Are large slices more expensive than smaller ones?

Programmers have a tendency to be superstitious. Particularly, when a programmer hears that copies are expensive, they start to see them everywhere, especially when they learn that, in Go, every assignment is a copy.

Consider this code; x is three orders of magnitude larger than y, is the assignment of x to a more expensive than the assignment of y to b?

func f() {
       x, y := make([]byte, 9000), make([]byte, 9)
       a := x
       b := y
       // ...

The answer is; no. x and y have the same type, []byte, that is, a slice of bytes. As both variables have the same type, their assignment involves copying the same amount of data. Both assignments have the same cost.

All slices are the same size; three machine words (three uintptrs). The first word in the slice is a pointer to the slice’s backing array, the storage for the slice, the second word is the slice’s length, and the third is the capacity. Assigning one slice variable to another copies just three machine words.

Further reading: Go slices: usage and internals (

The Zen of Go

This article was derived from my GopherCon Israel 2020 presentation. It’s also quite long. If you’d prefer a shorter version, head over to

A recording of the presentation is available on YouTube.

How should I write good code?

Something that I’ve been thinking about a lot recently, when reflecting on the body of my own work, is a common subtitle, how should I write good code? Given nobody actively seeks to write bad code, this leads to the question; how do you know when you’ve written good Go code?

If there’s a continuum between good and bad, how to do we know what the good parts are? What are its properties, its attributes, its hallmarks, its patterns, and its idioms?

Idiomatic Go

Which brings me to idiomatic Go. To say that something is idiomatic is to say that it follows the style of the time. If something is not idiomatic, it is not following the prevailing style. It is unfashionable.

More importantly, to say to someone that their code is not idiomatic does not explain why it’s not idiomatic. Why is this? Like all truths, the answer is found in the dictionary.

idiom (noun): a group of words established by usage as having a meaning not deducible from those of the individual words.

Idioms are hallmarks of shared values. Idiomatic Go is not something you learn from a book, it’s something that you acquire by being part of a community.

My concern with the mantra of idiomatic Go is, in many ways, it can be exclusionary. It’s saying “you can’t sit with us.” After all, isn’t that what we mean when critique of someone’s work as non-idiomatic? They didn’t do It right. It doesn’t look right. It doesn’t follow the style of time.

I offer that idiomatic Go is not a suitable mechanism for teaching how to write good Go code because it is defined, fundamentally, by telling someone they did it wrong. Wouldn’t it be better if the advice we gave didn’t alienate the author right at the point they were most willing to accept it?


Stepping away problematic idioms, what other cultural artefacts do Gophers have? Perhaps we can turn to Rob Pike’s wonderful Go Proverbs. Are these suitable teaching tools? Will these tell newcomers how to write good Go code?

In general, I don’t think so. This is not to dismiss Pike’s work, it is just that the Go Proverbs, like Segoe Kensaku’s original, are observations, not statements of value. Again, the dictionary comes to the rescue:

proverb (noun): a short, well-known pithy saying, stating a general truth or piece of advice.

The goal of the Go Proverbs are to reveal a deeper truth about the design of the language, but how useful is advice like the empty interface says nothing to a novice from a language that doesn’t have structural typing?

It’s important to recognise that, in a growing community, at any time the people learning Go far outnumber those who claim to have mastered the language. Thus proverbs are perhaps not the best teaching tool in this scenario.

Engineering Values

Dan Luu found an old presentation by Mark Lucovsky about the engineering culture of the windows team around the windows NT-windows 2000 timeframe. The reason I mention it is Lukovsky’s description of a culture as a common way of evaluating designs and making tradeoffs.

There are many ways of discussing culture, but with respect to an engineering culture Lucovsky’s description is apt. The central idea is values guide decisions in an unknown design space. The values of the NT team were; portability, reliability, security, and extensibility. Engineering values are, crudely translated, the way things are done around here.

Go’s values

What are the explicit values of Go? What are the core beliefs or philosophy that define the way a Go programmer interprets the world? How are they promulgated? How are they taught? How are they enforced? How do they change over time?

How will you, as a newly minted Go programmer, inculcate the engineering values of Go? Or, how will you, a seasoned Go professional promulgate your values to a future generations? And just so we’re clear, this process of knowledge transfer is not optional. Without new blood and new ideas, our community become myopic and wither.

The values of other languages

To set the scene for what I’m getting at we can look to other languages we see examples of their engineering values.

For example, C++ (and by extension Rust) believe that a programmer should not have to pay for a feature they do not use. If a program does not use some computationally expensive feature of the language, then it shouldn’t be forced to shoulder the cost of that feature. This value extends from the language, to its standard library, and is used as a yardstick for judging the design of all code written in C++.

