Some of the design decisions seem to me to be a bit more driven by being Rust-like than addressing Go's thorns though. In particular, using `impl` to define methods on types (https://borgo-lang.github.io/#methods), the new syntax for channels and goroutines (https://borgo-lang.github.io/#channels), and the `zeroValue()` built-in (https://borgo-lang.github.io/#zero-values-and-nil) seem a bit out of place. Overall though, if I had a choice, I would still rather write Borgo by the looks of it.

From my experience, things I think Go could really benefit from, like I believe it has benefited from generics:

* A way to implement new type constraints[1] for existing types, like `impl Trait for T`. This would obsolete most uses of reflection in the wild, in a way generics alone haven't. This isn't about syntax, it's about an entirely different way to associate methods to types, and with both Go and Rust being "data-oriented" languages, it's weird that Go is so limited in this particular regard. It has many other consequences when combined with generics, such as ...

* Ability to attach receiverless methods to types. Go concrete types may not need associated methods like "constructors", but generic types do, and there's no solution yet. You can provide factory functions everywhere and they infect the whole call graph (though this seems to be "idiomatic"), or make a special method that ignores its receiver and call that on a zero instance of the type, which is more hacky but closer to how free functions can be resolved by type. There's no reason this should be limited to constructors, that's just the easiest example to explain, in Rust associated methods are used for all kinds of things. Speaking of which...

* `cmp.Ordered` for custom types. Come on people. We shouldn't still have this much boilerplate to sort/min/max custom types, especially two full years after generics. Even boilerplate aside, `sort.Sort()` and especially `sort.Slice()` generate extremely un-optimized machine code, but it's not "idiomatic" to BYO sort implementations either because no other interface for this is standard. However, we would basically get this for free if both of the above points were solved, but it's also possible we get something else entirely that solves only ordering and not e.g. construction or serialization.

* Enums, especially if the need for exhaustiveness checking could be balanced with Go's values of making code easy to evolve later. When I need them, I use the `interface Foo { isFoo() }` idiom and accept heap boxing overhead, but even the official `deadcode` analysis tool still to this day does not recognize this idiom or support enough configuration to force it to understand. The Go toolchain could at the very least recognize idioms people are using to work around Go's own limitations.

If we had solutions to these problems, I think most Go folks would find enough value in them that they would still be "Go". In fact, I think people would have an easier time consolidating on a new standard way to do things rather than each come up with their own separate workarounds for it.

This is where I feel "The code I write, is going to look like the code written by most other on my team" the least, because that's only true when a Go idiom has some official status, it's not nearly as true for workarounds that the Go team has not yet chosen to either endorse or obsolete.

[1] Note that I'm not saying only interfaces, but type constraints in general. `cmp.Ordered` isn't an interface, it's a type constraint, so a full solution here has to work for that too.

Having worked on multiple very large Go codebases with many engineers, the lack of actual enums and a built-in optional type instead of nil drive me crazy.

I think I'm in love.

Edit: Looks like last commit was 7 months ago. Was this abandoned, or considered feature complete? I hope it's not abandoned!

You could look at Lisp. It's kind of the opposite of Go in this regard. You can use whatever paradigm you like, generate new code on the fly, use types or leave them. It even allows you to easily extend the language to your taste, all the way down to how code is read.

But Lisp might violate your set of absolutes.

I see an effect where the languages whose primary goal is a particular set language design choices (such as strict memory safety over all else) grow a cult following that enforces said design choices. Maybe in the pursuit of an opinionated language, even if the designers are reasonable at the language's inception, the community throws out logic and "opinionated" becomes a in-group out-group tribal caveman situation.

True.

> We could probably argue for all eternity about code formatting, for instance. But Go went and set it in stone.

This is part of the story that Rob Pikes uses to justify how opinionated Go is, but it's a bit stupid given that most language do fine and I've never seen any debates about the code formatting after the very beginning of a project (where it's always settled quickly in the few case where it happens in the first place).

The real reason why Go is opinionated is much more mundane: Rob is an old man who think he has seen it all and that the younger folks are children, and as a result he is very opinionated. (remember his argument against syntax coloring because “it's for babies” or something).

It's not bad to be opinionated when designing a language, it give some kind of coherence to it (looking at you Java and C++) but it can also get into the way of users sometimes. Fortunately Go isn't just Rob anymore and isn't impervious to changes, and there is finally generics and a package manager in the language!

Firstly, Ken Thompson is a master at filtering out unnecessary complexities and I highly rate his opinion of the important and unimportant things.

Secondly, the Go team were never against generics, the three early designers agreed the language needed generics but they couldn't figure out a way to add it orthogonally.

Go has gone on to be very successful in cloud and networked applications (which it was designed to cater for), which lends credit to the practicalities of what the designers thought as important, HN sentiments notwithstanding.

This is a PR statement that has been introduced only after Go generics landed, for years generics were dubbed “unnecessary complexity” in user code (Go has had generics from the beginning but only for internal use of the standard library).

> Go has gone on to be very successful in cloud and networked applications (which it was designed to cater for), which lends credit to the practicalities of what the designers thought as important

Well, given that the k8s team inside Google developed their own Go dialect with pre-processing to get access to generics, it seems that its limitations proved harmful enough.

