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".
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.
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:"
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.
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.
That's why you find it in the comments. That is where tools have found a place to add their own syntax without breaking the Go code.
Absolutely you can do the same in Java. It exists to the exact same degree as it does in Go. I expect it isn't done as often in the Java world because it is much harder to parse Java code so the tools don't get built.
> 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.
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.
It only seems right because there are no other implementations and the distinction is meaningless. You don't want to live in the world of multiple slightly different implementations all of which you need to support, this is, roughly, what the C++ build story looks like, and it's not fun.
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.
I suppose, to minimize its use. If annotations have the same syntactic weight as normal statements, such as “if” or “for” statements, there’s a temptation to use them liberally, which is clearly not a good fit for Go.
By making them comments, Go subtly signals that these are exceptional, making them less prominent and harder to abuse.
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.)
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)?
> 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.
That does seem a strange argument, it could simply be `%%` (or whatever) to introduce a 'metadata comment', and then a Go implementation that doesn't support metadata would simply lex both `%%` and `//` as comments and treat them identically.
I'd buy it. AFAIK the goal of Go is to have as little breaking changes as possible between versions.
Although they introduced backwards compat breaking features such as generics have they not ?
"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.
You are right that there could be new syntax, like, say, `@tool:name args` or `#tool.name args`, but is that any different than `//tool:name args`? They all read the same to me.
The upside of that particular syntax is that only the parser used by tools needs to understand directives. All other parser implementations can be blissfully unaware, negating the need for special no-ops. The downside is...?
I mean, technically you could write your entire business logic inside comments and have some tool parse it successfully. But we don't do that, because intuitively we know that's not the right place for it.
The issue isn't that this approach is incorrect, it's that it feels out of place. A comment should be a comment, nothing more. When comments start carrying executable meaning or structured directives, they stop serving their primary purpose.
It also becomes difficult to represent anything moderately complex in a clear way. Once the structure grows beyond something trivial, readability suffers quickly.
To me, it ends up feeling like command-line arguments.. technically workable, but messy and hard to reason about. Just look at something like "ffmpeg" arguments.. and then compare that to defining the same configuration through a structured format like Yaml or Json. The latter is simply clearer and easier to maintain.
> I mean, technically you could write your entire business logic inside comments and have some tool parse it successfully.
It sounds like you are talking about cgo. I think you have a stronger case there, but it is much the same situation: It's conceptually a third-party add-on that the Go language doesn't know anything about. "Cgo is not Go"[1]
I mean, if you really did have your own business logic language that you needed to include in Go code, where else would you put it if not in the "comments"? You, a random third-party, won't be able to add syntax to Go, so you cannot reasonably consider that to be an option. What syntax could Go add that is flexible enough to allow anyone to add anything they need, but that doesn't end up being comments by another token?
> A comment should be a comment, nothing more.
It's not really a comment, though. It is a directive. A comment is structured like `// Who the hell wrote this crap?`, while this is structured like `//tool:name args`.
I think what you are saying is that you don't like overloaded syntax, which is fair, but Go overloads syntax in several places so this is not a unique case. Besides, most languages end up with overloaded syntax in practice, so it isn't even something unique to Go.
To be clear, this isn't a feature of Go. There is nothing in the spec about it. It is what outside tools have hacked on top of Go because it doesn't offer that infinitely flexible feature mentioned above. However, it remains to be seen how you add infinitely flexible syntax without it turning into just being comments again.
I think your latest comment clarifies a lot of things for me here... primary being that it's something language designer did not wanted... just that tool developers went ahead with because there wasn't somethinf else (or something better) they could use.
If that is indeed the case, I believe it's fair. Im not into language/compiler design etc., but if I have to take a guess, this is where metaprogramming would have helped, right? Like the ones provided by zig or rust?
This is typical Go design, other languages do it worse, it isn't really needed, and then it gets added half way as it was supposed to be if done early on, and everyone cheers how Go is a "simple" language.
Just like some other famous languages of the authors.