In Java, and Ruby, and Smalltalk, the core value that everything is an object drives the design of programs around message passing, information hiding, and polymorphism. Designs that shoehorn a procedural style, or even a functional style, into these languages are considered to be wrong–or as Gophers would say, non idiomatic.

Turning to our own community, what are the engineering values that bind Go programmers? Discourse in our community is often fractious, so deriving a set of values from first principles would be a formidable challenge. Consensus is critical, but exponentially more difficult as the number of contributors to the discussion increases. But what if someone had done the hard work for us.

The Zen of Python Go

Several decades ago Tim Peters sat down and penned PEP-20, the Zen of Python. Peters’ attempted to document the engineering values that he saw Guido van Rossum apply in his role as BDFL for Python.

For the remainder of this article, I’m going to look towards the Zen of Python and ask, is there anything that can inform the engineering values of Go programmers?

A good package starts with a good name

Let’s start with something spicy,

“Namespaces are one honking great idea–let’s do more of those!”

The Zen of Python, Item 19

This is pretty unequivocal, Python programmers should use namespaces. Lots of them.

In Go parlance a namespace is a package. I doubt there is any question that grouping things into packages is good for design and potentially reuse. But there might be some confusion, especially if you’re coming with a decade of experience in another language, about the right way to do this.

In Go each package should have a purpose, and the best way to know a package’s purpose is by its name—a noun. A package’s name describes what it provides. So too reinterpret Peters’ words, every Go package should have a single purpose.

This is not a new idea, I’ve been saying this a while, but why should you do this rather than approach where packages are used for fine grained taxonomy? Why, because change.

“Design is the art of arranging code to work today, and be changeable forever.”

Sandi Metz

Change is the name of the game we’re in. What we do as programmers is manage change. When we do that well we call it design, or architecture. When we do it badly we call it technical debt, or legacy code.

If you are writing a program that works perfectly, one time, for one fixed set of inputs then nobody cares if the code is good or bad because ultimately the output of the program is all the business cares about.

But this is never true. Software has bugs, requirements change, inputs change, and very few programs are written solely to be executed once, thus your program will change over time. Maybe it’s you who’ll be tasked with this, more likely it will be someone else, but someone has to change that code. Someone has to maintain that code.

So, how can we make it easy to for programs to change? Interfaces everywhere? Make everything mockable? Pernicious dependency injection? Well, maybe, for some classes of programs, but not many, those techniques will be useful. However, for the majority of programs, designing something to be flexible up front is over engineering.

What if, instead, we take a position that rather than enhancing components, we replace them. Then the best way to know when something needs to be replaced, is when it doesn’t do what it says on the tin.

A good package starts with choosing a good name. Think of your package’s name as an elevator pitch, using just one word, to describe what it provides. When the name no longer matches the requirement, find a replacement.

Simplicity matters

“Simple is better than complex.”

The Zen of Python, Item 3

PEP-20 says simple is better than complex, I couldn’t agree more. A couple of years ago I made this tweet;

My observation, at least at the time, was that I couldn’t think of a language introduced in my life time that didn’t purport to be simple. Each new language offered as a justification, and an enticement, their inherent simplicity. But as I researched, I found that simplicity was not a core value of the many of the languages considered Go’s contemporaries. 1 Maybe this is just a cheap shot, but could it be that either these languages aren’t simple, or they don’t think of themselves as being simple. They don’t consider simplicity to be a core value.

Call me old fashioned, but when did being simple fall out of style? Why does the commercial software development industry continually, gleefully, forget this fundamental truth?

“There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies, and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult.”

C. A. R. Hoare, The Emperor’s Old Clothes, 1980 Turing Award Lecture

Simple does not mean easy, we know that. Often it is more work to make something simple to use, than easy to build.

“Simplicity is prerequisite for reliability.”

Edsger W Dijkstra, EWD498, 18 June 1975

Why should we strive for simplicity? Why is important that Go programs be simple? Simple doesn’t mean crude, it means readable and maintainable. Simple doesn’t mean unsophisticated, it means reliable, relatable, and understandable.

“Controlling complexity is the essence of computer programming.”

Brian W. Kernighan, Software Tools (1976)

Whether Python abides by its mantra of simplicity is a matter for debate, but Go holds simplicity as a core value. I think that we can all agree that when it comes to Go, simple code is preferable to clever code.

Avoid package level state

“Explicit is better than implicit.”

The Zen of Python, Item 2

This is a place where I think Peters’ was more aspirational than factual. Many things in Python are not explicit; decorators, dunder methods, and so on. Without doubt they are powerful, there’s a reason those features exists. Each feature is something someone cared enough about to do the work to implement it, especially the complicated ones. But heavy use of those features makes is harder for the reader to predict the cost of an operation.