The main reason why Go has been successful in back-end code is the same as the reason why any given language thrive in certain environments: it's a social phenomenon. Node.js has been even more successful despite JavaScript being a far from perfect language (especially in the ES 5 era where Node was born), which shows that you cannot credit success to particular qualities of the language.

I have nothing against Go, it's a tool that does its job fairly well and has very interesting qualities (fast compile time, self-contained binaries, decent performance out of the box), but the religious worship of “simplicity” is really annoying. Especially so when it comes in a discussion about error handling, where Go is by far the language which makes it the most painful because it lacks the syntactic sugar that would make the error as return value bearable (in fact the Go team was in favor of adding it roughly at the same time as generics, but the “simplicity at all cost” religion they had fostered among their users turned back against them and they had to cancel it…).

Nodejs has been more successful than Go in cloud?

For those who want to feel the wind of coding freedom blow through their hair, I can recommend to spend some time learning Lisp. It offers the most freedom you can possibly have in a programming language. It might enlighten you in many other ways. It won't be the last language you learn, but it might be the most fruitful experience.

You can format on save or in a pre commit hook. But that the language has a single canonical format makes it kind of new.

A simple settings file set in stone by the CTO, such a hard task to do.

The fact that is even a novelty unaware of, only confirms the target group for the language.

That said, suggesting adding exceptions to Go is about as reasonable as adding a GC to Zig. How much effort would you spend arguing against someone bringing that up as a serious proposal?

Suggesting the addition of exceptions to Go is as reasonable as suggesting the addition of loops to Rust. Which is to say that it already has exceptions, and always has. Much of the language's design is heavily dependent on the presence of exceptions.

Idioms dictate that you probably shouldn't use exceptions for errors (nor should you in any language, to be fair), but even that's not strictly adhered to. encoding/json in the standard library famously propagates errors using the exception handlers.

The only error-related concept the Go language has is the error type, but it comes with no handling semantics. Which stands to reason as there is nothing special about errors. Originally, Go didn't even have an error type, but it was added as a workaround to deal with the language not supporting cyclic imports.

Your pedantry is hilarious and contradictory.

Basically, discussions have context and I have no intention of prepending 10 disclaimers to every statement I make to preemptively guard against people interpreting my comments as absolutes in a vacuum.

It has its flaws, but the latest version (Scala 3) is really really good. The community is open to different styles of programming - from using it as a "better Java" to "pure functional programming like in Haskell".

Hilariously I was using a gen AI (phind) and asked it to generate some scala code and it no joke suggested the code in both implematic scala and in java style, and all you had to do is look at it and you could see java style was 1000X easier to read/maintain.

On the other hand, if you carefully select your team or work alone, then this is not a problem at all.

Btw, there isn't really "one" idiomatic scala style - therefore I tend to believe that you are not familiar with the language and the community.

    func try(fn func()) { fn() }
    func catch(fn func(any)) {
        if v := recover(); v != nil {
            fn(v)
        }
    }
    func throw(v any) { panic(v) }

    func fail() {
        throw("Bad things have happened")
    }

    func main() {
        try(func() {
            defer catch(func(v any) {
                fmt.Println(v)
            })
            fail()
        })
    }

The syscall/js package [0] throws panics if something goes wrong, rather than returning errors.

Go already has try/catch exceptions. We just don't use them most of the time because they're a really bad way of handling errors.

[0] https://pkg.go.dev/syscall/js

If you do naked if-err-return-err, you are likely doing error handling wrong.

Plenty of The Go Way arguments apply to software we were writing from the dawn of computing until the 1990's, and there are plenty of reasons why, with exception of Go (pun intended), the industry has moved beyond that.

In our Go codebases, the error reporting and subsequent debugging and bug fixing is night and day from our Python, Perl, Ruby, PHP, Javascript, and Elixir experiences.

The one glaring case where this is untrue is in our usage of Helm which, having been written in Go, I would expect better error handling. Instead we get, "you have an error on line 1; good luck" - and, inspired, I looked at their code just now. Littered with empty if-err-return-err blocks - tossing out all that beautiful context, much like an exception would, but worse.

https://github.com/search?q=repo%3Ahelm%2Fhelm%20%20if%20err...

In JS, for example, people don‘t even know which functions could throw exceptions, and just ignore them, most of the time. Fast to write and looks nice, but is horrible quality and a nightmare to debug.

Why do you think so? Maybe I'm an odd case, but my main use case for enums is for APIs and database designs, where I want to lock down some field to a set of acceptable values and make sure anything else is a mistake. Or for state machines. Error handling is manageable without enums (but I love Option/Result types more than Go's error approach, especially with the ? operator).

The thing is, these don't add much on their own. You'd have to bring in pattern matching and/or a bunch of other things* that would significantly complicate the language.

For example, with what's currently in the language, you could definitely have an option type. You'd just be limited to roughly an api that's `func (o Option[T]) IsEmpty() bool` and `func (o Option[T]) Get() T`. And these would just check if the underlying point is nil and dereference it. You can already do that with pointers. Errors/Result are similar.

