Internets of Interest #4: Niall Murphy’s Polemic Against On-Call

The interaction between career development and on-call is actually really, really, bad. Bluntly, the profession takes on-call seriously, tries to be good at it, yet it is very very rare for this to be rewarded in any meaningful way. In 11 years at my previous employer, I never saw anyone get promoted for on-call performance.

Not once.

Internets of interest #2: John Ousterhout discusses a Philosophy of Software Design

Ousterhout’s opus is tearing up tech twitter at the moment. But for those outside the North American prime shipping service area, we’re shit out of luck until a digital version is available. Until then, here’s Ousterhout’s Google Tech talk:

Slides: https://platformlab.stanford.edu/Seminar%20Talks/retreat-2017/John%20Ousterhout.pdf
CS190: https://web.stanford.edu/~ouster/cgi-bin/cs190-winter18/index.php

Internets of interest #1: Brian Kernighan on the Elements of Programming Style

“It turns out that style matters in programming for the same reason that it matters in writing. It makes for better reading.”

Douglas Crockford

I stumbled across this old (in internet years) presentation a few weeks ago and it’s been on high rotation since. If you can look past the recording difficulties (and the evacuation siren) this presentation is chock full of sound advice applicable to all programmers.

Source: https://video.ias.edu/PiTP2009-Kernighan

Maybe adding generics to Go IS about syntax after all

This is a short response to the recently announced Go 2 generics draft proposals

Update: This proposal is incomplete. It cannot replace two common use cases. The first is ensuring that several formal parameters are of the same type:

contract comparable(t T) {
t > t
}

func max(type T comparable)(a, b T) T

Here a, and b must be the same parameterised type — my suggestion would only assert that they had at least the same contract.

Secondly the it would not be possible to parameterise the type of return values:

contract viaStrings(t To, f From) {
var x string = f.String()
t.Set(string(""))
}

func SetViaStrings(type To, From viaStrings)(s []From) []To

Thanks to Ian Dawes and Sam Whited for their insight.

Bummer.


My lasting reaction to the Generics proposal is the proliferation of parenthesis in function declarations.

Although several of the Go team suggested that generics would probably be used sparingly, and the additional syntax would only be burden for the writer of the generic code, not the reader, I am sceptical that this long requested feature will be sufficiently niche as to be unnoticed by most Go developers.

It is true that type parameters can be inferred from their arguments, the declaration of generic functions and methods require a clumsy (type parameter declaration in place of the more common <T> syntaxes found in C++ and Java.

The reason for (type, it was explained to me, is Go is designed to be parsed without a symbol table. This rules out both <T> and [T] syntaxes as the parser needs ahead of time what kind of declaration a T is to avoid interpreting the angle or square braces as comparison or indexing operators respectively.

Contract as a superset of interfaces

The astute Roger Peppe quickly identified that contracts represent a superset of interfaces

Any behaviour you can express with an interface, you can do so and more, with a contract.

The remainder of this post are my suggestions for an alternative generic function declaration syntax that avoids add additional parenthesis by leveraging Roger’s observation.

Contract as a kind of type

The earlier Type Functions proposal showed that a type declaration can support a parameter. If this is correct, then the proposed contract declaration could be rewritten from

contract stringer(x T) {
var s string = x.String()
}

to

type stringer(x T) contract {
var s string = x.String()
}

This supports Roger’s observation that a contract is a superset of an interface. type stringer(x T) contract { ... } introduces a new contract type in the same way type stringer interface { ... }introduces a new interface type.

If you buy my argument that a contract is a kind of type is debatable, but if you’re prepared to take it on faith then the remainder of the syntax introduced in the generics proposal could be further simplified.

If contracts are types, use them as types

If a contract is an identifier then we can use a contract anywhere that a built-in type or interface is used. For example

func Stringify(type T stringer)(s []T) (ret []string) {
for _, v := range s {
ret = append(ret, v.String())
} return ret
}

Could be expressed as

func Stringify(s []stringer) (ret []string) {
for _, v := range s {
ret = append(ret, v.String())
} return ret
}

That is, in place of explicitly binding T to the contract stringer only for T to be referenced seven characters later, we bind the formal parameter s to a slice of stringer s directly. The similarity with the way this would previously be done with a stringer interface emphasises Roger’s observation.

