heap[uint32] to a function expecting *heap.Heap[uint64], so the type system is what allows it to optimize.
And on top of that, now the user also gets assurance at compile time that heap.Heap[uint32].Pop returns a uint32, preventing bugs from type confusion and also so you don’t have to add type assertions everywhere you use the heap.
So now heap, sort, etc. can benefit from this improved performance; users don’t have to write wrapper types and interface implementations just so their type can be sorted; and bugs are prevented at compile time.
For [1] I posted a reply. It’s true that there are overheads with some slice-returning routines but I explained how in the reply how I viewed the tradeoffs.
type customStruct struct { ... }
func (cs *customStruct) Less(i, j int) bool { ... }
func (cs *customStruct) Swap(i, j int) { ... }
func (cs *customStruct) Len() int { ... }
func sortMyCustomStruct(cs *customStruct) {
sort.Sort(cs)
}
They aren't widely used because the ergonomics suck, because they aren't generic yet.