A `try` keyword that expands `x := try thingThatProducesErr()` to:

    x, err := thingThatProducesErr()
    if err != nil {
        return {zero values of the rest of the function signature}, err
    }

* at the very least generic methods for flat map shenanigans

Just need to make sure the Option exposes the internal value only through:

    func (o Option[Value]) Get() (Value, bool) {
        return o.value, o.exists
    }

    if value, ok := option.Get(); ok {
        // value is valid
    }
    // value is invalid

A Result type would bring a similar benefit of fixing the few edge cases where an error may accidentally not be handled. Although I don't think it'd be worth the cost of switching over.

    if value != nil {
        // value is valid
    }
    // value is invalid

Of course, this is often left out, but you can just as easily do:

    value, _ := option.Get()

> Using an Option instead of a pointer buys you the inability to forget to check for nil.

> Of course, this is often left out, but you can just as easily do:

Unfortunately it gets brought up pretty much every time in these discussions.

Deliberate attempts to circumvent safety are not part of the threat model. The goal is prevention of accidental mistakes. Nothing can ultimately stop you from disabling all safeties, pointing the shotgun at your foot and pulling the trigger.

Then what you are really looking for is sum types (what Rust calls enums, but unusually so), not enums. Go does not have sum types, but you can use interfaces to archive a rough facsimile and most certainly to satisfy your specific expectation:

    type Hot struct{}
    func (Hot) temp() {}

    type Cold struct{}
    func (Cold) temp() {}

    type Temperature interface {
        temp()
    }

    func SetThermostat(temperature Temperature) {
        switch temperature.(type) {
        case Hot:
            fmt.Println("Hot")
        case Cold:
            fmt.Println("Cold")
        }
    }

They can certainly help solve some of the same problems. Does that make them related? I don't know.

By definition, an enumeration is something that counts one-by-one. In other words, as is used in programming languages, a construct that numbers a set of named constants. Indeed you can solve the problem using that:

    type Temperature int

    const (
        Hot Temperature = iota
        Cold
    )

    func SetThermostat(temperature Temperature) {
        switch temperature {
        case Hot:
            fmt.Println("Hot")
        case Cold:
            fmt.Println("Cold")
        }
    }

    type Temperature int

    const (
        Hot  Temperature = 0
        Cold Temperature = 1 
    )

    func SetThermostat(temperature Temperature) {
        switch temperature {
        case Hot:
            fmt.Println("Hot")
        case Cold:
            fmt.Println("Cold")
        }
    }

    const (
        Hot  Temperature = 0
        Cold Temperature = 1 
    )

Refactoring is harder - when you add a new value to the enum, you can't easily find all those places that may require logic changes to handle the new value.

Enums are a big thing I miss when writing Go, compared to when writing C.

Enumeration isn't a type, it's a numbering construct. Literally, by dictionary definition. Granted, if you use the Rust definition of enum then it is a type, but that's because it refers to what we in this thread call sum types. Rust doesn't support "true" enums at all.

> The compiler can't complain when you use a value that isn't in the list of enumerations.

Well, of course. But that's not a property of enums. That's a property of value constraints. If Go supported value constraints, then it could. Consider:

    type Temperature 0..1

    const (
        Hot  Temperature = 0
        Cold Temperature = 1 
    )

    type Email "{string}@{string}"

> Enums are a big thing I miss when writing Go, compared to when writing C.

Go has enums. They are demonstrated in the earlier comment. The compiler doesn't attempt to perform any static analysis on the use of the use of the enumerated values because, due to not having value constraints, "improper" use is not a fatal state[1] and Go doesn't subscribe to warnings, but all the information you need to perform such analysis is there. You are probably already using other static analysis tools to assist your development. Go has a great set of tools in that space. Why not add an enum checker to your toolbox?

[1] Just like it isn't in C. You will notice this compiles just fine:

    typedef enum {
        Hot,
        Cold
    } Temperature;

    void setThermostat(Temperature temperature) {
        switch (temperature) {
        case Hot:
            printf("Hot\n");
        }
    }

    int main() {
        setThermostat(10);
    }

No, it isn't, unlike C, in which it is. The C compiler can actually differentiate between an enum with one name and an enum with a different name.

There's no real reason the compiler vendor can't add in warnings when you pass in `myenum_one_t` instead of `myenum_two_t`. They may not be detecting it now, but it's possible to do so because nothing in the C standard says that any enum must be swappable for a different enum.

IOW, the compiler can distinguish between `myenum_one_t` and `myenum_two_t` because there is a type name for those.

Go is different: an integer is an integer, no matter what symbol it is assigned to. The compiler, now and in the future, can not distinguish between the value `10` and `MyConstValue`.

> Just like it isn't in C. You will notice this compiles just fine:

Actually, it doesn't compile "just fine". It warns you: https://www.godbolt.org/z/bn5ffbWKs

That's about as far as you can get from "compiling just fine" without getting to "doesn't compile at all".

And the reason it is able to warn you is because the compiler can detect that you're mixing one `0` value with a different `0` value. And it can detect that, while both are `0`, they're not what the programmer intended, because an enum in C carries with it type information. It's not simply an integer.

It warns you when you pass incorrect enums, even if the two enums you are mixing have identical values. See https://www.godbolt.org/z/eT861ThhE ?

