Home

Awesome

Go Report Card license

Table of Contents

math-compiler

This project contains the simplest possible compiler, which converts mathematical operations into assembly language, allowing all the speed in your sums!

Because this is a simple project it provides only a small number of primitives:

Despite this toy-functionality there is a lot going on, and we support:

Some errors will be caught at run-time, as the generated code has support for:

Installation

If you just need a binary you can find them upon the project release page, however if you wish to build and install locally you can do that in either of the standard ways:

  1. Install from the latest revision:
$ go install github.com/skx/math-compiler@master
  1. Or you can clone the source, and build from it:
$ git clone https://github.com/skx/math-compiler
$ cd math-compiler
$ go install .

Quick Overview

The intention of this project is mostly to say "I wrote a compiler", because I've already experimented with a language, an embedded evaluation engine, and implemented a BASIC interpreter. The things learned from those projects were pretty useful, even if the actual results were not so obviously useful in themselves.

Because there are no shortages of toy-languages, and there is a lot of complexity in writing another for no real gain, I decided to just focus upon a simple core:

In theory this would allow me to compile things like this:

2 + ( 4 * 54 )

However I even simplified that, via the use of a "Reverse Polish" notation, so if you want to run that example you'd enter the expression as:

4 54 * 2 +

About Our Output

The output of math-compiler will be an assembly-language file, which then needs to be compiled before it may be executed.

Given our previous example of 2 + ( 4 * 54) we can compile & execute that program like so:

$ math-compiler '4 54 * 2+' > sample.s
$ gcc -static -o sample ./sample.s
$ ./sample
Result 218

There you see:

If you prefer you can also let the compiler do the heavy-lifting, and generate an executable for you directly. Simply add -compile, and execute the generated a.out binary:

$ math-compiler -compile=true '2 8 ^'
$ ./a.out
Result 256

Or to compile and execute directly:

$ math-compiler -run '3 45 * 9 + 12 /'
Result 12

Test Cases

The codebase itself contains some simple test-cases, however these are not comprehensive as a large part of our operation is merely to populate a simple template-file, and it is hard to test that.

To execute the integrated tests use the standard go approach:

$ go test [-race] ./...

In addition to the internal test cases there are also some functional tests contained in test.sh - these perform some calculations and verify they produce the correct result.

frodo ~/go/src/github.com/skx/math-compiler $ ./test.sh
...
Expected output found for '2 0 ^' [0]
Expected output found for '2 1 ^' [2]
Expected output found for '2 2 ^' [4]
Expected output found for '2 3 ^' [8]
Expected output found for '2 4 ^' [16]
Expected output found for '2 5 ^' [32]
...
Expected output found for '2 30 ^' [1073741824]
...

Debugging the generated programs

If you run the compiler with the -debug flag a breakpoint will be generated immediately at the start of the program. You can use that breakpoint to easily debug the generated binary via gdb.

For example you might generate a program "2 3 + 4 /" like so:

$ math-compiler -compile -debug '2 3 + 4 /'

Now you can launch that binary under gdb, and run it:

$ gdb ./a.out
(gdb) run
..
Program received signal SIGTRAP, Trace/breakpoint trap.
0x00000000006b20cd in main ()

Dissassemble the code via disassemble, and step over instructions one at a time via stepi. If your program is long you might see a lot of output from the disassemble step:

(gdb) disassemble
Dump of assembler code for function main:
   0x00000000006b20cb:	push   %rbp
   0x00000000006b20cc:	int3
=> 0x00000000006b20cd:	fldl   0x6b20b3
   0x00000000006b20d4:	fstpl  0x6b2090
   0x00000000006b20db:	mov    0x6b2090,%rax
   0x00000000006b20e3:	push   %rax
   0x00000000006b20e4:	fldl   0x6b20bb
   0x00000000006b20eb:	fstpl  0x6b2090
   0x00000000006b20f2:	mov    0x6b2090,%rax
   0x00000000006b20fa:	push   %rax
   ...
   ...

You can set a breakpoint at a line in the future, and continue running till you hit it, with something like this:

 (gdb) break *0x00000000006b20fa
 (gdb) cont

Once there inspect the registers with commands like these two:

 (gdb) print $rax
 (gdb) info registers

My favourite is info registers float, which shows you the floating-point values as well as the raw values:

 (gdb) info registers float
 st0            0.140652076786443369638	(raw 0x3ffc90071917a6263000)
 st1            0	(raw 0x00000000000000000000)
 st2            0	(raw 0x00000000000000000000)
 ...
 ...

Further documentation can be found in the gdb manual, which is worth reading if you've an interest in compilers, debuggers, and decompilers.

Possible Expansion?

The obvious thing to improve in this compiler is to add support for more operations. At the moment support for the most obvious/common operations is present, but perhaps more functions could be added.

See Also

If you enjoyed this repository, then you might also enjoy my compiler for the Brainfuck language. The compiler there compiles brainfuck programs to x86-64 assembly-language:

Github Setup

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.

Steve