How To Build MRI n Times Faster

Clang, a compiler front end which you can use instead of gcc, can build MRI: Matz Ruby Implementation, only if you succeeded in avoiding all pitfalls.

Build the latest LLVM and Clang

Mac OS X Snow Leopard has builtin clang command but it's too old to build ruby. First of all, get the latest clang and it's dependency, the latest LLVM.

$ svn checkout http://llvm.org/svn/llvm-project/llvm/trunk llvm && cd llvm/tools && svn checkout http://llvm.org/svn/llvm-project/cfe/trunk clang && cd ..
$ ./configure && time make

Note that it takes really long time. I recommend you not to build before you go away from keyboard for a while but to build before you go to bed.

A simple test

hello.c

#include<stdio.h>
int main() {
  puts("Hello, world!");
  return 0;
}

building on gcc and clang

$ time gcc hello.c -o hello-gcc
!!!        0.34 real         0.01 user         0.03 sys!!!

[%] /private/tmp
$ time ~/src/llvm/Debug+Asserts/bin/clang hello.c -o hello-clang
!!!        0.11 real         0.06 user         0.02 sys!!!

clang is 3 times faster than gcc to build the hello world code. The executable files work samel.

Ruby

On GCC

$ git clone https://github.com/ruby/ruby.git ruby193-gcc
$ cd ruby193-gcc
$ autoconf
$ ./configure --prefix=`pwd`/local --enable-pthread --disable-install-doc --with-out-ext=tk\* --program-suffix=193 --with-readline-dir=/usr/local --with-arch=x86_64
$ time make
...
!!!      214.93 real       191.58 user        23.57 sys!!!

On Clang

$ git clone https://github.com/ruby/ruby.git ruby193
$ cd ruby193
$ ./configure --prefix=`pwd`/local --enable-pthread --disable-install-doc --with-out-ext=tk\* --program-suffix=193 --with-readline-dir=/usr/local PATH=/Users/ujihisa/src/llvm/Debug+Asserts/bin/:$PATH CC=/Users/ujihisa/src/llvm/Debug+Asserts/bin/clang CFLAGS=-Qunused-arguments CPPFLAGS=-Qunused-arguments --with-arch=x86_64
$ time make
!!!      932.06 real       883.44 user        31.09 sys!!!

It took 3 min 34.93 sec by GCC while it took 15 min 32.06 sec by Clang. GCC is 4.34 times faster than Clang to build MRI.

Pitfalls

If you use clang command of Mac OS X Snow Leopard, even though you'll succeed in finishing make ruby 1.9.3, but you cannot do anything by the Ruby.

$ ./local/bin/ruby -ve 1
ruby 1.9.3dev (2011-01-25 trunk 30651) [x86_64-darwin10.6.0]
!!!dyld: lazy symbol binding failed: Symbol not found: _rb_encdb_declare!!!
!!!  Referenced from: /Users/ujihisa/git/ruby193/local/lib/ruby/1.9.1/x86_64-darwin10.6.0/enc/encdb.bundle!!!
!!!  Expected in: flat namespace!!!

!!!dyld: Symbol not found: _rb_encdb_declare!!!
!!!  Referenced from: /Users/ujihisa/git/ruby193/local/lib/ruby/1.9.1/x86_64-darwin10.6.0/enc/encdb.bundle!!!
!!!  Expected in: flat namespace!!!

!!!/Users/ujihisa/git/ruby193/local/lib/ruby/1.9.1/x86_64-darwin10.6.0/enc/encdb.bundle: [BUG] Segmentation fault!!!
!!!ruby 1.9.3dev (2011-01-25 trunk 30651) [x86_64-darwin10.6.0]!!!

!!!-- Control frame information -----------------------------------------------!!!
!!!c:0002 p:-537663266 s:0004 b:0004 l:000003 d:000003 TOP   !!!
!!!c:0001 p:0000 s:0002 b:0002 l:001dd8 d:001dd8 TOP   !!!


!!!-- See Crash Report log file under ~/Library/Logs/CrashReporter or ---------!!!
!!!-- /Library/Logs/CrashReporter, for the more detail of ---------------------!!!
!!!-- C level backtrace information -------------------------------------------!!!

!!!-- Other runtime information -----------------------------------------------!!!

