This project is a golang based compiler and interpreter for a simple virtual machine. It is a port of the existing project:
(The original project has a perl based compiler/decompiler and an interpreter written in C.)
You can get a feel for what it looks like by referring to either the parent project, or the examples contained in this repository.
This particular virtual machine is intentionally simple, but despite that it is hopefully implemented in a readable fashion. ("Simplicity" here means that we support only a small number of instructions, and the 16-registers the virtual CPU possesses can store strings and integers, but not floating-point values.)
There are two ways to install this project from source, which depend on the version of the go version you're using.
If you prefer you can fetch a binary from our release page. Currently there is only a binary for Linux (amd64) due to the use of cgo
in our dependencies.
go get -u github.com/skx/go.vm
git clone https://github.com/skx/go.vm ;# make sure to clone outside of GOPATH
cd go.vm
go install
Once installed there are three sub-commands of interest:
go.vm compile $file.in
- Compiles the given program into bytecode.
go.vm execute $file.raw
- Given the path to a file of bytecode, then interpret it.
go.vm run $file.in
- Compiles the specified program, then directly executes it.
So to compile the input-file examples/hello.in
into bytecode:
$ go.vm compile examples/hello.in
Then to execute the resulting bytecode:
$ go.vm execute examples/hello.raw
Or you can handle both steps at once:
$ go.vm run examples/hello.in
The virtual machine has 16 registers, each of which can store an integer or a string. For example to set the first two registers you might write:
store #0, "This is a string"
store #1, 0xFFFF
In addition to this there are several mathematical operations which have the general form:
$operation $result, $src1, $src2
For example to add the contents of register #1 and register #2, storing the result in register #0 you would write:
add #0, #1, #2
Strings and integers may be displayed to STDOUT via:
print_str #1
print_int #3
Control-flow is supported via call
, ret
(for subroutines) and jmp
for absolute jumps. You can also use the Z
-flag which is set by
comparisons and the inc
/dec
instructions and make conditional jumps:
store #1, 0x42
cmp #1, 0x42
jmpz ok
store #1, "Something weird happened!\n"
print_str #1
exit
:ok
store #1, "Comparing register #01 to 0x42 succeeded!\n"
print_str #1
exit
Further instructions are available and can be viewed beneath examples/. The instruction-set is pretty limited, for example there is no notion of reading from STDIN - however this is supported via the use of traps, as documented below.
Some brief notes on parts of the code / operation:
The compiler is built in a traditional fashion:
- Input is split into tokens via lexer.go
- This uses the token.go for the definition of constants.
- The stream of tokens is iterated over by compiler.go
- This uses the constants in opcode.go for the bytecode generation.
The approach to labels is the same as in the inspiring-project: Every time we come across a label we output a pair of temporary bytes in our bytecode. Later, once we've read the whole program and assume we've found all existing labels, we go back up and fix the generated addresses.
You can use the dump
command to see the structure the lexer generates:
$ go.vm dump ./examples/hello.in
{STORE store}
{IDENT #1}
{, ,}
{STRING Hello, World!
}
{PRINT_STR print_str}
{IDENT #1}
{EXIT exit}
The core of the interpreter is located in the file cpu.go and is as simple and naive as you would expect. There are some supporting files in the same directory:
- register.go
- The implementation of the register-related functions.
- stack.go
- The implementation of the stack.
- traps.go
- The implementation of the traps, to be described below.
Compared to the original project there are two main changes:
- The
DB
/DATA
operation allows storing string data directly in the generated bytecode. - There is a notion of
traps
.- Rather than defining opcodes for complex tasks it is now possible to callback into the CPU-emulator to do work.
For example in simple.vm project this is possible:
DB 0x01, 0x02,
But this is not:
DB "This is a string, with terminator to follow"
DB 0x00
go.vm
supports this, and it is demonstrated in examples/peek-strlen.in.
The instruction int
can be used to call back to the emulator to do some work
on behalf of a program. The following traps are currently defined & available:
int 0x00
- Set the contents of the register
#0
with the length of the string in register#0
.
- Set the contents of the register
int 0x01
- Set the contents of the register
#0
with a string entered by the user. - See examples/trap.stdin.in.
- Set the contents of the register
int 0x02
- Update the (string) contents of register
#0
to remove any trailing newline. - See examples/trap.box.in.
- Update the (string) contents of register
Adding your own trap-functions should be as simple as editing cpu/traps.go.
Fuzz-testing is a powerful technique to discover bugs, in brief it consists of running a program with numerous random inputs and waiting for it to die.
I've fuzzed this repository repeatedly via go-fuzz and fixed a couple of minor issues.
Note however that fuzzing will trigger some expected failures. Our virtual CPU has only 16 registers, so for example a program that tries to set register #30 to a particular value is invalid, and will terminate the virtual machine.
Because fuzzing involves using "random" input it is possible there are bugs lurking in the virtual-machine which I've not been lucky enough to catch, so if you wish to fuzz this is how you do it. First of all install the tool:
$ go get github.com/dvyukov/go-fuzz/go-fuzz
$ go get github.com/dvyukov/go-fuzz/go-fuzz-build
Now you can build the interpreter using it:
$ go-fuzz-build github.com/skx/go.vm/fuzz
Finally you can launch the fuzzer:
$ go-fuzz -nprocs=1 -bin=fuzz-fuzz.zip -workdir=workdir
Interesting results will appear in workdir/crashers/
some crashes will be invalid and you can see that via the *.output
files which contain STDOUT from the run (more or less). For example this is an "expected" failure:
$ cat workdir/crashers/92108737efbd0ac6b42ae4473db5a257314b36cf.output
Register 99 out of range
exit status 1
This repository is configured to run tests upon every commit, and when pull-requests are created/updated. The testing is carried out via .github/run-tests.sh which is used by the github-action-tester action.
Releases are automated in a similar fashion via .github/build, and the github-action-publish-binaries action.