1

Unifying unknown type parameters

The first example in the design proposal introduces an unknown type parameter.

func Print(type T)(s []T) {
for _, v := range s {
fmt.Println(v)
}
}

The operations on unknown types are limited, they are in some senses values that can only be read. Again drawing on Roger’s observation above, the syntax could potentially be expressed as:

func Print(s []contract{}) {
for _, v := range s {
fmt.Println(v)
}
}

Or maybe even

type T contract {} func Print(s []T) {
for _, v := range s {
fmt.Println(v)
}
}

In essence the literal contract{} syntax defines an anonymous unknown type analogous to interface{}‘s anonymous interface type.

Conclusion

The great irony is, after years of my bloviation that “adding generics to Go has nothing to do with the syntax”

2

, it turns out that, actually, yes, the syntax is crucial.

Internets of interest #0: The future of Microprocessors

This weekend I’ve been freshening up the introductory material for a workshop that Francesc Campoy and I are teaching at GopherCon this month. As part of my research, these videos have been on high rotation.

The first video by Sophie Wilson, the designer of the first ARM chip from which both the company and the line of RISC microprocessors we know today were born.

The second presentation by John Hennessy of Computer Architecture: A Quantitative Approach fame is a reprisal of the 2017 Turing Award lecture he gave with his co-author David Patterson.

The theme of both presentations is the same; the end of Dennard scaling and the tremendous technical and economic challenges in bringing extreme UV lithography to the commercial processor production will cap the growth in processor performance to 2-3% per year for the foreseeable future.

Source: John Hennessy and David Patterson, Computer Architecture: A Quantitative Approach, 6/e. 2018

Both Wilson and Hennessy see the future of processor design as a gestalt of CPUs, GPUs, DSPs and VLIW architectures. Issue of adapting mainstream imperative programming languages to these architectures remains very much an open question.

Using Go modules with Travis CI

In my previous post I converted httpstat to use Go 1.11’s upcoming module support. In this post I continue to explore integrating Go modules into a continuous integration workflow via Travis CI.

Life in mixed mode

The first scenario is probably the most likely for existing Go projects, a library or application targeting Go 1.10 and Go 1.11. httpstat has an existing CI story–I’m using Travis CI for my examples, if you use something else, please blog about your experience–and I wanted to test against the current and development versions of Go.

GO111MODULE

The straddling of two worlds is best accomplished via the GO111MODULE environment variable. GO111MODULE dictates when the Go module behaviour will be preferred over the Go 1.5-1.10’s vendor/ directory behaviour. In Go 1.11 the Go module behaviour is disabled by default for packages within $GOPATH (this also includes the default $GOPATH introduced in Go1.8). Thus, without additional configuration, Go1.11 inside Travis CI will behave like Go 1.10.

In my previous post I chose the working directory ~/devel/httpstat to ensure I was not working within a $GOPATH workspace. However CI vendors have worked hard to make sure that their CI bots always check out of the branch under test inside a working $GOPATH.

Fortunately there is a simple workaround for this, add env GO111MODULE=on before any go buildor test invocations in your .travis.yml to force Go module behaviour and ignore any vendor/ directories that may be present inside your repo.

1
language: go
go:
- 1.10.x
- master
os:
- linux
- osx
dist: trusty
sudo: false
install: true
script:
- env GO111MODULE=on go build
- env GO111MODULE=on go test

Creating a go.mod on the fly

You’ll note that I didn’t check in the go.mod module manifest I created in my previous post. This was initially an accident on my part, but one that turned out to be beneficial. By not checking in the go.mod file, the source of truth for dependencies remained httpstat’s Gopkg.toml file. When the call to env GO111MODULE=on go build executes on the Travis CI builder, the go tool converts my Gopkg.toml on the fly, then uses it to fetch dependencies before building.

$ env GO111MODULE=on go build
go: creating new go.mod: module github.com/davecheney/httpstat
go: copying requirements from Gopkg.lock
go: finding github.com/fatih/color v1.5.0
go: finding golang.org/x/sys v0.0.0-20170922123423-429f518978ab
go: finding golang.org/x/net v0.0.0-20170922011244-0744d001aa84
go: finding golang.org/x/text v0.0.0-20170915090833-1cbadb444a80
go: finding github.com/mattn/go-colorable v0.0.9
go: finding github.com/mattn/go-isatty v0.0.3
go: downloading github.com/fatih/color v1.5.0
go: downloading github.com/mattn/go-colorable v0.0.9
go: downloading github.com/mattn/go-isatty v0.0.3
go: downloading golang.org/x/net v0.0.0-20170922011244-0744d001aa84
go: downloading golang.org/x/text v0.0.0-20170915090833-1cbadb444a80

If you’re not using a dependency management tool that go mod knows how to convert from this advice may not work for you and you may have to maintain a go.mod manifest in parallel with you previous dependency management solution.