It's abject hackery because such "comment" directives don't even necessarily have to be a comment to be active:
const x = `This whole thing is
a
//go:generate worse-is-better
multiline
string literal`
A lot of people in this discussion are beating up Go for using syntactical comments for directives, but in reality the implementation is even less principled than that!
I know this is off-topic, and likely to get me down-voted, but hey! I'll live dangerously for the sake of repeating a hilarious (to me) .sig in a now ancient Usenet post.
These //go.* commands always remind me of this:
"""
//GO.SYSIN DD *
DOO DAH
DOO DAH
"""
(Why yes, that is IBM System 360 JCL from circa 1975. Why do you ask?)
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.
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]()
}
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.
https://wiki.c2.com/?HotComments
//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.
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.
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.
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:"
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.
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).
That's why you find it in the comments. That is where tools have found a place to add their own syntax without breaking the Go code.
Absolutely you can do the same in Java. It exists to the exact same degree as it does in Go. I expect it isn't done as often in the Java world because it is much harder to parse Java code so the tools don't get built.
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.
I keep being impressed at subtle but meaningful things that Go does right.
For other things, like `//go:noinline`, this is fair criticism. `//go:fix inline` is quite different in every way.
By making them comments, Go subtly signals that these are exceptional, making them less prominent and harder to abuse.
(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.)
(Though for the record, talking about alternative implementations when discussing Go is kind of a funny joke.)
But to be fair to alternative toolchains, TinyGo and TamaGo are also a thing.
https://news.ycombinator.com/item?id=47395574
[1] https://pkg.go.dev/go/ast#Directive
> 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.
"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.
The upside of that particular syntax is that only the parser used by tools needs to understand directives. All other parser implementations can be blissfully unaware, negating the need for special no-ops. The downside is...?
The issue isn't that this approach is incorrect, it's that it feels out of place. A comment should be a comment, nothing more. When comments start carrying executable meaning or structured directives, they stop serving their primary purpose.
It also becomes difficult to represent anything moderately complex in a clear way. Once the structure grows beyond something trivial, readability suffers quickly.
To me, it ends up feeling like command-line arguments.. technically workable, but messy and hard to reason about. Just look at something like "ffmpeg" arguments.. and then compare that to defining the same configuration through a structured format like Yaml or Json. The latter is simply clearer and easier to maintain.
It's not wrong, but, it doesn't feel right.
It sounds like you are talking about cgo. I think you have a stronger case there, but it is much the same situation: It's conceptually a third-party add-on that the Go language doesn't know anything about. "Cgo is not Go"[1]
I mean, if you really did have your own business logic language that you needed to include in Go code, where else would you put it if not in the "comments"? You, a random third-party, won't be able to add syntax to Go, so you cannot reasonably consider that to be an option. What syntax could Go add that is flexible enough to allow anyone to add anything they need, but that doesn't end up being comments by another token?
> A comment should be a comment, nothing more.
It's not really a comment, though. It is a directive. A comment is structured like `// Who the hell wrote this crap?`, while this is structured like `//tool:name args`.
I think what you are saying is that you don't like overloaded syntax, which is fair, but Go overloads syntax in several places so this is not a unique case. Besides, most languages end up with overloaded syntax in practice, so it isn't even something unique to Go.
To be clear, this isn't a feature of Go. There is nothing in the spec about it. It is what outside tools have hacked on top of Go because it doesn't offer that infinitely flexible feature mentioned above. However, it remains to be seen how you add infinitely flexible syntax without it turning into just being comments again.
[1] https://go-proverbs.github.io
If that is indeed the case, I believe it's fair. Im not into language/compiler design etc., but if I have to take a guess, this is where metaprogramming would have helped, right? Like the ones provided by zig or rust?
Just like some other famous languages of the authors.
These //go.* commands always remind me of this:
"""
//GO.SYSIN DD *
DOO DAH
DOO DAH
"""
(Why yes, that is IBM System 360 JCL from circa 1975. Why do you ask?)
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.
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.
It is just a demo.
There was even a more upvoted post between your triple dupe and this https://news.ycombinator.com/item?id=47347322 #scp