The good news is we have a choice, as Go programmers, to choose to make our code explicit. Explicit could mean many things, perhaps you may be thinking explicit is just a nice way of saying bureaucratic and long winded, but that’s a superficial interpretation. It’s a misnomer to focus only on the syntax on the page, to fret about line lengths and DRYing up expressions. The more valuable, in my opinon, place to be explicit are to do with coupling and with state.

Coupling is a measure of the amount one thing depends on another. If two things are tightly coupled, they move together. An action that affects one is directly reflected in another. Imagine a train, each carriage joined–ironically the correct word is coupled–together; where the engine goes, the carriages follow.

Another way to describe coupling is the word cohesion. Cohesion measures how well two things naturally belong together. We talk about a cohesive argument, or a cohesive team; all their parts fit together as if they were designed that way.

Why does coupling matter? Because just like trains, when you need to change a piece of code, all the code that is tightly coupled to it must change. A prime example, someone release a new version of their API and now your code doesn’t compile.

APIs are an unavoidable source of coupling but there are more insidious forms of coupling. Clearly everyone knows that if an API’s signature changes the data passing into and out of that call changes. It’s right there in the signature of the function; I take values of these types and return values of other types. But what if the API passed data another way? What if every time you called this API the result was based on the previous time you called that API even though you didn’t change your parameters.

This is state, and management of state is the problem in computer science.

package counter

var count int

func Increment(n int) int {
        count += n
        return count

Suppose we have this simple counter package. You can call Increment to increment the counter, you can even get the value back if you Increment with a value of zero.

Suppose you had to test this code, how would you reset the counter after each test? Suppose you wanted to run those tests in parallel, could you do it? Now suppose that you wanted to count more than one thing per program, could you do it?

No, of course not. Clearly the answer is to encapsulate the count variable in a type.

package counter

type Counter struct {
        count int

func (c *Counter) Increment(n int) int {
        c.count += n
        return c.count

Now imagine that this problem isn’t restricted to just counters, but your applications main business logic. Can you test it in isolation? Can you test it in parallel? Can you use more than one instance at a time? If the answer those question is no, the reason is package level state.

Avoid package level state. Reduce coupling and spooky action at a distance by providing the dependencies a type needs as fields on that type rather than using package variables.

Plan for failure, not success

“Errors should never pass silently.”

The Zen of Python, Item 10

It’s been said of languages that favour exception handling follow the Samurai principle; return victorious or not at all. In exception based languages functions only return valid results. If they don’t succeed then control flow takes an entirely different path.

Unchecked exceptions are clearly an unsafe model to program in. How can you possibly write code that is robust in the presence of errors when you don’t know which statements could throw an exception? Java tried to make exceptions safer by introducing the notion of a checked exception which, to the best of my knowledge, has not been repeated in another mainstream language. There are plenty of languages which use exceptions but they all, with the singular exception of Java, do so in the unchecked variety.

Obviously Go chose a different path. Go programmers believe that robust programs are composed from pieces that handle the failure cases before they handle the happy path. In the space that Go was designed for; server programs, multi threaded programs, programs that handle input over the network, dealing with unexpected data, timeouts, connection failures and corrupted data must be front and centre of the programmer’s mind if they are to produce robust programs.

“I think that error handling should be explicit, this should be a core value of the language.”

Peter Bourgon, GoTime #91

I want to echo Peter’s assertion, as it was the impetus for this article. I think so much of the success of Go is due to the explicit way errors are handled. Go programmers thinks about the failure case first. We solve the “what if…​” case first. This leads to programs where failures are handled at the point of writing, rather than the point they occur in production.

The verbosity of

if err != nil {
    return err

is outweighed by the value of deliberately handling each failure condition at the point at which they occur. Key to this is the cultural value of handling each and every error explicitly.

Return early rather than nesting deeply

“Flat is better than nested.”

The Zen of Python, Item 5

This is sage advice coming from a language where indentation is the primary form of control flow. How can we interpret this advice in terms of Go? gofmt controls the overall whitespace of a Go program so there’s not thing doing there.

I wrote earlier about package names, and there is probably some advice here about avoiding a complicated package hierarchy. In my experience the more a programmer tries to subdivide and taxonimise their Go codebase the more they risk hitting the dead end that is package import loops.

I think the best application of item 5’s advice is the control flow within a function. Simply put, avoid control flow that requires deep indentation.

“Line of sight is a straight line along which an observer has unobstructed vision.”

May Ryer, Code: Align the happy path to the left edge

Mat Ryer describes this idea as line of sight coding. Light of sight coding means things like:

  • Using guard clauses to return early if a precondition is not met.
  • Placing the successful return statement at the end of the function rather than inside a conditional block.
  • Reducing the overall indentation level of the function by extracting functions and methods.

Key to this advice is the thing that you care about, the thing that the function does, is never in danger of sliding out of sight to the right of your screen. This style has a bonus side effect that you’ll avoid pointless arguments about line lengths on your team.