A clean slate

The second option I investigated, but ultimately did not pursue, was to treat the Travis CI builder, like my fresh Ubuntu 18.04 install, as a blank canvas. Rather than working around Travis CI’s attempts to check the branch out inside a working $GOPATH I experimented with treating the build as a C project

2

then invoking gimme directly. This also required me to check in my go.mod file as without Travis’ language: go support, the checkout was not moved into a $GOPATH folder. The latter seems like a reasonable approach if your project doesn’t intend to be compatible with Go 1.10 or earlier.

language: c
os:
- linux
- osx
dist: trusty
sudo: false
install:
- eval "$(curl -sL https://raw.githubusercontent.com/travis-ci/gimme/master/gimme | GIMME_GO_VERSION=master bash)"
script:
- go build
- go test

You can see the output from this branch here.

Sadly when run in this mode gimme is unable to take advantage of the caching provided by the language: go environment and must build Go 1.11 from source, adding three to four minutes delay to the install phase of the build. Once Go 1.11 is released and gimme can source a binary distribution this will hopefully address the setup latency.

Ultimately this option may end up being redundant if GO111MODULE=on becomes the default behaviour in Go 1.12 and the location Travis places the checkout becomes immaterial.

Taking Go modules for a spin

Update: Since this post was written, Go 1.11beta2 has been released. I’ve updated the setup section to reflect this. Russ Cox kindly wrote to me to explain the reasoning behind storing the Go module cache in $GOPATH. I’ve included his response inline.


This weekend I wanted to play with Ubuntu 18.04 on a spare machine. This gave me a perfect excuse to try out the modules feature recently merged into the Go 1.11 development branch.

TL;DR: When Go 1.11 ships you’ll be able to download the tarball and unpack it anywhere you like. When Go 1.11 ships you’ll be able to write Go modules anywhere you like. 

Setup

The recently released Go 1.11beta2 has support for Go modules.

% curl https://dl.google.com/go/go1.11beta2.linux-amd64.tar.gz | \
tar xz --transform=s/^go/go1.11/g
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 169M 100 169M 0 0 23.6M 0 0:00:07 0:00:07 --:--:-- 21.2M
% go1.11/bin/go version go version go1.11beta2 linux/amd64

That’s all you need to do to install Go 1.11beta2. Out of shot, I’ve added $HOME/go1.11/bin to my $PATH.

Kicking the tires

Now we have a version of Go with module support installed, I wanted to try to use it to manage the dependencies for httpstat, a clone of the Python tool of the same name that many collaborators swarmed on to build in late 2016.

To show that Go 1.11 won’t need you to declare a $GOPATH or use a specific directly layout for the location of your project, I’m going to use my favourite directory for source code, ~/devel.

% git clone https://github.com/davecheney/httpstat devel/httpstat
Cloning into 'devel/httpstat'...
remote: Counting objects: 2326, done.
remote: Total 2326 (delta 0), reused 0 (delta 0), pack-reused 2326
Receiving objects: 100% (2326/2326), 8.73 MiB | 830.00 KiB/s, done.
Resolving deltas: 100% (673/673), done.
Checking out files: 100% (1361/1361), done.
% cd devel/httpstat % go mod -init -module github.com/davecheney/httpstat
go: creating new go.mod: module github.com/davecheney/httpstat
go: copying requirements from Gopkg.lock

Nice, go mod -init translated my existing Gopkg.lock file into its own go.mod format.

% cat go.mod
module github.com/davecheney/httpstat
require (
github.com/fatih/color v1.5.0
github.com/mattn/go-colorable v0.0.9
github.com/mattn/go-isatty v0.0.3
golang.org/x/net v0.0.0-20170922011244-0744d001aa84
golang.org/x/sys v0.0.0-20170922123423-429f518978ab
golang.org/x/text v0.0.0-20170915090833-1cbadb444a80
)

Let’s give it a try

% go build
go: finding golang.org/x/net v0.0.0-20170922011244-0744d001aa84
go: finding github.com/mattn/go-colorable v0.0.9
go: finding github.com/mattn/go-isatty v0.0.3
go: finding golang.org/x/sys v0.0.0-20170922123423-429f518978ab
go: finding github.com/fatih/color v1.5.0
go: finding golang.org/x/text v0.0.0-20170915090833-1cbadb444a80
go: downloading github.com/fatih/color v1.5.0
go: downloading github.com/mattn/go-isatty v0.0.3
go: downloading golang.org/x/net v0.0.0-20170922011244-0744d001aa84
go: downloading github.com/mattn/go-colorable v0.0.9
go: downloading golang.org/x/text v0.0.0-20170915090833-1cbadb444a80

Very nice, go build ignored the vendor/ folder in this repository (because we’re outside $GOPATH) and fetched the revisions it needed. Let’s try out the binary and make sure it works.