!!!* Loaded script: ./local/bin/ruby!!!

!!!* Loaded features:!!!

!!!    0 enumerator.so!!!

!!![NOTE]!!!
!!!You may have encountered a bug in the Ruby interpreter or extension libraries.!!!
!!!Bug reports are welcome.!!!
!!!For details: http://www.ruby-lang.org/bugreport.html!!!


!!!vimshell: signal 6(SIGABRT) "./local/bin/ruby -ve 1"!!!

Conclusion

You can build Ruby 1.9.3 by Clang. It's 0.23 times faster than building by GCC.

A Contributed Article to Kernel/VM Advent Calendar: VIM=VM

There are two kinds of programmers; one uses Vim and the other uses Emacs. I don't think there are positive counter opinions about that Emacs is not an editor but an environment. On the other hand, there are some opinions that Vim is an editor or not, even though everyone agrees with the fact that vi is an editor. Some people, including the author of this article, claim that Vim is a virtual machine. Here I give you instructions of a Vim plugin shadow.vim instead of explaining why Vim is VM directly.

shadow.vim

The Vim plugin shadow.vim supports wrapping a virtual file with a program automatically with thin configuration.

Here it's a quote from a programmer:

"Nobody knows the Java code you committed is originally written in Scheme."

Usage

Assuming the product is a.pl, create a.pl.shd first.

a.pl (in Vim):

## ruby -e 'puts $<.read.gsub(/$/, ";")'
$a = 1
print($a)

Open a.pl in Vim. The Vim actually shows the contents of a.pl.shd. When you save the file, the command in the first line without ## runs, then the actual a.pl will be the result.

a.pl (actually):

$a = 1;
print($a);

Install

Unarchive the zip file into a directory that is under &rtp of your Vim, which stands for run time path, including ~/.vim dir.

Use Case

Here there are three examples, but you can use more general purposes.

• Commit JavaScript files which was written in CoffeeScript

  • before

        ## coffee -csb
        f = (x) -> x + 1
        print f 10
    
        # vim: set ft=coffee :
    

  • after

        var f;
        f = function(x) {
          return x + 1;
        };
        print(f(10));
    

• Use cpp command before committing Java files.

• Markdown, Haml or something else to HTML

More examples

You can write code in C and save the file as GNU Assembly Language on your Linux automatically, so you can avoid portability.

References

• VIM=VM http://www.slideshare.net/Shougo/vim-presentation-5330807 by Shougo
• http://atnd.org/events/10701

Búsqueda/Instalación Librerías RubyGems Fácilmente

¿Cómo se puede explorar y tratar de librerías RubyGems? Hay una gran cantidad de joyas como las estrellas en el cielo. Si lo desea, estar de pie sobre los hombros de gigantes.

Una escena típica de cuando usted está buscando una librería que cumple su trabajo sería como el siguiente, por ejemplo, que busca una librería que está relacionado con DataMapper.

$ gem search dm

No consigues nada porque se le olvidó especificar -r opción.

$ gem search dm -r

Usted obtiene demasiados resultados, porque hay demasiadas joyas que incluyen "dm" en el nombre, por ejemplo "admin" algo.

De todos modos, que finalmente encontró una buena librería cuyo nombre es "dm-is-persistent_state_machine".

 $ gem install dm-is-persistent_states_machine

Este failes porque el nombre de la librería correcta es "dm-is-persistent_state_machine", mientras que ha escrito "dm-is-persistent_states_machine". Sí, eso es muy difícil de escribir correctamente cuando el nombre de la librería es muy larga. Puede ser difícil si el nombre no está en su lengua materna. (*1)

unite-gem

Una solución es usar un plugin Unite unite-gem. Después de instalar este plugin, basta con ejecutar

 :Unite gem

y el tipo nada. Verá los siguientes resultados.

y se puede reducir el número de candidatos.

a continuación, simplemente pulse <Cr>... el proceso de instalación se produjo. Usted no tiene que introducir el nombre completo de la librería.

Nota

• (*1): por ejemplo, muchas personas trataron de tipo "nokogiri", pero mal escrita como "nokogirl". Por último los autores Nokogiri realizó una joya alias "nokogirl" y todo el mundo se puso feliz. Este blog es sólo givin una versión más general de la solución.