Go on. Given:

    type E int
    const (
        A E = iota
        B
        C
    )

    enum E {
        A,
        B,
        C
    }

That's all C gives you. It provides nothing more. They are exactly the same (syntax aside).

> It warns you

Warnings are not fatal. It compiles just fine. The Go compiler doesn't give warnings of any sort, so naturally it won't do such analysis. But, again, you can use static analysis tools to the same effect. You are probably already using other static analysis tools as there are many other things that are even more useful to be warned about, so why not here as well?

> enum in C carries with it type information.

Just as they do in Go. That's not a property of enums in and of themselves, but there is, indeed, an associated type in both cases. Of course there is. There has to be.

    type Option1 struct { ... }
    type Option2 struct { ... }
    type MyEnum interface { Option1 | Option2 }

    var myValue MyEnum // currently not legal Go

So when I looked at Go for the first time, the error handling was one of the many positive features.

Is there any good reason for wanting try/catch other than being lazy?

sounds good on paper, but seeing "if err!=nil" repeated million times in golang codebases does not create positive impression at all

Okay, but other than exceptions, whats the alternative?

I only briefly tried Rust and was turned off by the poor ergonomics; I don't think (i.e. open to correction) that the Rust way (using '?') is a 1:1 replacement for the use-cases covered by Go error management or exceptions.

Sometimes (like in the code I wrote about 60m ago), you want both the result as well as the error, like "Here's the list of files you recursively searched for, plus the last error that occurred". Depending on the error, the caller may decide to use the returned value (or not).

Other times you want an easy way to ignore the error, because a nil result gets checked anyway two lines down: Even when an error occurs, I don't necessarily want to stop or return immediately. It's annoying to the user to have 30 errors in their input, and only find out about #2 after #1 is fixed, and #3 after #2 is fixed ... and number #30 after #29 is fixed.

Go allows these two very useful use-cases for errors. I agree it's not perfect, but with code-folding on by default, I literally don't even see the `if err != nil` blocks.

Somewhat related: In my current toy language[1], I'm playing around with the idea of "NULL-safety" meaning "Results in a runtime-warning and a no-op", not "Results in a panic" and not "cannot be represented at all in a program"[2].

This lets a function record multiple errors at runtime before returning a stack of errors, rather than stack-tracing, segfaulting or returning on the first error.

[1] Everyone is designing their own best language, right? :-) I've been at this now since 2016 for my current toy language.

[2] I consider this to be pointless: every type needs to indicate lack of a value, because in the real world, the lack of a value is a common, regular and expected occurrence[3]. Using an empty value to indicate the lack of a value is almost certainly going to result in an error down the line.

[3] Which is where there are so many common ways of handling lack of a value: For PODs, it's quite popular to pick a sentinel value, such as `(size_t)-1`, to indicate this. For composite objects, a common practice is for the programmer to check one or two fields within the object to determine if it is a valid object or not. For references NULL/null/nil/etc is used. I don't like any of those options.

Golang code has a rhythm: you do the thing, you check the error, you do the thing, you check the error. After a while it becomes automatic and easy to read, like any other syntax/formatting. You notice if the error isn't checked.

Yes, at first it's jarring. But to be honest, the jarring thing is because Go code checks the error every time it does something, not because of the actual "if err != nil" syntax.

I've gotten used to Javas getter/setter spam, does that make it a good idea?

Moreover, don't you think that something like Rusts ? operator wouldn't be a perfect solution for handling the MOST common type of error handling, aka not handling it, just returning it up the stack?

  val, err := doAThing()
  if err != nil {
    return nil, err
  }

  val := doAThing()?

I'd be very surprised if, in Rust codebases, there's not an implicit bias against wrapping and towards using `?`, just to help keep things "clean"; which has implications not only for debugging, but also for situations where doing something more is required for correctness.

Rust's error handling is clearly better than Go's, but Go's is better than exceptions and the complaints about verbosity are largely complaints about having to actually consider errors.

It's the best strategy for short running programs, or scripts if you will. You just write code without thinking about error handling at all. If anything goes wrong at runtime, the program aborts with a stacktrace, which is exactly you want and you get it for free.

For long-running programs you want reliability, which implies the need to think about and explicitly handle each possible error condition, making exceptions a subpar choice.

Because fundamentally the function you called can return different errors at any point so if you just propagate the error the code paths are in fact not spelled out at all because the function one above in the hierarchy has to deal with all the possible errors two calls down which are not transparent at all.

There's no requirement for the calling function to handle each possible type of error of the callee. It can, as long as the callee properly wrapped the error, but it's relatively rare for that to be required. Usually the exact error is not important, just that there was one, so it gets handled generically.

1. err != nil, nondefault return value

2. err != nil, default return value

3. err == nil, nondefault return value

4. err == nil, default return value

when often what you want to express only has two: either you return an error and there's no meaningful output, or there's output and no error. A type system with tuples but no sum types can only express "and", not "or".

In a hot path it’s often beneficial to not have lots of branches for error handling. Exceptions make it cheap on success (yeah, no branches!) and pretty expensive on failure (stack unwinding). It is context specific but I think that can be seen as a good reason to have try catch.

Now of course in practice people throw exceptions all the time. But in a tight, well controlled environment I can see them as being useful.

This is true but the branch isn't taken unless there's an error in Go.

