Robert Griesemer, Rob Pike and Ken Thompson started sketching the goals for a new language on the white board on September 21, 2007. Within a few days the goals had settled into a plan to do something and a fair idea of what it would be. Design continued part-time in parallel with unrelated work. By January 2008, Ken had started work on a compiler with which to explore ideas; it generated C code as its output. By mid-year the language had become a full-time project and had settled enough to attempt a production compiler. In May 2008, Ian Taylor independently started on a GCC front end for Go using the draft specification. Russ Cox joined in late 2008 and helped move the language and libraries from prototype to reality.
Many others have contributed ideas, discussions, and code.
Go was born out of frustration with existing languages and environments for systems programming. Programming had become too difficult and the choice of languages was partly to blame. One had to choose either efficient compilation, efficient execution, or ease of programming; all three were not available in the same mainstream language. Programmers who could were choosing ease over safety and efficiency by moving to dynamically typed languages such as Python and JavaScript rather than C++ or, to a lesser extent, Java.
Go is an attempt to combine the ease of programming of an interpreted, dynamically typed language with the efficiency and safety of a statically typed, compiled language. It also aims to be modern, with support for networked and multicore computing. Finally, it is intended to be fast: it should take at most a few seconds to build a large executable on a single computer. To meet these goals required addressing a number of linguistic issues: an expressive but lightweight type system; concurrency and garbage collection; rigid dependency specification; and so on. These cannot be addressed well by libraries or tools; a new language was called for.
Go is mostly in the C family (basic syntax), with significant input from the Pascal/Modula/Oberon family (declarations, packages), plus some ideas from languages inspired by Tony Hoare's CSP, such as Newsqueak and Limbo (concurrency). However, it is a new language across the board. In every respect the language was designed by thinking about what programmers do and how to make programming, at least the kind of programming we do, more effective, which means more fun.
Programming today involves too much bookkeeping, repetition, and clerical work. As Dick Gabriel says, “Old programs read like quiet conversations between a well-spoken research worker and a well-studied mechanical colleague, not as a debate with a compiler. Who'd have guessed sophistication bought such noise?” The sophistication is worthwhile—no one wants to go back to the old languages—but can it be more quietly achieved?
Go attempts to reduce the amount of typing in both senses of the word.
Throughout its design, we have tried to reduce clutter and
complexity. There are no forward declarations and no header files;
everything is declared exactly once. Initialization is expressive,
automatic, and easy to use. Syntax is clean and light on keywords.
Stuttering (foo.Foo* myFoo = new(foo.Foo)) is reduced by
simple type derivation using the :=
declare-and-initialize construct. And perhaps most radically, there
is no type hierarchy: types just are, they don't have to
announce their relationships. These simplifications allow Go to be
expressive yet comprehensible without sacrificing, well, sophistication.
Another important principle is to keep the concepts orthogonal. Methods can be implemented for any type; structures represent data while interfaces represent abstraction; and so on. Orthogonality makes it easier to understand what happens when things combine.
Other than declaration syntax, the differences are not major and stem from two desires. First, the syntax should feel light, without too many mandatory keywords, repetition, or arcana. Second, the language has been designed to be easy to analyze and can be parsed without a symbol table. This makes it much easier to build tools such as debuggers, dependency analyzers, automated documentation extractors, IDE plug-ins, and so on. C and its descendants are notoriously difficult in this regard.
They're only backwards if you're used to C. In C, the notion is that a
variable is declared like an expression denoting its type, which is a
nice idea, but the type and expression grammars don't mix very well and
the results can be confusing; consider function pointers. Go mostly
separates expression and type syntax and that simplifies things (using
prefix * for pointers is an exception that proves the rule). In C,
the declaration
int* a, b;
declares a to be a pointer but not b; in Go
var a, b *int;
declares both to be pointers. This is clearer and more regular.
Also, the := short declaration form argues that a full variable
declaration should present the same order as := so
var a uint64 = 1;has the same effect as
a := uint64(1);
Parsing is also simplified by having a distinct grammar for types that
is not just the expression grammar; keywords such as func
and chan keep things clear.
Safety. Without pointer arithmetic it's possible to create a language that can never derive an illegal address that succeeds incorrectly. Compiler and hardware technology have advanced to the point where a loop using array indices can be as efficient as a loop using pointer arithmetic. Also, the lack of pointer arithmetic can simplify the implementation of the garbage collector.
++ and -- statements and not expressions? And why postfix, not prefix?
