gomacro - interactive Go interpreter and debugger with generics and macros
gomacro is an almost complete Go interpreter, implemented in pure Go. It offers both
an interactive REPL and a scripting mode, and does not require a Go toolchain at runtime
(except in one very specific case: import of a 3rd party package at runtime).
It has two dependencies beyond the Go standard library:
github.com/peterh/liner
and
golang.org/x/tools/go/packages
Gomacro can be used as:
-
a standalone executable with interactive Go REPL, line editing and code completion:
just run gomacro
from your command line, then type Go code. Example:
$ gomacro
[greeting message...]
gomacro> import "fmt"
gomacro> fmt.Println("hello, world!")
hello, world!
14 // int
<nil> // error
gomacro>
press TAB to autocomplete a word, and press it again to cycle on possible completions.
Line editing follows mostly Emacs: Ctrl+A or Home jumps to start of line,
Ctrl+E or End jumps to end of line, Ald+D deletes word starting at cursor...
For the full list of key bindings, see https://github.com/peterh/liner
-
a tool to experiment with Go generics: see Generics
-
a Go source code debugger: see Debugger
-
an interactive tool to make science more productive and more fun.
If you use compiled Go with scientific libraries (physics, bioinformatics, statistics...)
you can import the same libraries from gomacro REPL (immediate on Linux and Mac OS X,
requires restarting on other platforms,
see Importing packages below), call them interactively,
inspect the results, feed them to other functions/libraries, all in a single session.
The imported libraries will be compiled, not interpreted,
so they will be as fast as in compiled Go.
For a graphical user interface on top of gomacro, see Gophernotes.
It is a Go kernel for Jupyter notebooks and nteract, and uses gomacro for Go code evaluation.
-
a library that adds Eval() and scripting capabilities to your Go programs in few lines
of code:
package main
import (
"fmt"
"reflect"
"github.com/lifepod-solutions/gomacro/fast"
)
func RunGomacro(toeval string) reflect.Value {
interp := fast.New()
vals, _ := interp.Eval(toeval)
// for simplicity, only use the first returned value
return vals[0].ReflectValue()
}
func main() {
fmt.Println(RunGomacro("1+1"))
}
Also, github issue #13 explains
how to have your application's functions, variable, constants and types
available in the interpreter.
Note: gomacro license is MPL 2.0, which imposes some restrictions
on programs that use gomacro.
See MPL 2.0 FAQ for common questions
regarding the license terms and conditions.
-
a way to execute Go source code on-the-fly without a Go compiler:
you can either run gomacro FILENAME.go
(works on every supported platform)
or you can insert a line #!/usr/bin/env gomacro
at the beginning of a Go source file,
then mark the file as executable with chmod +x FILENAME.go
and finally execute it
with ./FILENAME.go
(works only on Unix-like systems: Linux, *BSD, Mac OS X ...)
-
a Go code generation tool:
gomacro was started as an experiment to add Lisp-like macros to Go, and they are
extremely useful (in the author's opinion) to simplify code generation.
Macros are normal Go functions, they are special only in one aspect:
they are executed before compiling code, and their input and output is code
(abstract syntax trees, in the form of go/ast.Node)
Don't confuse them with C preprocessor macros: in Lisp, Scheme and now in Go,
macros are regular functions written in the same programming language
as the rest of the source code. They can perform arbitrary computations
and call any other function or library: they can even read and write files,
open network connections, etc... as a normal Go function can do.
See doc/code_generation.pdf
for an introduction to the topic.
Installation
Prerequites
Gomacro is pure Go, and in theory it should work on any platform supported by the Go compiler.
The following combinations are tested and known to work:
- Linux: amd64, 386, arm64, arm, mips, ppc64le
- Mac OS X: amd64, 386 (386 binaries running on amd64 system)
- Windows: amd64, 386
- FreeBSD: amd64, 386
- Android: arm64, arm (tested with Termux and the Go compiler distributed with it)
How to install
The command
go get -u github.com/lifepod-solutions/gomacro
downloads, compiles and installs gomacro and its dependencies
Current Status
Almost complete.
The main limitations and missing features are:
- importing 3rd party libraries at runtime currently only works on Linux and Mac OS X.
On other systems as Windows, Android and *BSD it is cumbersome and requires recompiling - see Importing packages.
- conversions from/to unsafe.Pointer are not supported
- some corner cases using interpreted interfaces, as interface -> interface type assertions and type switches, are not implemented yet.
- some corner cases using recursive types may not work correctly.