Given that the Go compiler emits the equivalent of `if (__unlikely(err != nil)) {...}` and that any modern CPUs are decently good at branch prediction (especially in a hot path that repeats), I find it hard to believe that the cost would be greater than exceptions.

You generally need to skip all lines that the exception invalidates. That's why it's a block or conditional.

Try/catch is super confusing because the catch is often far away from the try. And in Python I just put try/catch around big chunks of code just in case for production.

I think Go is more stable and readable because they force you not to use the lazy unreadable way of error handling.

Enums I honestly never used in Go also not the not-type-safe ones.

But I'm also someone who used interfaces in Go maybe I think 4 times only in years and years of development.

I just never really need all those fancy things.

For example, and option type with enums combined can ensure return values are checked by providing a compile time error if a case is missing (as expressed in the first few examples of the readme).

But they decided not to add enums because it conflicted and overlapped too much with interfaces.

I just want to add "my" experience that personally, yes maybe you can argue enums are nice, but I never missed them in Go.

I personally agree with the Go team on how they argue and for me it would be a step back if they listened to the herd that does not take all sides of the story into consideration but just keeps pushing enums.

Try/catch is just a really bad thing all "hacky solution" alarm bells go off for me if you want to change error handling to giant try/catch blocks.

I'm very curious now about how it might conflict and/or overlap with interfaces.

To reach the goal of an enumeration type (and all the strong type-checking that that brings with it), enums could look as simple as:

    type DayNames enum {
       Sunday
       Monday
       Tuesday
       Wednesday
       Thursday
       Friday
       Saturday
    }
    ...
    func isFunDay (dow DayNames) {
       // This must fail to compile, because there is an unhandled enumeration
       switch {
          case Sunday: ...
          case Monday: ...
          case Tuesday: ...
          case Thursday: ...
          case Friday: ...
          case Saturday: ...
       }
       ...
    }
    ...
    isFunDay (0)   // Compile failure
    var x int
    isFunDay (x)   // Compile failure

Hah :-)

> Here you Go: https://go.dev/doc/faq#variant_types

Not quite the same: Variants are a constrained list of types. Enums are a constrained list of values.

Let's assume that I agree with the reasoning for not having a constrained list of types.

It still doesn't tell me why we can't have a constrained list of values.

But you ruined it with "fancy things" which shows offhand disregard and disrespect.

A question like "what do you need these features for?" would have been a better contribution to the forum.

I also wanted to add that I used inheritance only ONCE in all my years of writing Python in all other millions of lines of code inheritance was not the best solution.

This is my daily struggle as a CTO. People using waaayy too many "fancy" features of languages making it totally unreadable and unmaintainable.

It's their ego they want to show off how many complex language features they know. And it's ruining my codebases.

Enumerated types are simply named integers in most languages, exactly the sort you get with const / iota in Go: https://en.wikipedia.org/wiki/Enumerated_type

Rather than the tagged union which the word represents in Rust, and only Rust. Java's enums are close, since they're classes and one can add arbitrary behaviors and extra data associated with the enum.

It remains a strange choice to refer to this as the true enum, actual enum, real enum, as has started occurring since Rust became prominent. If that's a meaningful concept, it means a set of numeric values which may be referred to by name. This is the original definition and remains the most popular.

From a PL theory perspective, enum denotes an enumerable set of values within a type. It just happens that sums slot in well enough with that.

I didn't read GP as saying "Actual enums are what Rust has", I read it more as "Go doesn't have actual enums", where "enum" is a type that is constrained to a specified set of values, which is what all mainstream non-Rust languages with enums call "Enums".

I mean, even if Rust never existed, the assertion "Go doesn't have actual enums" is still true, no?

I'd love to be able to use a bit more type-y Go such as Borgo, and have a Pythonesque dynamic scripting language that latches onto it effortlessly.

Dynamic typing is great for exploratory work, whether that's ML research or developing new features for a web app. But it would be great to be able to morph it over time into a more specified strongly typed language without having to refactor loads of stuff.

Like building out of clay and firing the parts you are happy with.

Could even have a three step - Python-esque -> Go/Java-esque -> Rust/C++esque.

We do exactly that with Common Lisp. It compiles to different languages/frameworks depending on what we require (usually sbcl is more than enough, but for instance for embedded or ML we need another step. All dev (with smaller data etc) is in sbcl so with all the advantages.

For learning cl, the reddit lisp subreddit is good and has the current best ones on it. Lately there is a guy making a gui (clog) who is doing good work for spreading general lisp love by making it modern. Including tutorials. And there are others too.

The larger problem is building an ecosystem and a stdlib that's written in python, not C. Use ffi or similar instead of C-API.

Also types are self documenting to an extent. Could be helpful for a shared codebase. Again Python just now getting round to adding type definitions.

At the end of the day good tooling/ecosystem and sheer developer hours is more important than what I'm suggesting but it would be nice anyway. I dream about cool programming languages but I stick to boring for work.

The rest of the readme focuses on the delta between Go and Borgo. It doesn't say much about the delta between Borgo and Rust.

I think the delta there is mainly no lifetimes/ownership?

I wonder if Borgo's compiler messages are as nice as Rust's/Gleam's, though.

