//go:fix inline and the source-level inliner
go.dev125 points by commotionfever 4 days ago
125 points by commotionfever 4 days ago
If I follow, this isn't a compile time inline directive, it's a `go fix` time source transformation of client code calling the annotated function.
Per the post, it sounds like this is most effective in closed-ecosystem internal monorepo-like contexts where an organisation has control over every instance of client code & can `go fix` all of the call sites to completely eradicate all usage of a deprecated APIs:
> For many years now, our Google colleagues on the teams supporting Java, Kotlin, and C++ have been using source-level inliner tools like this. To date, these tools have eliminated millions of calls to deprecated functions in Google’s code base. Users simply add the directives, and wait. During the night, robots quietly prepare, test, and submit batches of code changes across a monorepo of billions of lines of code. If all goes well, by the morning the old code is no longer in use and can be safely deleted. Go’s inliner is a relative newcomer, but it has already been used to prepare more than 18,000 changelists to Google’s monorepo.
It could still have some incremental benefit for public APIs where client code is not under centralised control, but would not allow deprecated APIs to be removed without breakage.
yeah this is the part that got me excited honestly. we're not google-scale by any stretch but we have ~8 internal Go modules and deprecating old helper functions is always this awkward dance of "please update your imports" in slack for weeks. even if it doesn't let you delete the function immediately for external consumers, having the tooling nudge internal callers toward the replacement automatically is huge. way better than grep + manual PRs
it could be better than a nudge -- if you could get a mandatory `go fix` call into internal teams' CI pipelines that either fixes in place (perhaps risky) or fails the build if code isn't already identical to fixed code.
I'm not sure what all of the hazards are, but I could imagine a language (or a policy) where public APIs ship with all of the inline fix directives packaged as robust transactions (some kind of "API-version usage diffs"). When the client pulls the new API version they are required to run the update transaction against their usage as part of the validation process. The catch being that this will only work if the fix is entirely semantically equivalent, which is sometimes hard to guarantee. The benefits would be huge in terms of allowing projects to refine APIs and fix bad design decisions early rather than waiting or never fixing things "because too many people already depend on the current interface".
I wonder why they chose to add these directives as comments as opposed to adding new syntax for them. It feels like a kludge.
Go designers distinguish between Go language as defined by Go spec and implementation details.
//go:fix is something understood by a particular implementation of Go. Another implementation could implement Go without implementing support for //go:fix and it would be a fully compliant implementation of Go, the language.
If they made it part of the syntax, that would require other implementations to implement it.
If the comments impact correctness (which inlining doesn't, but I believe there are other directives that do), saying it's "an implementation detail" waves away "it's an implementation detail that everyone needs" aka part of the spec.
The reason it feels like a kludge is that "comments" are normally understood to be non-impactful. Is a source transformation that removes all comments valid? If comments have no impact per the spec, yes. But that's not the case here.
In practice comments in go are defined to be able to carry semantic meaning extensibly. Whether they're safe to ignore depends on what meaning is given to the directives, e.g. conditional compilation directives.
There are directives and packages that affect correctness. E.g. the embed package allows you to initialize a variable using a directive. E.g. //go:embed foo.json followed by var jsonFile string initializes the jsonFile variable with the contents of the foo.json file. A compiler or tooling that doesn't support this results in broken code.
There's nothing unique to Go about this kind of tooling. It exists in C, Java, Rust, Typescript, and probably dozens of other settings as well. It's the standard way of implementing "after-market" opt-in directives.
Are we referring to 'go fix' as after market tooling?
It's certainly done in many places. JsDoc is the biggest example I can think of. But they're all walking the line of "this doesn't have an impact, except when it does".
It being done by the language owners just makes them the ones walking the line.
That's exactly how this works: it doesn't have an impact, except when you ask it to. This is an idiomatic approach to this problem.
The part I object to is overloading comments, which aren't meant to be load bearing. A tool developed outside the language has no choice but to take something that has no meaning and overload it, but language owners weren't forced to take this route; they could've made directives not comments.
In practice, the Go language developers carved syntax out of comments, so that a comment is "anything that starts with //, unless the next characters are go:"
In the listed examples, the compiler will emit a diagnostic upon encountering those comments:
https://go.dev/blog/inliner#example-fixing-api-design-flaws
So these comments carry more weight than how those comment annotations might be consumed by optional tools for other languages.
For most of the listed examples, I think the corresponding C annotation would have been "[[deprecated]]", which has been added to the syntax as of C23.
It does not exist in Java. Comments in Java do not change code.
This also does not change th code. It is an advertisement to a linter-loke tool to take some action on the source code. Its most similar to linter directives which usually are comments.
We're talking about the "//go" comments in general I think here.
Things like "//go:embed" and "//go:build" very much do change the semantics of source code.
The comments above 'import "C"' containing C function definitions and imports change the compilation of go source code.
The "//go" comments contain a mix of ones that must be respected to compile, to being optional, to being entirely ignorable (like generate and fix).