- goto can only jump backward, not forward
- out-of-order code is under testing - some corner cases, as for example out-of-order declarations
used in keys of composite literals, are not supported.
Clearly, at REPL code is still executed as soon as possible, so it makes a difference mostly
if you separate multiple declarations with ; on a single line. Example:
var a = b; var b = 42
Support for "batch mode" is in progress - it reads as much source code as possible before executing it,
and it's useful mostly to execute whole files or directories.
The documentation also contains the full list of features and limitations
Extensions
Compared to compiled Go, gomacro supports several extensions:
-
generics (experimental) - see Generics
-
an integrated debugger, see Debugger
-
configurable special commands. Type :help
at REPL to list them,
and see cmd.go:37
for the documentation and API to define new ones.
-
untyped constants can be manipulated directly at REPL. Examples:
gomacro> 1<<100
{int 1267650600228229401496703205376} // untyped.Lit
gomacro> const c = 1<<100; c * c / 100000000000
{int 16069380442589902755419620923411626025222029937827} // untyped.Lit
This provides a handy arbitrary-precision calculator.
Note: operations on large untyped integer constants are always exact,
while operations on large untyped float constants are implemented with go/constant.Value
,
and are exact as long as both numerator and denominator are <= 5e1232.
Beyond that, go/constant.Value
switches from *big.Rat
to *big.Float
with precision = 512, which can accumulate rounding errors.
If you need exact results, convert the untyped float constant to *big.Rat
(see next item) before exceeding 5e1232.
-
untyped constants can be converted implicitly to *big.Int
, *big.Rat
and *big.Float
. Examples:
import "math/big"
var i *big.Int = 1<<1000 // exact - would overflow int
var r *big.Rat = 1.000000000000000000001 // exact - different from 1.0
var s *big.Rat = 5e1232 // exact - would overflow float64
var t *big.Rat = 1e1234 // approximate, exceeds 5e1232
var f *big.Float = 1e646456992 // largest untyped float constant that is different from +Inf
Note: every time such a conversion is evaluated, it creates a new value - no risk to modify the constant.
Be aware that converting a huge value to string, as typing f
at REPL would do, can be very slow.
-
zero value constructors: for any type T
, the expression T()
returns the zero value of the type
-
macros, quoting and quasiquoting: see
doc/code_generation.pdf
and slightly relaxed checks:
- unused variables and unused return values never cause errors
Examples
Some short, notable examples - to run them on non-Linux platforms, see Importing packages first.
plot mathematical functions
- install libraries:
go get gonum.org/v1/plot gonum.org/v1/plot/plotter gonum.org/v1/plot/vg
- start the interpreter:
gomacro
- at interpreter prompt, paste the whole Go code listed at https://github.com/gonum/plot/wiki/Example-plots#functions
(the source code starts after the picture under the section "Functions", and ends just before the section "Histograms")
- still at interpreter prompt, enter
main()
If all goes well, it will create a file named "functions.png" in current directory containing the plotted functions.
simple mandelbrot web server
- install libraries:
go get github.com/sverrirab/mandelbrot-go
- chdir to mandelbrot-go source folder:
cd; cd go/src/github.com/sverrirab/mandelbrot-go
- start interpreter with arguments:
gomacro -i mbrot.go
- at interpreter prompt, enter
init(); main()
- visit http://localhost:8090/
Be patient, rendering and zooming mandelbrot set with an interpreter is a little slow.
Further examples are listed by Gophernotes
Importing packages
Gomacro supports the standard Go syntax import
, including package renaming. Examples:
import "fmt"
import (
"io"
"net/http"
r "reflect"
)
Third party packages - i.e. packages not in Go standard library - can also be imported
with the same syntax, as long as the package is already installed.
To install a package, follow its installation procedure: quite often it is the command go get PACKAGE-PATH
The next steps depend on the system you are running gomacro on:
Linux and Mac OS X
If you are running gomacro on Linux or Mac OS X, import
will then just work:
it will automatically download, compile and import a package. Example:
$ gomacro
[greeting message...]
gomacro> import "gonum.org/v1/plot"
// debug: looking for package "gonum.org/v1/plot" ...
// debug: compiling "/home/max/go/src/gomacro.imports/gonum.org/v1/plot/plot.go" ...
go: finding module for package gonum.org/v1/plot/vg/draw
go: finding module for package gonum.org/v1/plot
go: found gonum.org/v1/plot in gonum.org/v1/plot v0.0.0-20200226011204-b25252b0d522
gomacro> plot.New()
&{...} // *plot.Plot
<nil> // error
Note: internally, gomacro will compile and load a Go plugin containing the package's exported declarations.