[1] https://gleam.run/

Ymmv, you might be surprised if you actually bothered to benchmark. Depending on the workload, either JS or erlang can ultimately turn out on top.

They're all optimized to a degree that each has a niche it excells at and leaves the others in the dust.

even with heavily scewed benchmark like techempower fortunes (https://www.techempower.com/benchmarks/#hw=ph&test=fortune&s...) you end up with JS getting ahead of Go with raw requests. And not just slightly, but by 1.5 times the throughput.

In other benchmarks, Golang does indeed win out with similar or even bigger advantages... so the only thing you can ultimately say is ... that it depends. Its a different story if you chose other languages though. But JS, Golang and Erlang are all extremely optimized for their ideal usecase.

Scroll down to the "nodejs" entry for a more realistic comparison.

I'm not saying that JS is "just as fast as golang" generally. My argument is specifically that it's optimized to a degree that there are cases in which JS, an interpreted language, does end up on top.

The same applies to erlang and it's optimization for efficient concurrency.

On average you'll likely get better performance with go, but depending on the workload the results can differ

Structure privacy, OTOH, does. Count me in as someone who really enjoys the case-based approach. It’s not the only one which could work, but it does work.

Today I use idiomatic names - MyName in Go, myName in JS/JSON, my_name in SQL. There are many reasons but generally speaking, for me, it’s less effort and code is more readable.

Curious what your rationale is?

Snake case is not idiomatic for xml, but we still happen to use it for leaf config options.

The main benefit is reducing ambiguity to what maps to what across files. Ease of grepability is also an advantage.

I had a YAML file using `some_property_name`, which was turned into `SomePropertyName`, and it's a small annoyance. It's not a huge deal, but it adds friction where some languages have none. (Or alternately, getting reordered in a separate system like `property_name_some`.)

Especially with database code, something that's fine in Go, like EmployeeID, ends up being employeeid in SQL. You can use underscores in Go but that can trigger other behaviours. If you mix your own JSON with JSON from other sources, you get inconsistent capitalisation. And so on.

And when you have hundreds or thousands of identifiers like this, it gets really hard to read.

You can of course capitalise in SQL - even though it's not semantic - but that becomes inconsistent, too. And then of course the lifecycles of each of these things can be different, which adds another layer of complexity - maybe you refactor your Go code before you upgrade the database, so you end up with two identifiers anyway.

Ultimately I switched to using idiomatic names everywhere, and I really haven't looked back. The boundaries between these systems tend to be pretty clear, as mentioned by someone else, so finding things shouldn't be hard regardless of what they're named.

It's certainly takes slightly longer to deal with idiomatic names - but you read code way more than you write it, and it's easier to read idiomatic code.

The moment you integrate with a third party your US centre zip_code field is suddenly coming over the wire as postCode. The conversions are going to go on, at least in go I can define all of that conversion with ease in one place.

(Some linguist will point out that Bongo-Bongo has half-capitalization, or half has capitalization).

I don't disagree, the problem I have with it is, I have to pay for that up front and have to factor it into my design immediately. This also combines with the fact that the namespace is very flat with no heirarchy, so, choosing good public names is something I feel like I spend way too much time on.

Go is the only language that causes me to pull out a thesaurus when trying to name methods and struct members. It's kinda maddening. Although, after going through this exercise, I end up with code that reads like English. I just wish I could refactor my way into that state, rather than having to try to nail it up front.

I'm guessing you are talking about something else entirely, like, perhaps, decoding JSON into a struct using reflection and encountering a situation where the field names didn't match? Indeed, implicitness can bite you there. That's true in every language. But, then again, as you prefer explicitness why would you be using that approach in the first place?

I forget what it was, but basically my code wasn’t working the way I thought it should and it was solely due to a lowercased struct field. It happened twice where I spent at least a little while trying to figure it out.

And yeah I would guess that I tried to compile. Would be very dumb if I hadn’t although wouldn’t be the dumbest thing I’ve ever done

Frankly if they insisted on visibility being part of the name, I would have preferred they go with the age-old C++/ancientPHP tradition of using a _ prefix for private members.

The latter will be there in one of the future versions and is in an active design phase, which luckily focuses on tagged-union implementation strategy.

With that said, you can already easily use one of the Option/Result libraries or write your own structure - switching on either is trivial (though you have to sometimes choose between zero-cost-ness and convenience).

It already has struct generics, iterator expressions (LINQ), switch pattern matching, good C interop and easy concurrency primitives (no WaitGroup nonsense, also has Channel). Oh, and also portable SIMD, C pointers and very strong performance in general.

* True as in proper tagged unions with either a tag or another type of discriminant and aliased layout, instead of tag + flattening all constituent parts into a one jumbo struct. Or making it an opaque pointer to a box (like Java does, or C# if you go inheritance/interface route). These don't count. I'm curious about Borgo's lowering strategy for enums, but given Go doesn't have those, I'm not holding my breath and expecting something like F# struct unions at best.