> The reason it feels like a kludge is that "comments" are normally understood to be non-impactful. Is a source transformation that removes all comments valid? If comments have no impact per the spec, yes. But that's not the case here.
This is not inlining in the compiler. It's a directive to a source transformation (refactoring) tool. So yes, this has no impact on the code. It will do things if you run `go fix` on your codebase, otherwise it won't.
And yet it still breaks "comments aren't semantic". That transformation I described is still invalid.
I don’t understand why that wouldn’t be valid. As far as I understand if you compile code with these go:fix comments, they will be ignored. But if instead of compiling the code you run ‘go fix’, the source code will be modified to inline the function call. Only after the source code has been modified in this way would compiling reflect the inlining. Do you have a different understanding?
I mean that directives other than inlining impact correctness. If you have a source file that only builds for one OS, stripping the build tag will break your build.
That's such an elegant solution.
I keep being impressed at subtle but meaningful things that Go does right.
Because these are instructions for users for making tool-assisted changes to their source code, not a behavior that exists at runtime (or even compile time). A new syntax wouldn't make sense for it.
For other things, like `//go:noinline`, this is fair criticism. `//go:fix inline` is quite different in every way.
The //go:xyz comments are an established pattern in the Go tooling.
This is begging the question. Yes, but why did they do that over dedicated syntax?
(My personal theory is that early go had a somewhat misguided idea of simplicity, and preferred overloading existing concepts with special cases over introducing new keywords. Capitalization for visibility is another example of that.)
//go:xyz is dedicated syntax that is compatible with both the language spec and other toolchains that don't know about it.
It's an overloaded comment. I am personally quite fine with it, I don't think it's bad. but it is an overloaded comment.
I'm no longer sure what you're saying. You asked why they didn't go with dedicated syntax, I listed two advantageous aspects of the chosen syntax. We know it's an overloaded comment: that's literally one of the advantages.
Well, I've been unable to follow you as well, then. Obviously if they'd used a different type of syntax (e.g. using # for annotations), those would also be compatible with the language spec, and other implementations would still be just as capable of ignoring all unknown annotations.
(Though for the record, talking about alternative implementations when discussing Go is kind of a funny joke.)
Can't golang devs prioritize something like annotations or other attribute/metadata system instead of writing these in comments? I'm pretty sure this must have been raised a lot of times before, so just wanted to ask if there is/are any specific reason(s)?
These are called directives [1], and are treated as metadata by the compiler.
Understood... but why in comments?
Someone else said this below...
> Go designers distinguish between Go language as defined by Go spec and implementation details. > //go:fix is something understood by a particular implementation of Go. Another implementation could implement Go without implementing support for //go:fix and it would be a fully compliant implementation of Go, the language. > > If they made it part of the syntax, that would require other implementations to implement it.
...I'm not sure I buy that argument TBH.
hmm... thanks... And yes, I don't buy it either.
"If they made it part of the syntax, that would require other implementations to implement it." ... I mean, so what? Has golang stopped ading new features to the spec? If not (which I guess so), then how is this any different? Unless you have freezed the language, this reasoning doesn't make sense to me.
Good illustration that a seemingly simple feature could require a ton of functionality under the hood. Would be nice to have this in Python.
It looks the following code will be rewritten badly, but no ways to avoid it? If this is true, maybe the blog article should mention this.
package main
//go:fix inline
func handle() {
recover()
}
func foo() {
handle()
}
func main() {
defer foo()
panic("bye")
}recover()'s semantics make it so that "pointless" use like this can be inlined in a way that changes its semantics, but "correct" use remains unchanged.
Yes, maybe some code uses recover() to check if its being called as a panic handler, and perhaps `go fix` should add a check for this ("error: function to be inlined calls recover()"), but this isn't a particularly common footgun.
> ... and perhaps `go fix` should add a check for this (
This is an impossible task. For a library function, you can't know whether or not the function is defer called.
Maybe this is not an important problem. But it would be better if the blog article mentions this.
Great example, illustrating go1.26.1 go fix source inline transformation breaking program semantics. Raise it as a bug against go fix?
As I have mentioned, no ways to fix it. Because it is hard to know whether or not the handle function is called in a deferred call.
Another example (fixable):
package main
import "unsafe"
//go:fix inline
func foo[T any]() {
var t T
_ = 1 / unsafe.Sizeof(t)
}
func main() {
foo[struct{}]()
}
another:
package main
type T = [8]byte
var a T
//go:fix inline
func foo() T {
return T{}
}
func main() {
if foo() == a {
}
}
similar:
package main
//go:fix inline
func foo[T [8]byte | [4]uint16]() {
var v T
var n byte = 1 << len(v) >> len(v)
if n == 0 {
println("T is [8]byte")
} else {
println("T is [4]uint16]")
}
}
func main() {
foo[[8]byte]()
}Earlier submission: https://news.ycombinator.com/item?id=47385766
Far later submission. Check the ID again.. you were 2 days later.
There was even a more upvoted post between your triple dupe and this https://news.ycombinator.com/item?id=47347322 #scp