Without pointer arithmetic, the convenience value of pre- and postfix
increment operators drops. By removing them from the expression
hierarchy altogether, expression syntax is simplified and the messy
issues around order of evaluation of ++ and --
(consider f(i++) and p[i] = q[++i])
are eliminated as well. The simplification is
significant. As for postfix vs. prefix, either would work fine but
the postfix version is more traditional; insistence on prefix arose
with the STL, a library for a language whose name contains, ironically, a
postfix increment.
Go uses brace brackets for statement grouping, a syntax familiar to programmers who have worked with any language in the C family. Semicolons, however, are for parsers, not for people, and we wanted to eliminate them as much as possible. To achieve this goal, Go borrows a trick from BCPL: the semicolons that separate statements are in the formal grammar but are injected automatically, without lookahead, by the lexer at the end of any line that could be the end of a statement. This works very well in practice but has the effect that it forces a brace style. For instance, the opening brace of a function cannot appear on a line by itself.
Some have argued that the lexer should do lookahead to permit the
brace to live on the next line. We disagree. Since Go code is meant
to be formatted automatically by
gofmt,
some style must be chosen. That style may differ from what
you've used in C or Java, but Go is a new language and
gofmt's style is as good as any other. More
important—much more important—the advantages of a single,
programmatically mandated format for all Go programs greatly outweigh
any perceived disadvantages of the particular style.
Note too that Go's style means that an interactive implementation of
Go can use the standard syntax one line at a time without special rules.
One of the biggest sources of bookkeeping in systems programs is memory management. We feel it's critical to eliminate that programmer overhead, and advances in garbage collection technology in the last few years give us confidence that we can implement it with low enough overhead and no significant latency. (The current implementation is a plain mark-and-sweep collector but a replacement is in the works.)
Another point is that a large part of the difficulty of concurrent and multi-threaded programming is memory management; as objects get passed among threads it becomes cumbersome to guarantee they become freed safely. Automatic garbage collection makes concurrent code far easier to write. Of course, implementing garbage collection in a concurrent environment is itself a challenge, but meeting it once rather than in every program helps everyone.
Finally, concurrency aside, garbage collection makes interfaces simpler because they don't need to specify how memory is managed across them.
It was important to us to extend the space of identifiers from the confines of ASCII. Go's rule—identifier characters must be letters or digits as defined by Unicode—is simple to understand and to implement but has restrictions. Combining characters are excluded by design, for instance. Until there is an agreed external definition of what an identifier might be, plus a definition of canonicalization of identifiers that guarantees no ambiguity, it seemed better to keep combining characters out of the mix. Thus we have a simple rule that can be expanded later without breaking programs, one that avoids bugs that would surely arise from a rule that admits ambiguous identifiers.
On a related note, since an exported identifier must begin with an
upper-case letter, identifiers created from “letters”
in some languages can, by definition, not be exported. For now the
only solution is to use something like X日本語, which
is clearly unsatisfactory; we are considering other options. The
case-for-visibility rule is unlikely to change however; it's one
of our favorite features of Go.
Generics may well be added at some point. We don't feel an urgency for them, although we understand some programmers do.
Generics are convenient but they come at a cost in complexity in the type system and run-time. We haven't yet found a design that gives value proportionate to the complexity, although we continue to think about it. Meanwhile, Go's built-in maps and slices, plus the ability to use the empty interface to construct containers (with explicit unboxing) mean in many cases it is possible to write code that does what generics would enable, if less smoothly.
This remains an open issue.
We believe that coupling exceptions to a control
structure, as in the try-catch-finally idiom, results in
convoluted code. It also tends to encourage programmers to label
too many ordinary errors, such as failing to open a file, as
exceptional.
Go takes a different approach. Instead of exceptions, it has a couple of built-in functions to signal and recover from truly exceptional conditions. The recovery mechanism is executed only as part of a function's state being torn down after an error, which is sufficient to handle catastrophe but requires no extra control structures and, when used well, can result in clean error-handling code.
This is answered in the general FAQ.
Object-oriented programming, at least in the best-known languages, involves too much discussion of the relationships between types, relationships that often could be derived automatically. Go takes a different approach.
Rather than requiring the programmer to declare ahead of time that two types are related, in Go a type automatically satisfies any interface that specifies a subset of its methods. Besides reducing the bookkeeping, this approach has real advantages. Types can satisfy many interfaces at once, without the complexities of traditional multiple inheritance. Interfaces can be very lightweight—having one or even zero methods in an interface can express useful concepts. Interfaces can be added after the fact if a new idea comes along or for testing—without annotating the original types. Because there are no explicit relationships between types and interfaces, there is no type hierarchy to manage or discuss.