Every time you indent you add another precondition to the programmers stack, consuming one of their 7 ±2 short term memory slots. Rather than nesting deeply, keep the successful path of the function close to the left hand side of your screen.

If you think it’s slow, prove it with a benchmark

“In the face of ambiguity, refuse the temptation to guess.”

The Zen of Python, Item 12

Programming is based on mathematics and logic, two concepts which rarely involve the element of chance. But there are many things we, as programmers, guess about every day. What does this variable do? What does this parameter do? What happens if I pass nil here? What happens if I call Register twice? There’s actually a lot of guesswork in modern programming, especially when it comes to using libraries you didn’t write.

“APIs should be easy to use and hard to misuse.”

Josh Bloch

One of the best ways I know to help a programmer avoid having to guess is to, when building an API, focus on the default use case. Make it as easy as you can for the caller to do the most common thing. However, I’ve written and talked a lot about API design in the past, so instead my interpretation of item 12 is; don’t guess about performance.

Despite how you may feel about Knuth’s advice, one of the drivers of Go’s success is its efficient execution. You can write efficient programs in Go and thus people will choose Go because of this. There are a lot of misconceptions about performance, so my request is, when you’re looking to performance tune your code or you’re facing some dogmatic advice like defer is slow, CGO is expensive, or always use atomics not mutexes, don’t guess.

Don’t complicate your code because of outdated dogma, and, if you think something is slow, first prove it with a benchmark. Go has excellent benchmarking and profiling tools that come in the distribution for free. Use them to find your bottlenecks.

Before you launch a goroutine, know when it will stop

At this point I think I think I’ve mined the valuable points from PEP-20 and possibly stretched its reinterpretation beyond the point of good taste. I think that’s fine, because although this was a useful rhetorical device, ultimately we are talking about two different languages.

“You type g o, a space, and then a function call. Three keystrokes, you can’t make it much shorter than that. Three keystrokes and you’ve just started a sub process.”

Rob Pike, Simplicity is Complicated, dotGo 2015

The next two suggestions I’ll dedicate to goroutines. Goroutines are the signature feature of the language, our answer for first class concurrency. They are so easy to use, just put the word go in front of the statement and you’ve launched that function asynchronously. It’s so simple, no threads, no stack sizes, no thread pool executors, no ID’s, no tracking completion status.

Goroutines are cheap. Because of the runtime’s ability to multiplex goroutines onto a small pool of threads (which you don’t have to manage), hundreds of thousands, millions of goroutines are easily accommodated. This opens up designs that would be not be practical under competing concurrency models like threads or evented callbacks.

But as cheap as goroutines are, they’re not free. At a minimum there’s a few kilobytes for their stack, which, when you’re getting up into the 10^6 goroutines, does start to add up. This is not to say you shouldn’t use millions of goroutines if that is what the design calls for, but when you do, it’s critical that you keep track of them because 10^6 of anything can consume a non trivial amount of resources in aggregate.

Goroutines are the key to resource ownership in Go. To be useful a goroutine has to do something, and that means it almost always holds reference to, or ownership of, a resource; a lock, a network connection, a buffer with data, the sending end of a channel. While that goroutine is alive, the lock is held, the network connection remains open, the buffer retained and the receivers of the channel will continue to wait for more data.

The simplest way to free those resources is to tie them to the lifetime of the goroutine–when the goroutine exits, the resource has been freed. So while it’s near trivial to start a goroutine, before you write those three letters, g o and a space, make sure you have an answer to these questions:

  • Under what condition will a goroutine stop? Go doesn’t have a way to tell a goroutine to exit. There is no stop or kill function, for good reason. If we cannot command a goroutine to stop, we must instead ask it, politely. Almost always this comes down to a channel operation. Range loops over a channel exit when the channel is closed. A channel will become selectable if it is closed. The signal from one goroutine to another is best expressed as a closed channel.
  • What is required for that condition to arise? If channels are both the vehicle to communicate between goroutines and the mechanism for them to signal completion, the next question to the programmer becomes, who will close the channel, when will that happen?
  • What signal will you use to know the goroutine has stopped? When you signal a goroutine to stop, that stopping will happen at some time in the future relative to the goroutine’s frame of reference. It might happen quickly in terms of human perception, but computers execute billions of instructions every second, and from the point of view of each goroutine, their execution of instructions is unsynchronised. The solution is often to use a channel to signal back or a waitgroup where a fan in approach is needed.

Leave concurrency to the caller

It is likely that in any serious Go program you write there will be concurrency involved. This raises the problem, many of the libraries and code that we write fall into this a one goroutine per connection, or worker pattern. How will you manage the lifetime of those goroutines?

net/http is a prime example. Shutting down the server owning the listening socket is relatively straight forward, but what about a goroutines spawned from that accepting socket? net/http does provide a context object inside the request object which can be used to signal–to code that is listening–that the request should be canceled, thereby terminating the goroutine, however it is less clear how to know when all of these things have been done. It’s one thing to call context.Cancel, its another to know that the cancellation has completed.2

The point I want to make about net/http is that its a counter example to good practice. Because each connection is handled by a goroutine spawned inside the net/http.Server type, the program, living outside the net/http package, does not have an ability to control the goroutines spawned for the accepting socket.

