HookRace Blog     Posts     Feed     DDNet     Me     Switch Theme

What makes Nim practical?

In my last post I showed what makes the Nim programming language special. Today, let’s consider Nim from another angle: What makes Nim a practical programming language?

Binary Distribution

Programs written in interpreted languages like Python are difficult to distribute. Either you require Python (in a specific version even) to be installed already, or you ship it with your program. This even causes some to reconsider Python as a teaching language.

How does Nim work around this problem? For starters your program gets statically linked against the Nim runtime. That means you end up with a single binary that depends solely on the standard C library, which we can take for granted on any operating system we’re interested in.

Let’s write a small program and give this a try:

echo "Hello World"

If you want to follow along, get the Nim compiler. Save this code as hello.nim. Let’s compile it now so that we can distribute the hello binary:

$ nim -d:release c hello
CC: hello
CC: stdlib_system
[Linking]
$ ./hello
Hello World
$ ls -lha hello
-rwxr-xr-x 1 def def 57K Jan 22 09:37 hello*
$ ldd hello
    linux-vdso.so.1 (0x00007fffd5973000)
    libdl.so.2 => /lib64/libdl.so.2 (0x00007f0f92c6b000)
    libc.so.6 => /lib64/libc.so.6 (0x00007f0f928c3000)
    /lib64/ld-linux-x86-64.so.2 (0x00007f0f92e6f000)

Now our little program starts growing and we get interested in using a few Nim libraries. Do we have to add compiled versions of these libraries to our distributions now? Do not fret! Nim libraries are statically compiled into our binary as well. Let’s get a library using Nim’s package manager, nimble:

$ nimble install strfmt
Installing strfmt-0.5.4
strfmt installed successfully.

And use the library in our, admittedly not very useful, program:

import strfmt

echo "Hello {} number {:04.1f}".fmt("World", 6.0)

If we want to compile with Clang instead of GCC as the backend, that’s easy as well and Clang is usually much faster to compile and yields a smaller binary:

$ nim --cc:clang -d:release c hello
$ ./hello
Hello World number 06.0
$ ls -lha hello
-rwxr-xr-x 1 def def 54K Jan 22 10:06 hello*

The only dynamic libraries we have to worry about are the C libraries we use, but that’s a story for another time.

Source Distribution

We can’t compile our program for all possible combinations of operating systems and CPU architectures of course. But we can be prepared for them and create a source distribution that can be compiled on the target systems with just a C compiler:

$ tree
.
├── hello.ini
├── hello.nim
└── lib
    └── nimbase.h
$ cat hello.ini
[Project]
Platforms: """
  windows: i386;amd64
  linux: i386;amd64;powerpc64;arm;sparc;mips;powerpc
  macosx: i386;amd64;powerpc64
  solaris: i386;amd64;sparc
  freebsd: i386;amd64
  netbsd: i386;amd64
  openbsd: i386;amd64
  haiku: i386;amd64
"""

[Lib]
Files: "lib/nimbase.h"

[Windows]
binPath: "bin"
$ niminst csource hello.ini -d:release

Here’s the resulting build script if you want to try it out on any of these platforms: hello-csources.zip

$ unzip hello-csources.zip
$ ./build.sh
SUCCESS
$ bin/hello
Hello World

niminst can also be used to build full installers for your program. It’s what the Nim compiler itself uses.

Debug and Release Builds

Debug and release builds behave quite differently in Nim by default. Here’s an overview:

Option Debug build Release build
objChecks On Off
fieldChecks On Off
rangeChecks On Off
boundChecks On Off
overflowChecks On Off
assertions On Off
stackTrace On Off
lineTrace On Off
lineDir On Off
deadCodeElim Off On

As you develop your program you will be thankful for the range, bound and overflow checks. As you release your program your users will be thankful for the additional speed of disabling those.

If this is not what you want, and you require runtime checks even in release builds, you can enable them project wide or for parts of your code.

For example here we have an expensive max function that takes most of the running time:

proc max[T](x: varargs[T]): T =
  if x.len == 0:
    return

  result = x[0]

  for i in 1 .. x.high:
    if result < x[i]:
      result = x[i]

var s = newSeq[int]()
for i in 1..1_000:
  s.add(i)

echo max(s)

We want our program to never access out of sequence bounds, so we want to compile with nim -d:release --checks:on max. As it’s unreasonable to write this every time, we can create a max.nim.cfg file in our directory instead and enable runtime checks for every compilation of our program:

checks: on

Now we run into the problem that the max function runs too slowly (it doesn’t, but let’s imagine). So we disable runtime checks just for this function:

{.push checks: off.}
proc max[T](x: varargs[T]): T =
  if x.len == 0:
    return

  result = x[0]

  for i in 1 .. x.high:
    if result < x[i]:
      result = x[i]
{.pop.}

Great, now we can control which part of our program has runtime checks and which doesn’t. checks enables and disables all runtime checks at once, but there are more fine-grained controls as well.

Nim instead of Python + C++

High performance languages like C++ may require some boilerplate. A higher level language can be used at compile time to automatically create the boilerplate. In DDNet (which inherited them from Teeworlds) there are many examples for this.