It's possible to use these ideas to construct something analogous to
type-safe Unix pipes. For instance, see how fmt.Fprintf
enables formatted printing to any output, not just a file, or how the
bufio package can be completely separate from file I/O,
or how the crypto packages stitch together block and
stream ciphers. All these ideas stem from a single interface
(io.Writer) representing a single method
(Write). And that's only scratching the surface.
It takes some getting used to but this implicit style of type dependency is one of the most exciting things about Go.
len a function and not a method?
We debated this issue but decided
implementing len and friends as functions was fine in practice and
didn't complicate questions about the interface (in the Go type sense)
of basic types.
Method dispatch is simplified if it doesn't need to do type matching as well. Experience with other languages told us that having a variety of methods with the same name but different signatures was occasionally useful but that it could also be confusing and fragile in practice. Matching only by name and requiring consistency in the types was a major simplifying decision in Go's type system.
Regarding operator overloading, it seems more a convenience than an absolute requirement. Again, things are simpler without it.
The convenience of automatic conversion between numeric types in C is outweighed by the confusion it causes. When is an expression unsigned? How big is the value? Does it overflow? Is the result portable, independent of the machine on which it executes? It also complicates the compiler; “the usual arithmetic conversions” are not easy to implement and inconsistent across architectures. For reasons of portability, we decided to make things clear and straightforward at the cost of some explicit conversions in the code. The definition of constants in Go—arbitrary precision values free of signedness and size annotations—ameliorates matters considerably, though.
A related detail is that, unlike in C, int and int64
are distinct types even if int is a 64-bit type. The int
type is generic; if you care about how many bits an integer holds, Go
encourages you to be explicit.
The same reason strings are: they are such a powerful and important data structure that providing one excellent implementation with syntactic support makes programming more pleasant. We believe that Go's implementation of maps is strong enough that it will serve for the vast majority of uses. If a specific application can benefit from a custom implementation, it's possible to write one but it will not be as convenient syntactically; this seems a reasonable tradeoff.
Map lookup requires an equality operator, which structs and arrays do not implement. They don't implement equality because equality is not well defined on such types; there are multiple considerations involving shallow vs. deep comparison, pointer vs. value comparison, how to deal with recursive structures, and so on. We may revisit this issue—and implementing equality for structs and arrays will not invalidate any existing programs—but without a clear idea of what equality of structs and arrays should mean, it was simpler to leave it out for now.
There's a lot of history on that topic. Early on, maps and channels were syntactically pointers and it was impossible to declare or use a non-pointer instance. Also, we struggled with how arrays should work. Eventually we decided that the strict separation of pointers and values made the language harder to use. Introducing reference types, including slices to handle the reference form of arrays, resolved these issues. Reference types add some regrettable complexity to the language but they have a large effect on usability: Go became a more productive, comfortable language when they were introduced.
Concurrency and multi-threaded programming have a reputation for difficulty. We believe the problem is due partly to complex designs such as pthreads and partly to overemphasis on low-level details such as mutexes, condition variables, and even memory barriers. Higher-level interfaces enable much simpler code, even if there are still mutexes and such under the covers.
One of the most successful models for providing high-level linguistic support for concurrency comes from Hoare's Communicating Sequential Processes, or CSP. Occam and Erlang are two well known languages that stem from CSP. Go's concurrency primitives derive from a different part of the family tree whose main contribution is the powerful notion of channels as first class objects.
Goroutines are part of making concurrency easy to use. The idea, which has been around for a while, is to multiplex independently executing functions—coroutines, really—onto a set of threads. When a coroutine blocks, such as by calling a blocking system call, the run-time automatically moves other coroutines on the same operating system thread to a different, runnable thread so they won't be blocked. The programmer sees none of this, which is the point. The result, which we call goroutines, can be very cheap: unless they spend a lot of time in long-running system calls, they cost little more than the memory for the stack.
To make the stacks small, Go's run-time uses segmented stacks. A newly minted goroutine is given a few kilobytes, which is almost always enough. When it isn't, the run-time allocates (and frees) extension segments automatically. The overhead averages about three cheap instructions per function call. It is practical to create hundreds of thousands of goroutines in the same address space. If goroutines were just threads, system resources would run out at a much smaller number.
After long discussion it was decided that the typical use of maps did not require safe access from multiple threads, and in those cases where it did, the map was probably part of some larger data structure or computation that was already synchronized. Therefore requiring that all map operations grab a mutex would slow down most programs and add safety to few. This was not an easy decision, however, since it means uncontrolled map access can crash the program.
The language does not preclude atomic map updates. When required, such as when hosting an untrusted program, the implementation could interlock map access.