% ./httpstat golang.org
Connected to 216.58.196.145:443
HTTP/2.0 200 OK
Server: Google Frontend
Alt-Svc: quic=":443"; ma=2592000; v="44,43,39,35"
Cache-Control: private
Content-Type: text/html; charset=utf-8
Date: Sat, 14 Jul 2018 08:20:43 GMT Strict-Transport-Security: max-age=31536000; preload
Vary: Accept-Encoding
X-Cloud-Trace-Context: 323cd59570cc084fed506f7e85d79d9f
Body discarded

Move along, nothing to see here.

Go module source cache

In the previous version of this article I included a footnote mentioning that go get in module mode stored its downloaded source in $GOPATH/src/mod not the cache added in Go 1.10. Russ Cox kindly wrote to me to explain the rational behind this choice and also copied this to a recent thread on golang-dev. For completeness, here is his response:

The build cache ($GOCACHE, defaulting to $HOME/.cache/go-build) is for storing recent compilation results, so that if you need to do that exact compilation again, you can just reuse the file. The build cache holds entries that are like “if you run this exact compiler on these exact inputs. this is the output you’d get.” If the answer is not in the cache, your build uses a little more CPU to run the compiler nstead of reusing the output. But you are guaranteed to be able to run the compiler instead, since you have the exact inputs and the compiler binary (or else you couldn’t even look up the answer in the cache).

The module cache ($GOPATH/src/mod, defaulting to $HOME/go/src/mod) is for storing downloaded source code, so that every build does not redownload the same code and does not require the network or the original code to be available. The module cache holds entries that are like “if you need to download mymodule@v1.2.3, here are the files you’d get.” If the answer is not in the cache, you have to go out to the network. Maybe you don’t have a network right now. Maybe the code has been deleted. It’s not anywhere near guaranteed that you can redownload the sources and also get the same result. Hopefully you can, but it’s not an absolute certainty like for the build cache. (The go.sum file will detect if you get a different answer on re-download, but knowing you got the wrong bits doesn’t help you make progress on actually building your code. Also these paths end up in file-line information in binaries, so they show up in stack traces, and the like and feed into tools like text editors or debuggers that don’t necessarily know how to trigger the right cache refresh.)

Wrap up

You can build Go 1.11 from source right now anywhere you like. You don’t need to set an environment variable or follow a predefined location.

With Go 1.11 and modules you can write your Go modules anywhere you like. You’re no longer forced into having one copy of a project checked out in a specific sub directory of your $GOPATH.

Slices from the ground up

This blog post was inspired by a conversation with a co-worker about using a slice as a stack. The conversation turned into a wider discussion on the way slices work in Go, so I thought it would be useful to write it up.

Arrays

Every discussion of Go’s slice type starts by talking about something that isn’t a slice, namely, Go’s array type. Arrays in Go have two relevant properties:

  1. They have a fixed size; [5]int is both an array of 5 ints and is distinct from [3]int.
  2. They are value types. Consider this example:
    package main
    
    import "fmt"
    
    func main() {
            var a [5]int
            b := a
            b[2] = 7
            fmt.Println(a, b) // prints [0 0 0 0 0] [0 0 7 0 0]
    }

    The statement b := a declares a new variable, b, of type [5]int, and copies the contents of a to b. Updating b has no effect on the contents of a because a and b are independent values.1

Slices

Go’s slice type differs from its array counterpart in two important ways:

  1. Slices do not have a fixed length. A slice’s length is not declared as part of its type, rather it is held within the slice itself and is recoverable with the built-in function len.2
  2. Assigning one slice variable to another does not make a copy of the slices contents. This is because a slice does not directly hold its contents. Instead a slice holds a pointer to its underlying array3 which holds the contents of the slice.