For others wanting to play along at home:

  $ docker run --name net8 --rm -it mcr.microsoft.com/dotnet/sdk:8.0-jammy-arm64v8 bash -c '
  cd /root
  dotnet new -d -o console0 console
  cd console0
  dotnet publish --nologo --self-contained -v d -r osx-arm64 -o console0-darwin-arm64
  sleep 600
  '

  $ docker cp net8:/root/console0/console0-darwin-arm64/console0 ./console0
  $ ./console0
  Killed: 9

    dotnet publish -p:PublishSingleFile=true -p:PublishTrimmed=true -o {folder}

    dotnet publish -p:PublishAot=true -o {folder}

Cross-architecture compilation is, however, supported (but requires the same extra dependencies as e.g. Rust).

If you just want to publish for every target from a single docker container (you can't easily do that with e.g. Rust as noted), then you can go with JIT+single-file using the other command.

Keep in mind that Go makes concessions in order for cross-compile to work, and invested extra engineering effort in that, while .NET makes emitting "canonical" native binaries using specific system environment a priority and also cares a lot about JIT instead (there aren't that many people working on .NET compiler infrastructure, so it effectively punches above its weight bypassing Go and Java and matching C++ if optimized).

Generally, there are three publishing options that each make sense depending on scenario:

JIT + host runtime: by definition portable, includes slim launcher executable for convenience, the platform for which can be specified with e.g. -r osx-arm64[0].

JIT + self-contained runtime: this includes IL assemblies and runtime together, either within a single file or otherwise (so it looks like AOT, just one bin/exe). This requires specifying RID, like in the previous option, for cross-compilation.

AOT: statically linked native binary, cross-OS compilation is not supported officially[1] because macOS is painful in general, and Windows<->Linux/FreeBSD is a configuration nightmare - IL AOT Compiler depends on Clang or MSVC and a native linker so it is subject to restrictions of those as a start. But it can be done and there are alternate, more focused toolchains, that offer it, like Bflat[1].

If you just want a hello world AOT application, then the shortest path to that is `dotnet new console --aot && dotnet publish -o {folder}`. Otherwise, the options above are selected based on the needs either via build properties or CLI arguments. I don't know which use case you have - let me know if you have something specific in mind ("Just like in Go" may or may not be optimal choice depending on scenario).

[0] https://learn.microsoft.com/en-us/dotnet/core/rid-catalog

[1] https://github.com/bflattened/bflat (can also build UEFI binaries, lol)

- A proposal for sum types by a Go team member: https://github.com/golang/go/issues/57644

- The community proposal with some comments from the Go team: https://github.com/golang/go/issues/19412

Here are some excerpts from the latest Go survey [1]:

- "The top responses in the closed-form were learning how to write Go effectively (15%) and the verbosity of error handling (13%)."

- "The most common response mentioned Go’s type system, and often asked specifically for enums, option types, or sum types in Go."

I think the problem is not the lack of will on the part of the Go team, but rather that these issues are not easy to fix in a way that fits the language and doesn't cause too many issues with backwards compatibility.

[1]: https://go.dev/blog/survey2024-h1-results

One of the proposals you linked has been raised in 2017 and it is still open with "No one assigned"; same fate for the other item. That doesn't inspire confidence in terms of Go team treating these things as top priority.

I think stuff developers are moaning about the most should be top priority but I guess that is just my simpleton thinking.

> I think the problem is not the lack of will on the part of the Go team, but rather that these issues are not easy to fix in a way that fits the language and doesn't cause too many issues with backwards compatibility.

They have made many changes to the language, some significant ones like Generics (which I would assume was also not an easy problem to solve) while they have largely left the elephant in the room unaddressed i.e. error handling - and the developers deal with that one on a daily basis and I would wager a lot more frequently compared to Generics. If I had to gauge their priority, I would go by where they they are putting their money instead of surveys and proposals. And their priorities seem to be different from what the populace is asking. And that is my point.

Sometimes I think we focus too much and formalize on the first order tooling we use, language being one of them, while we neglect the layers upon layers of abstractions built on top of them. I wonder whether a meta-language could exist that would be useful in these upper layers. Not a framework that imposes its own logic. More of a DSL that can capture both business logic and architecture.

There aren't any such techniques. If one were to create an academic discipline to study "transpilers" and one to study "compilers", all you'd end up with is an identical bunch of techniques and analyses with different names on them. (A thing that sometimes happens when diverse disciplines study what turns out to be the same thing; see machine learning versus statistics for a well-known example.)

Even "compiling" to machine code isn't special anymore, because CPUs don't actually execute assembly anymore. They themselves compile assembly into internal micro-ops, which is what they actually execute. So compilers don't even compile "machine language" anymore; it's just another target. This also dodges "is it a 'compiler' or a 'transpiler' if it targets WASM?", which is good, because there is no value in that question on any level.

2. Transpilers don’t typically optimize code, leaving those transformations to the compiler of the target language.

3. Compilers will typically have an internal representation (SSA) which they operate on to optimize. Transpilers typically operate on the AST (because they don’t need to do any but the most trivial optimizations).

There are exceptions to the rules but these cover the majority of the reasons on why people make the distinction.

The average hobby compiler—regardless of target—doesn't optimize code and works directly on the AST because that's simple to get started with. Most hobby compilers also target some other language rather than LLVM or machine code because that's simple to get started with, so the result is that most transpilers are hobby projects that don't optimize. But there's no reason why a transpiler shouldn't include optimization steps that adapt the output to use code paths that are known to be fast, and a production-grade transpiler typically will include these steps.