This is an area of design that is still evolving, with efforts like go-kit’s run.Group and the Go team’s ErrGroup which provide a framework to execute, cancel and wait on functions run asynchronously.

The bigger design maxim here is for library writers, or anyone writing code that could be run asynchronously, leave the responsibility of starting to goroutine to your caller. Let the caller choose how they want to start, track, and wait on your functions execution.

Write tests to lock in the behaviour of your package’s API

Perhaps you were hoping to read an article from me where I didn’t rant about testing. Sadly, today is not that day.

Your tests are the contract about what your software does and does not do. Unit tests at the package level should lock in the behaviour of the package’s API. They describe, in code, what the package promises to do. If there is a unit test for each input permutation, you have defined the contract for what the code will do in code, not documentation.

This is a contract you can assert as simply as typing go test. At any stage, you can know with a high degree of confidence, that the behaviour people relied on before your change continues to function after your change.

Tests lock in api behaviour. Any change that adds, modifies or removes a public api must include changes to its tests.

Moderation is a virtue

Go is a simple language, only 25 keywords. In some ways this makes the features that are built into the language stand out. Equally these are the features that the language sells itself on, lightweight concurrency, structural typing.

I think all of us have experienced the confusion that comes from trying to use all of Go’s features at once. Who was so excited to use channels that they used them as much as they could, as often as they could? Personally for me I found the result was hard to test, fragile, and ultimately overcomplicated. Am I alone?

I had the same experience with goroutines, attempting to break the work into tiny units I created a hard to manage hurd of Goroutines and ultimately missed the observation that most of my goroutines were always blocked waiting for their predecessor– the code was ultimately sequential and I had added a lot of complexity for little real world benefit. Who has experienced something like this?

I had the same experience with embedding. Initially I mistook it for inheritance. Then later I recreated the fragile base class problem by composing complicated types, which already had several responsibilities, into more complicated mega types.

This is potentially the least actionable piece of advice, but one I think is important enough to mention. The advice is always the same, all things in moderation, and Go’s features are no exception. If you can, don’t reach for a goroutine, or a channel, or embed a struct, anonymous functions, going overboard with packages, interfaces for everything, instead prefer simpler approach rather than the clever approach.

Maintainability counts

I want to close with one final item from PEP-20,

“Readability Counts.”

The Zen of Python, Item 7

So much has been said, about the importance of readability, not just in Go, but all programming languages. People like me who stand on stages advocating for Go use words like simplicity, readability, clarity, productivity, but ultimately they are all synonyms for one word–maintainability.

The real goal is to write maintainable code. Code that can live on after the original author. Code that can exist not just as a point in time investment, but as a foundation for future value. It’s not that readability doesn’t matter, maintainability matters more.

Go is not a language that optimises for clever one liners. Go is not a language which optimises for the least number of lines in a program. We’re not optimising for the size of the source code on disk, nor how long it takes to type the program into an editor. Rather, we want to optimise our code to be clear to the reader. Because its the reader who’s going to have to maintain this code.

If you’re writing a program for yourself, maybe it only has to run once, or you’re the only person who’ll ever see it, then do what ever works for you. But if this is a piece of software that more than one person will contribute to, or that will be used by people over a long enough time that requirements, features, or the environment it runs in may change, then your goal must be for your program to be maintainable. If software cannot be maintained, then it will be rewritten; and that could be the last time your company will invest in Go.

Can the thing you worked hard to build be maintained after you’re gone? What can you do today to make it easier for someone to maintain your code tomorrow?

Dynamically scoped variables in Go

This is a thought experiment in API design. It starts with the classic Go unit testing idiom:

func TestOpenFile(t *testing.T) {
        f, err := os.Open("notfound")
        if err != nil {

        // ...

What’s the problem with this code? The assertion. if err != nil { ... } is repetitive and in the case where multiple conditions need to be checked, somewhat error prone if the author of the test uses t.Error not t.Fatal, eg:

        f, err := os.Open("notfound")
        if err != nil {
        f.Close() // boom!

What’s the solution? DRY it up, of course, by moving the repetitive assertion logic to a helper:

func TestOpenFile(t *testing.T) {
        f, err := os.Open("notfound")
        check(t, err)

        // ...
func check(t *testing.T, err error) {
       if err != nil {

Using the check helper the code is a little cleaner, and clearer, check the error, and hopefully the indecision between t.Error and t.Fatal has been solved. The downside of abstracting the assertion to a helper function is now you need to pass a testing.T into each and every invocation. Worse, you need to pass a *testing.T to everything that needs to call check, transitively, just in case.