Go plugins are currently supported only on Linux and Mac OS X.
WARNING On Mac OS X, never execute strip gomacro
: it breaks plugin support,
and loading third party packages stops working.
Other systems
On all other systems as Windows, Android and *BSD you can still use import
,
but there are more steps: you need to manually download the package,
and you also need to recompile gomacro after the import
(it will tell you).
Example:
$ go get gonum.org/v1/plot
$ gomacro
[greeting message...]
gomacro> import "gonum.org/v1/plot"
// warning: created file "/home/max/go/src/github.com/lifepod-solutions/gomacro/imports/thirdparty/gonum_org_v1_plot.go", recompile gomacro to use it
Now quit gomacro, recompile and reinstall it:
gomacro> :quit
$ go install github.com/lifepod-solutions/gomacro
Finally restart it. Your import is now linked inside gomacro and will work:
$ gomacro
[greeting message...]
gomacro> import "gonum.org/v1/plot"
gomacro> plot.New()
&{...} // *plot.Plot
<nil> // error
Note: if you need several packages, you can first import
all of them,
then quit and recompile gomacro only once.
Generics
gomacro contains two alternative, experimental versions of Go generics:
-
the first version is modeled after C++ templates, and is appropriately named "C++ style"
See doc/generics-c++.md for how to enable and use them.
-
the second version is named "contracts are interfaces" - or more briefly "CTI".
It is modeled after several published proposals for Go generics,
most notably Ian Lance Taylor's Type Parameters in Go
It has some additions inspired from Haskell generics
and original contributions from the author - in particular to create a simpler alternative to
Go 2 contracts
For their design document and reasoning behind some of the design choices, see doc/generics-cti.md
The second version of generics "CTI" is enabled by default in gomacro.
They are in beta status, and at the moment only generic types and functions are supported.
Syntax and examples:
// declare a generic type with two type arguments T and U
type Pair#[T,U] struct {
First T
Second U
}
// instantiate the generic type using explicit types for T and U,
// and create a variable of such type.
var pair Pair#[complex64, struct{}]
// equivalent:
pair := Pair#[complex64, struct{}] {}
// a more complex example, showing higher-order functions
func Transform#[T,U](slice []T, trans func(T) U) []U {
ret := make([]U, len(slice))
for i := range slice {
ret[i] = trans(slice[i])
}
return ret
}
Transform#[string,int] // returns func([]string, func(string) int) []int
// returns []int{3, 2, 1} i.e. the len() of each string in input slice:
Transform#[string,int]([]string{"abc","xy","z"}, func(s string) int { return len(s) })
Contracts specify the available methods of a generic type.
For simplicity, they do not introduce a new syntax or new language concepts:
contracts are just (generic) interfaces.
With a tiny addition, actually: the ability to optionally indicate the receiver type.
For example, the contract specifying that values of type T
can be compared with each other
to determine if the first is less, equal or greater than the second is:
type Comparable#[T] interface {
// returns -1 if a is less than b
// returns 0 if a is equal to b
// returns 1 if a is greater than b
func (a T) Cmp(b T) int
}
A type T
implements Comparable#[T]
if it has a method func (T) Cmp(T) int
.
This interface is carefully chosen to match the existing methods of
*math/big.Float
, *math/big.Int
and *math/big.Rat
.
In other words, *math/big.Float
, *math/big.Int
and *math/big.Rat
already implement it.
What about basic types as int8
, int16
, int32
, uint
... float*
, complex*
... ?
Gomacro extends them, automatically adding many methods equivalent to the ones declared
on *math/big.Int
to perform arithmetic and comparison, including Cmp
which is
internally defined as (no need to define it yourself):
func (a int) Cmp(b int) int {
if a < b {
return -1
} else if a > b {
return 1
} else {
return 0
}
}
Thus the generic functions Min
and Max
can be written as
func Min#[T: Comparable] (a, b T) T {
if a.Cmp(b) < 0 { // also <= would work
return a
}
return b
}
func Max#[T: Comparable] (a, b T) T {
if a.Cmp(b) > 0 { // also >= would work
return a
}
return b
}
Where the syntax #[T: Comparable]
or equivalently #[T: Comparable#[T]]
indicates that T
must satisfy the contract (implement the interface) Comparable#[T]
Such functions Min
and Max
will then work automatically for every type T
that satisfies the contract (implements the interface) Comparable#[T]
:
all basic integers and floats, plus *math/big.Float
, *math/big.Int
and *math/big.Rat
,
plus every user-defined type T
that has a method func (T) Cmp(T) int
If you do not specify the contract(s) that a type must satisfy, generic functions
cannot access the fields and methods of a such type, which is then treated
as a "black box", similarly to interface{}
.