You have provided some defining properties that might allow for distinction, but you have not given any reasons for why people make a distinction.

But perhaps we can suss it out. Given the statement "Borgo compiles to Go", what important information is lost that would be saved if "Borgo transpiles to Go" was used instead?

In the statement "XYZ is a compiler/transpiler", it does. It doesn't hurt to have a word that is more specific than others. Otherwise we should just refer to compilers as an "app" :)

Okay. What important information is lost in "XYZ is a compiler" that would be gained in "XYZ is a transpiler"?

> It doesn't hurt to have a word that is more specific than others.

It can if the intent is not properly understood. And so far I'm not sure we do have that understanding.

It reminds me of how my 5-year-old son always corrects me when I tell him to get in the car—"you mean the van!". I have tried to explain to him that a minivan is a kind of car, and he's just about getting it, but it's been a challenge for him to grasp.

This thread chain is in response to jerf's comment "transpiler shouldn't be a word" (simplifying his comment for brevity's sake)

> The word "transpiler" propagates the misunderstanding that there is something special about a compiler that emits machine code, that requires some special "compiler" techniques for special "compiler" purposes that are not necessary for "transpiler" purposes because "transpiling" requires a completely different set of techniques.

In context of the parent comment I read this to be a reaction to someone insisting that we use "transpiler" instead of "compiler"—more an observation of what is happening here than a call to stop using the word altogether.

But if you dig in, it will turn out that every single one of them has a different interpretation, quite often fatally so to whatever the task at hand is.

I mention this because my impression has been that the distinction between "transpiler" and "compiler" is that the latter is into some machine code and the former is not. I think if we could get people to sit down and very clearly define the difference we'd discover it is not as universal a definition as we think.

My personal favorite is when I say a particular term is not well defined on the internet, and I get multiple commenters to jump up and tell me off about how wrong I am and how well-defined the term is and how universal the understanding is, while each of them gives a completely different definition. As I write this it hasn't happened in this thread yet, but stay tuned.

Anyhow, the simple solution is, there isn't a useful distinction between them. There's no sharp line anyhow. Plenty of "transpilers" produce things like Python that looks like

    def f000000001_bch(a0023, a0024, bf___102893):
        __a1 = f000000248_BCh1(a0024, const_00012)
        if __c_112__0:
            f0000000923(__a1)
        else:
            f0000000082(__a1)

It seems the term is not terribly useful even on its own terms... it is not as well defined as everyone thinks.

Ultimately, "compiler" isn't a bright shining line either... I can take anything and shade it down to the point where you might not be sure ("is that a 'compiler' or an 'interpreter'?"), but the "transpiler" term is trying to draw a line where there isn't even a seam in the landscape.

Something like "Assuming all concepts are universal to one's own peculiar definition"

Maybe "semantic egocentrism" could fit the bill?

I love the "draw me a tree" idea of a Platonic form, we all have an idealized model of what that is, that is uniquely our own. With that in mind isnt everything subject to some sort of semantics?

No, a transpiler emits code that another system is meant to understand (often another compiler or interpreter). Whether a human can understand it or not is immaterial to the objective of transpiling.

Bizarre take. No, compilers in the classical sense target byte code and machine code which is meant to be interpreted by a byte code interpreter or a hardware machine.

> Or are you saying that 'transpile' is no more than another word for 'compile'?

Yes. Compilers translate from one language to another. Transpilers translate from one language to another. Both have the objective of preserving the behavior of the program across translation. Neither has the objective of making something intended for humans as a general rule.

That transpiled code (if we draw a distinction) targets languages meant for humans to read/write means that many transpiled programs can be read by people, but it's not the objective.

Bizarre in what way? If compilers are somehow different, then they mustn't target systems, as that's what your previous comment says transpilers do. Which leaves even your own classical definition to be contradictory, if they are somehow different. What does that leave?

> Yes.

But it seems you do not consider them different, which was the divergent path in the previous comment. But evaluating both the "if" and the "else" statement is rather illogical. The evaluation of both branches is what is truly bizarre here.

What possible compiler target couldn't you write? Compilers are not exactly magic.

The difference (to the parent comment) in my eyes is that the target language is the thing intended for humans, not the target output itself. As another commenter points out, transpiled code is often not intended for humans, even if the language is.

If it's not clear, these are just my opinions. Not an attempt at an objective fact or anything.

Like, as in flipping toggle switches? Isn't that just an input device? They were still producing machine code from that switching.

What is meaningful about such a distinction?

The middles all look very similar, though, which is where the heart of "compiler" comes from; that process of some sort of parsing and then transforming down to some other representation. This has a distinguishing set of characteristics and problems despite what frontends and backends get slapped on them.

> And it's written in rust. Kinda unholy.

Agreed, it's like, do you really hate writing Go so much that you'll really write all that Rust to get out of it? Haha. Reminds me of the web frameworks which generated the JS so you didn't have to touch it, like GWT of old.

I'm sure it was a fun exercise to create Borgo, though.

My favorite transpiler is Haxe. It targets so many other languages, the surface area is impressive.

https://haxe.org/

Tell us more about this compiler subset that does not produce binary output. What do they produce? Ternary? Qubits? Nothing?