This is ok, I guess, but I will make the observation that the t variable is only needed when the assertion fails — and even in a testing scenario, most of the time, most of the tests pass, so that means reading, and writing, all these t‘s is a constant overhead for the relatively rare occasion that a test fails.

What about if we did something like this instead?

func TestOpenFile(t *testing.T) {
        f, err := os.Open("notfound")
        // ...
func check(err error) {
        if err != nil {

Yeah, that’ll work, but it has a few problems

% go test
--- FAIL: TestOpenFile (0.00s)
panic: open notfound: no such file or directory [recovered]
        panic: open notfound: no such file or directory

goroutine 22 [running]:
        /Users/dfc/go/src/testing/testing.go:874 +0x3a3
panic(0x111b040, 0xc0000866f0)
        /Users/dfc/go/src/runtime/panic.go:679 +0x1b2
        /Users/dfc/src/ +0xa1
testing.tRunner(0xc0000b4400, 0x115ac90)
        /Users/dfc/go/src/testing/testing.go:909 +0xc9
created by testing.(*T).Run
        /Users/dfc/go/src/testing/testing.go:960 +0x350
exit status 2

Let’s start with the good; we didn’t have to pass a testing.T every place we call check, the test fails immediately, and we get a nice message in the panic — albeit twice. But where the assertion failed is hard to see. It occurred on expect_test.go:11 but you’d be forgiven for not knowing that.

So panic isn’t really a good solution, but there’s something in this stack trace that is — can you see it? Here’s a hint,

TestOpenFile has a t value, it was passed to it by tRunner, so there’s a testing.T in memory at address 0xc0000b4400. What if we could get access to that t inside check? Then we could use it to call t.Helper and t.Fatal. Is that possible?

Dynamic scoping

What we want is to be able to access a variable whose declaration is neither global, or local to the function, but somewhere higher in the call stack. This is called dynamic scoping. Go doesn’t support dynamic scoping, but it turns out, for restricted cases, we can fake it. I’ll skip to the chase:

// getT returns the address of the testing.T passed to testing.tRunner
// which called the function which called getT. If testing.tRunner cannot
// be located in the stack, say if getT is not called from the main test
// goroutine, getT returns nil.
func getT() *testing.T {
        var buf [8192]byte
        n := runtime.Stack(buf[:], false)
        sc := bufio.NewScanner(bytes.NewReader(buf[:n]))
        for sc.Scan() {
                var p uintptr
                n, _ := fmt.Sscanf(sc.Text(), "testing.tRunner(%v", &p)
                if n != 1 {
                return (*testing.T)(unsafe.Pointer(p))
        return nil

We know that each Test is called by the testing package in its own goroutine (see the stack trace above). The testing package launches the test via a function called tRunner which takes a *testing.T and a func(*testing.T) to invoke. Thus we grab a stack trace of the current goroutine, scan through it for the line beginning with testing.tRunner — which can only be the testing package as tRunner is a private function — and parse the address of the first parameter, which is a pointer to a testing.T. With a little unsafe we convert the raw pointer back to a *testing.T and we’re done.

If the search fails then it is likely that getT wasn’t called from a Test. This is actually ok because the reason we needed the *testing.T was to call t.Fatal and the testing package already requires that t.Fatal be called from the main test goroutine.

import ""

func TestOpenFile(t *testing.T) {
        f, err := os.Open("notfound")
        // ...

Putting it all together we’ve eliminated the assertion boilerplate and possibly made the expectation of the test a little clearer to read, after opening the file err is expected to be nil.

Is this fine?

At this point you should be asking, is this fine? And the answer is, no, this is not fine. You should be screaming internally at this point. But it’s probably worth introspecting those feelings of revulsion.

Apart from the inherent fragility of scrobbling around in a goroutine’s call stack, there are some serious design issues:

  1. The expect.Nil‘s behaviour now depends on who called it. Provided with the same arguments it may have different behaviour depending on where it appears in the call stack — this is unexpected.
  2. Taken to the extreme dynamic scoping effective brings into the scope of a single function all the variables passed into any function that preceded it. It is a side channel for passing data in to and out of functions that is not explicitly documented in function declaration.

Ironically these are precisely the critiques I have of context.Context. I’ll leave it to you to decide if they are justified.

A final word

This is a bad idea, no argument there. This is not a pattern you should ever use in production code. But, this isn’t production code, it’s a test, and perhaps there are different rules that apply to test code. After all, we use mocks, and stubs, and monkey patching, and type assertions, and reflection, and helper functions, and build flags, and global variables, all so we can test our code effectively. None of those, uh, hacks will ever show up in the production code path, so is it really the end of the world?