As a result of the second property, two slices can share the same underlying array. Consider these examples:

  1. Slicing a slice:
    package main
    
    import "fmt"
    
    func main() {
            var a = []int{1,2,3,4,5}
            b := a[2:]
            b[0] = 0
            fmt.Println(a, b) // prints [1 2 0 4 5] [0 4 5]
    }

    In this example a and b share the same underlying array–even though b starts at a different offset in that array, and has a different length. Changes to the underlying array via b are thus visible to a.

  2. Passing a slice to a function:
    package main
    
    import "fmt"
    
    func negate(s []int) {
            for i := range s {
                    s[i] = -s[i]
            }
    }
    
    func main() {
            var a = []int{1, 2, 3, 4, 5}
            negate(a)
            fmt.Println(a) // prints [-1 -2 -3 -4 -5]
    }

    In this example a is passed to negateas the formal parameter s. negate iterates over the elements of s, negating their sign. Even though negate does not return a value, or have any way to access the declaration of a in main, the contents of a are modified when passed to negate.

Most programmers have an intuitive understanding of how a Go slice’s underlying array works because it matches how array-like concepts in other languages tend to work. For example, here’s the first example of this section rewritten in Python:

Python 2.7.10 (default, Feb  7 2017, 00:08:15) 
[GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.34)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> a = [1,2,3,4,5]
>>> b = a
>>> b[2] = 0
>>> a
[1, 2, 0, 4, 5]

And also in Ruby:

irb(main):001:0> a = [1,2,3,4,5]
=> [1, 2, 3, 4, 5]
irb(main):002:0> b = a
=> [1, 2, 3, 4, 5]
irb(main):003:0> b[2] = 0
=> 0
irb(main):004:0> a
=> [1, 2, 0, 4, 5]

The same applies to most languages that treat arrays as objects or reference types.4

The slice header value

The magic that makes a slice behave both as a value and a pointer is to understand that a slice is actually a struct type. This is commonly referred to as a slice header after its counterpart in the reflect package. The definition of a slice header looks something like this:

package runtime

type slice struct {
        ptr   unsafe.Pointer
        len   int
        cap   int
}

This is important because unlike map and chan types slices are value types and are copied when assigned or passed as arguments to functions.

To illustrate this, programmers instinctively understand that square‘s formal parameter v is an independent copy of the v declared in main.

package main

import "fmt"

func square(v int) {
        v = v * v
}

func main() {
        v := 3
        square(v)
        fmt.Println(v) // prints 3, not 9
}

So the operation of square on its v has no effect on main‘s v. So too the formal parameter s of double is an independent copy of the slice s declared in mainnot a pointer to main‘s s value.

package main

import "fmt"

func double(s []int) {
        s = append(s, s...)
}

func main() {
        s := []int{1, 2, 3}
        double(s)
        fmt.Println(s, len(s)) // prints [1 2 3] 3
}

The slightly unusual nature of a Go slice variable is it’s passed around as a value, not than a pointer. 90% of the time when you declare a struct in Go, you will pass around a pointer to values of that struct.5 This is quite uncommon, the only other example of passing a struct around as a value I can think of off hand is time.Time.

It is this exceptional behaviour of slices as values, rather than pointers to values, that can confuses Go programmer’s understanding of how slices work. Just remember that any time you assign, subslice, or pass or return, a slice, you’re making a copy of the three fields in the slice header; the pointer to the underlying array, and the current length and capacity.

Putting it all together

I’m going to conclude this post on the example of a slice as a stack that I opened this post with:

package main

import "fmt"

func f(s []string, level int) {
        if level > 5 {
               return
        }
        s = append(s, fmt.Sprint(level))
        f(s, level+1)
        fmt.Println("level:", level, "slice:", s)
}

func main() {
        f(nil, 0)
}

Starting from main we pass a nil slice into f as level 0. Inside f we append to s the current level before incrementing level and recursing. Once level exceeds 5, the calls to f return, printing their current level and the contents of their copy of s.

level: 5 slice: [0 1 2 3 4 5]
level: 4 slice: [0 1 2 3 4]
level: 3 slice: [0 1 2 3]
level: 2 slice: [0 1 2]
level: 1 slice: [0 1]
level: 0 slice: [0]

You can see that at each level the value of s was unaffected by the operation of other callers of f, and that while four underlying arrays were created 6 higher levels of f in the call stack are unaffected by the copy and reallocation of new underlying arrays as a by-product of append.

Further reading

If you want to find out more about how slices work in Go, I recommend these posts from the Go blog:

Notes