A really simple use case is to get the current git revision at compile time in Python and put it into the config.h with a #define so it can be referred to at runtime in the program. In Nim you need neither Python nor a C preprocessor and can instead do it all directly at compile time in Nim:

const gitHash = staticExec "git rev-parse HEAD"
echo "Revision: ", gitHash

Use as a scripting language

With TinyCC as the backend Nim makes for a nice scripting language. All we need is a nimscript file and the TinyCC compiler:

#!/bin/sh
nim --cc:tcc --verbosity:0 -d:release -r c $*

To automatically update DDNet’s map files on our upcoming HTTP map server I’m using this script (with this crc32 module):

#!/usr/bin/env nimscript

import os, crc32, strutils

const
  baseDir = "/home/teeworlds/servers"
  mapdlDir = "/var/www-maps"

for kind, path in walkDir baseDir/"maps":
  if kind != pcFile:
    continue

  let (dir, name, ext) = splitFile(path)

  if ext != ".map":
    continue

  let
    sum = crc32FromFile(path).int64.toHex(8).toLower
    newName = name & "_" & sum & ext
    newPath = mapdlDir / newName
    tmpPath = newPath & ".tmp"

  if existsFile newPath:
    continue

  copyFile path, tmpPath
  moveFile tmpPath, newPath

Of course you can always compile your program if you need the full speed of Clang or GCC. But for quick tests while developing, or scripts you just need occasionally, this is good enough.

There is a nicer version of this idea now, called nimrun

Debugging Nim

GDB works rather well with Nim. The only problem I’ve had so far is that the variables get unique identifiers in C, but using tab completion you can get along fine:

echo "Hello World"
var x = 10
echo "Value of x: ", x
$ nim --lineDir:on --debuginfo c hello
$ gdb ./hello
(gdb) break hello.nim:3
Breakpoint 1 at 0x41f886: file /home/def/hello.nim, line 3.
(gdb) run
Starting program: /home/def/hello
Hello World
Breakpoint 1, helloInit () at /home/def/hello.nim:3
3       echo "Value of x: ", x
(gdb) print x_89010
$1 = 10
(gdb) print x_89010 = 200
$2 = 200
(gdb) c
Continuing.
Value of x: 200

Wrapping libraries with c2nim

Nice C libraries can automatically be converted into a Nim wrapper with c2nim. Let’s do this for Bellard’s new BPG image format. Since the library is so new, there is no shared library compilation included yet, so I added that myself in my fork.

We need to fix up the C header for c2nim by adding this at the top of libbpg.h:

#ifdef C2NIM
#  dynlib bpglib
#  cdecl

/* Dynamically link to the correct library for our system: */
#  if defined(windows)
#    define bpglib "libbpg.dll"
#  elif defined(macosx)
#    define bpglib "libbpg.dylib"
#  else
#    define bpglib "libbpg.so"
#  endif

/* Remove prefixes in our wrapper, we have modules in Nim: */
#  prefix bpg_decoder_
#  prefix BPG_
#  prefix BPG

/* These are not recognized by c2nim, but that's easy to fix: */
#  mangle uint8_t uint8
#  mangle uint16_t uint16
#  mangle uint32_t uint32
#endif

I made a patch so I can apply the changes again when the libbpg interface changes. Creating the wrapper is now as simple as this:

$ patch -p1 < libbpg.h.patch
patching file libbpg.h
$ c2nim -o:bpg.nim libbpg.h
Hint: operation successful (155 lines compiled; 0 sec total; 516.528KB) [SuccessX]

Look, no hands! The resulting wrapper doesn’t even look bad. Now we can write a simple program to decode a BPG file and save it in the PPM format:

import bpg, os

proc writePPM(img, filename) =
  var imgInfo: ImageInfo
  discard img.getInfo(addr imgInfo)

  let (w,h) = (imgInfo.width.int, imgInfo.height.int)
  var rgbLine = newSeq[uint8](w * 3)

  var f = open(filename, fmWrite)
  f.writeln "P6\n", w, " ", h, "\n255"

  discard img.start(OUTPUT_FORMAT_RGB24)
  for y in 1..h:
    discard img.getLine(addr rgbLine[0])
    discard f.writeBuffer(addr rgbLine[0], w * 3)

  f.close()

if paramCount() != 1:
  stderr.writeln "Usage: decode img.bpg"
  quit 1

var
  buf = readFile paramStr(1)
  img = bpg.open()

if img.decode(cast[ptr uint8](addr buf[0]), buf.len.cint) < 0:
  stderr.writeln "Could not decode image"
  quit 2

img.writePPM("out.ppm")
img.close()

Note that you can tell from the discard statements where I chose to ignore errors. If you actually want to run this, check out the instructions in the repository.

Final words

That’s Nim from a more practical angle. Hopefully you’ll consider Nim for your next project, there are many libraries available already. Also, the community always needs more helping hands!

To do my part in making Nim more practical, I’m trying to implement a new, working REPL using TinyCC. We may also soon get a working compiler as a service for proper IDE integration, directly from Andreas Rumpf.

For comments use Reddit, Hacker News or get in touch with the Nim community directly. You can reach me personally at [email protected].

Thanks to Andreas Rumpf and Dominik Picheta again, for proof-reading this post as well.