If you’ve read this far perhaps you’ll agree with me that as unconventional as this approach is, not having to pass a *testing.T into every function that could possibly need to assert something transitively, makes for clearer test code.

So maybe, in this case, the ends do justify the means.

If you’re interested, I’ve put together a small assertion library using this pattern. Caveat emptor.

Complementary engineering indicators

Last year I had the opportunity to watch Cat Swetel’s presentation The Development Metrics You Should Use (but Don’t). The information that could be gleaned from just tracking the start and finish date of work items was eye opening. If you’re using an issue tracker this information is probably already (perhaps with some light data munging) available — no need for TPS reports. Additionally, statistics obtained by data mining your project’s issue tracker are, perhaps, less likely to be juked.

Around the time I saw Cat’s presentation I finished reading Andy Grove’s High Output Management. The hidden gem in this book (assuming becoming a meeting powerhouse isn’t your bag) was Grove’s notion of indicator pairs. An example of a paired indicator might be the number of sales deals closed paired with the customer retention rate. The underling principle being optimising for one indicator will have an adverse impact on the other. In the example, overly aggressive or deceptive tactics could superficially raise the number of sales made, but would be reflected in a dip in the retention rate as customers returned the product or terminated their service prematurely.

These ideas lead me to thinking about indicators you could use for a team delivering a software product. Could those indicators be derived cheaply from the hand to hand combat of software delivery? Could they be structured in a way that aggressively pursuing one metric would be reflected negatively in another? I think so.

These are the three metrics that I’ve been using to track the health of the project that I lead.

  • Date; was the software done when we said it would be done. If you prefer this indicator as a scalar, how many days difference is there between the ship date agreed on at the start of the sprint/milestone/whatever and what was the actual date that you considered it done.
  • Completeness; when the software is done, how many of the things we said we’re going to do actually got delivered in that release.
  • Defects reported; once the software is in the field, what is the rate of bugs reported.

It is relatively easy, for example, to hit a delivery date if you aggressively descope anything risky or simply don’t do it. But in doing so this lack of promised functionality would impact the completeness metric.

Conversely, it’s straight forward to hit your milestone’s completeness target if you let the release date slip and slip. Bringing both the metics into line requires good estimation skills to judge how much can be attempted in milestone and provide direct feedback if your estimation skills needed work.

The third indicator, defects reported in the field, acts as a check on the other two. It would be easy to consistent hit your delivery date with 100% feature completion if your team does a shoddy job. The high fives and :tada: emojis will be short lived if each release brings with it a swathe of high priority bug reports. This indicator also tends to have a second order effect, rushed features to meet a deadline tend to generate remedial work in the following milestones, crowding out promised work or blowing later deadlines.

I consider these to be complementary metrics, they should be considered together, as a group, rather than individually. Ideally your team should be delivering what you promised, when you promised it, with a low defect rate. But more importantly, if that isn’t the case, if one of the indicators is unhealthy, addressing it shouldn’t result in the problem moving to another.

Use internal packages to reduce your public API surface

In the beginning, before the go tool, before Go 1.0, the Go distribution stored the standard library in a subdirectory called pkg/ and the commands which built upon it in cmd/. This wasn’t so much a deliberate taxonomy but a by product of the original make based build system. In September 2014, the Go distribution dropped the pkg/ subdirectory, but then this tribal knowledge had set root in large Go projects and continues to this day.

I tend to view empty directories inside a Go project with suspicion. Often they are a hint that the module’s author may be trying to create a taxonomy of packages rather than ensuring each package’s name, and thus its enclosing directory, uniquely describes its purpose. While the symmetry with cmd/ for package main commands is appealing, a directory that exists only to hold other packages is a potential design smell.

More importantly, the boilerplate of an empty pkg/ directory distracts from the more useful idiom of an internal/ directory. internal/ is a special directory name recognised by the go tool which will prevent one package from being imported by another unless both share a common ancestor. Packages within an internal/ directory are therefore said to be internal packages.

To create an internal package, place it within a directory named internal/. When the go command sees an import of a package with internal/ in the import path, it verifies that the importing package is within the tree rooted at the parent of the internal/ directory.

For example, a package /a/b/c/internal/d/e/f can only be imported by code in the directory tree rooted at /a/b/c. It cannot be imported by code in /a/b/g or in any other repository.

If your project contains multiple packages you may find you have some exported symbols which are intended to be used by other packages in your project, but are not intended to be part of your project’s public API. Although Go has limited visibility modifiers–public, exported, symbols and private, non exported, symbols–internal packages provide a useful mechanism for controlling visibility to parts of your project which would otherwise be considered part of its public versioned API.

You can, of course, promote internal packages later if you want to commit to supporting that API; just move them up a directory level or two. The key is this process is opt-in. As the author, internal packages give you control over which symbols in your project’s public API without being forced to glob concepts together into unwieldy mega packages to avoid exporting them.