Two values of type T
can be added if T
has an appropriate method.
But which name and signature should we choose to add values?
Copying again from math/big
, the method we choose is func (T) Add(T,T) T
If receiver is a pointer, it will be set to the result - in any case,
the result will also be returned.
Similarly to Comparable
, the contract Addable
is then
type Addable#[T] interface {
// Add two values a, b and return the result.
// If recv is a pointer, it must be non-nil
// and it will be set to the result
func (recv T) Add(a, b T) T
}
With such a contract, a generic function Sum
is quite straightforward:
func Sum#[T: Addable] (args ...T) T {
// to create the zero value of T,
// one can write 'var sum T' or equivalently 'sum := T()'
// Unluckily, that's not enough for math/big numbers, which require
// the receiver of method calls to be created with a function `New()`
// Once math/big numbers have such method, the following
// will be fully general - currently it works only on basic types.
sum := T().New()
for _, elem := range args {
// use the method T.Add(T, T)
//
// as an optimization, relevant at least for math/big numbers,
// also use sum as the receiver where result of Add will be stored
// if the method Add has pointer receiver.
//
// To cover the case where method Add has instead value receiver,
// also assign the returned value to sum
sum = sum.Add(sum, elem)
}
return sum
}
Sum#[int] // returns func(...int) int
Sum#[int] (1,2,3) // returns int(6)
Sum#[complex64] // returns func(...complex64) complex64
Sum#[complex64] (1.1+2.2i, 3.3) // returns complex64(4.4+2.2i)
Sum#[string] // returns func(...string) string
Sum#[string]("abc.","def.","xy","z") // returns "abc.def.xyz"
Partial and full specialization of generics is not supported in CTI generics,
both for simplicity and to avoid accidentally providing Turing completeness at compile-time.
Instantiation of generic types and functions is on-demand.
Current limitations:
- type inference on generic arguments #[...] is not yet implemented,
thus generic arguments #[...] must be explicit.
- generic methods are not yet implemented.
- types are not checked to actually satisfy contracts.
Debugger
Since version 2.6, gomacro also has an integrated debugger.
There are three ways to enter it:
- hit CTRL+C while interpreted code is running.
- type
:debug STATEMENT-OR-FUNCTION-CALL
at the prompt.
- add a statement (an expression is not enough)
"break"
or _ = "break"
to your code, then execute it normally.
In all cases, execution will be suspended and you will get a debug>
prompt, which accepts the following commands:
step
, next
, finish
, continue
, env [NAME]
, inspect EXPR
, list
, print EXPR-OR-STATEMENT
Also,
- commands can be abbreviated.
print
fully supports expressions or statements with side effects, including function calls and modifying local variables.
env
without arguments prints all global and local variables.
- an empty command (i.e. just pressing enter) repeats the last command.
Only interpreted statements can be debugged: expressions and compiled code will be executed, but you cannot step into them.
The debugger is quite new, and may have some minor glitches.
Why it was created
First of all, to experiment with Go :)
Second, to simplify Go code generation tools (keep reading for the gory details)
Problem: "go generate" and many other Go tools automatically create
Go source code from some kind of description - usually an interface
specifications as WSDL, XSD, JSON...
Such specification may be written in Go, for example when creating JSON
marshallers/unmarshallers from Go structs, or in some other language,
for example when creating Go structs from JSON sample data.
In both cases, a variety of external programs are needed to
generate Go source code: such programs need to be installed
separately from the code being generated and compiled.
Also, Go is currently lacking generics (read: C++-like templates)
because of the rationale "we do not yet know how to do them right,
and once you do them wrong everybody is stuck with them"
The purpose of Lisp-like macros is to execute arbitrary code
while compiling, in particular to generate source code.
This makes them very well suited (although arguably a bit low level)
for both purposes: code generation and C++-like templates, which
are a special case of code generation - for a demonstration of how
to implement C++-like templates on top of Lisp-like macros,
see for example the project https://github.com/lifepod-solutions/cl-parametric-types
from the same author.
Building a Go interpreter that supports Lisp-like macros,
allows to embed all these code-generation activities
into regular Go source code, without the need for external programs
(except for the interpreter itself).
As a free bonus, we get support for Eval()
LEGAL
Gomacro is distributed under the terms of Mozilla Public License 2.0
or any later version.