Be wary of functions which take several parameters of the same type

APIs should be easy to use and hard to misuse.

— Josh Bloch

A good example of a simple looking, but hard to use correctly, API is one which takes two or more parameters of the same type. Let’s compare two function signatures:

func Max(a, b int) int
func CopyFile(to, from string) error

What’s the difference between these functions? Obviously one returns the maximum of two numbers, the other copies a file, but that’s not the important thing.

Max(8, 10) // 10
Max(10, 8) // 10

Max is commutative; the order of its parameters does not matter. The maximum of eight and ten is ten regardless of if I compare eight and ten or ten and eight.

However, this property does not hold true for CopyFile.

CopyFile("/tmp/backup", "")
CopyFile("", "/tmp/backup")

Which one of these statements made a backup of your presentation and which one overwrite your presentation with last week’s version? You can’t tell without consulting the documentation. A code reviewer cannot know if you’ve got the order correct without consulting the documentation.

The general advice is to try to avoid this situation. Just like long parameter lists, indistinct parameter lists are a design smell.

A challenge

When this situation is unavoidable my solution to this class of problem is to introduce a helper type which will be responsible for calling CopyFile correctly.

type Source string

func (src Source) CopyTo(dest string) error {
	return CopyFile(dest, string(src))

func main() {
	var from Source = ""

In this way CopyFile is always called correctly and, given its poor API can possibly be made private, further reducing the likelihood of misuse.

Can you suggest a better solution?

Don’t force allocations on the callers of your API

This is a post about performance. Most of the time when worrying about the performance of a piece of code the overwhelming advice should be (with apologies to Brendan Gregg) don’t worry about it, yet. However there is one area where I counsel developers to think about the performance implications of a design, and that is API design.

Because of the high cost of retrofitting a change to an API’s signature to address performance concerns, it’s worthwhile considering the performance implications of your API’s design on its caller.

A tale of two API designs

Consider these two Read methods:

func (r *Reader) Read(buf []byte) (int, error)
func (r *Reader) Read() ([]byte, error)

The first method takes a []byte buffer and returns the number of bytes read into that buffer and possibly an error that occurred while reading. The second takes no arguments and returns some data as a []byte or an error.

This first method should be familiar to any Go programmer, it’s io.Reader.Read. As ubiquitous as io.Reader is, it’s not the most convenient API to use. Consider for a moment that io.Reader is the only Go interface in widespread use that returns both a result and an error. Meditate on this for a moment. The standard Go idiom, checking the error and iff it is nil is it safe to consult the other return values, does not apply to Read. In fact the caller must do the opposite. First they must record the number of bytes read into the buffer, reslice the buffer, process that data, and only then, consult the error. This is an unusual API for such a common operation and one that frequently catches out newcomers.

A trap for young players?

Why is it so? Why is one of the central APIs in Go’s standard library written like this? A superficial answer might be io.Reader‘s signature is a reflection of the underlying read(2) syscall, which is indeed true, but misses the point of this post.

If we compare the API of io.Reader to our alternative, func Read() ([]byte, error), this API seems easier to use. Each call to Read() will return the data that was read, no need to reslice buffers, no need to remember the special case to do this before checking the error. Yet this is not the signature of io.Reader.Read. Why would one of Go’s most pervasive interfaces choose such an awkward API? The answer, I believe, lies in the performance implications of the APIs signature on the caller.

Consider again our alternative Read function, func Read() ([]byte, error). On each call Read will read some data into a buffer1 and return the buffer to the caller. Where does this buffer come from? Who allocates it? The answer is the buffer is allocated inside Read. Therefore each call to Read is guaranteed to allocate a buffer which would escape to the heap. The more the program reads, the faster it reads data, the more streams of data it reads concurrently, the more pressure it places on the garbage collector.

The standard libraries’ io.Reader.Read forces the caller to supply a buffer because if the caller is concerned with the number of allocations their program is making this is precisely the kind of thing they want to control. Passing a buffer into Read puts the control of the allocations into the caller’s hands. If they aren’t concerned about allocations they can use higher level helpers like ioutil.ReadAll to read the contents into a []byte, or bufio.Scanner to stream the contents instead.

The opposite, starting with a method like our alternative func Read() ([]byte, error) API, prevents callers from pooling or reusing allocations–no amount of helper methods can fix this. As an API author, if the API cannot be changed you’ll be forced to add a second form to your API taking a supplied buffer and reimplementing your original API in terms of the newer form. Consider, for example, io.CopyBuffer. Other examples of retrofitting APIs for performance reasons are the fmt package and the net/http package which drove the introduction of the sync.Pool type precisely because the Go 1 guarantee prevented the APIs of those packages from changing.

If you want to commit to an API for the long run, consider how its design will impact the size and frequency of allocations the caller will have to make to use it.