How To Write And Show Multibyte Characters With Snap Framework

I'm playing with Snap these days. Snap is a web framework for Haskell.

The default template of application code will be like below

{-# LANGUAGE OverloadedStrings #-}
module Main where

import           Control.Applicative
import           Snap.Types
import           Snap.Util.FileServe
import           Text.Templating.Heist
import           Text.Templating.Heist.TemplateDirectory

import           Glue
import           Server


main :: IO ()
main = do
    td <- newTemplateDirectory' "templates" emptyTemplateState
    quickServer $ templateHandler td defaultReloadHandler $ \ts ->
        ifTop (writeBS "hello") <|>
        route [ ("foo", writeBS "bar")
              , ("echo/:echoparam", echoHandler)
              ] <|>
        templateServe ts <|>
        dir "static" (fileServe ".")


echoHandler :: Snap ()
echoHandler = do
    param <- getParam "echoparam"
    maybe (writeBS "must specify echo/param in URL")
          writeBS param

See the following line in main function:

        ifTop (writeBS "hello") <|>

This just shows a text written in ByteString "hello" to browser.

$ cabal install
$ {your application name} 3000
$ curl http://localhost:3000
hello

You can change the message "hello" to something and show the message on browser, but if the text is non-ascii like utf-8, you have to change other places as well. See the following patch:

diff --git snapfib.cabal snapfib.cabal
index d7a98fe..35b5c92 100644
--- snapfib.cabal
+++ snapfib.cabal
@@ -28,7 +28,8 @@ Executable snapfib
     text,
     containers,
     MonadCatchIO-transformers,
-    filepath >= 1.1 && <1.2
+    filepath >= 1.1 && <1.2,
+    utf8-string

   if impl(ghc >= 6.12.0)
     ghc-options: -threaded -Wall -fwarn-tabs -funbox-strict-fields -O2
diff --git src/Main.hs src/Main.hs
index d5b24c4..036b2db 100644
--- src/Main.hs
+++ src/Main.hs
@@ -9,13 +9,14 @@ import           Text.Templating.Heist.TemplateDirectory

 import           Glue
 import           Server
+import           Data.Text


 main :: IO ()
 main = do
     td <- newTemplateDirectory' "templates" emptyTemplateState
     quickServer $ templateHandler td defaultReloadHandler $ \ts ->
-        ifTop (writeBS "hello") <|>
+        ifTop (writeText $ pack "こんにちはこんにちは!") <|>
         route [ ("foo", writeBS "bar")
               , ("echo/:echoparam", echoHandler)
               ] <|>

The function writeBS in Snap receives ByteString and shows the message to browser as a response. This cannot handle UTF-8 strings for some reason, so instead of using writeBS with ByteString you have to use writeText with Text represented message. You can convert into Text with Data.Text.pack function.

See also:

• http://ujihisa.blogspot.com/2010/12/writing-fibonacci-web-server-in-haskell.html

Ruby-like Object-oriented Notation in Haskell

This article doesn't discuss about something practical but about something experimental.

import Prelude hiding ((.))

send :: Object -> Method -> Object
send (Fixnum value) (Method "to_s") = Str $ show value
send _ _ = undefined

(.) = send

data Object = Fixnum Int | Str String deriving Show
data Method = Method String deriving Show

main = print $ x.to_s
  where
    x = Fixnum 10
    to_s = Method "to_s"

See the line in main function definition: print $ x.to_s. This looks like Ruby.

• You can hide the definition of Haskell Prelude "." and can define your own function which name is "."; here I defined send first then made an alias of it.
• Object and sending method into an object are like just a data and a function which receives an argument of a specific type unless you think about class or prototype; or just, say, "performance."

Writing Fibonacci Web Server in Haskell

A practice: implementing a web service which receives a number by URL parameter and returns the corresponding Fibonacci sequence number.

GET
http://localhost:8000/fib/10

returns

fib(10) = 55

I used Snap Framework.

$ mkdir snapfib
$ cd snapfib
$ snap init

This generates an empty snap project files. then

$ cabal install
$ snapfib

builds the codes and installs a command snapfib in your ~/.cabal/bin/snapfib. You may run the app locally with the default port 8000; you may open the default page, hello world, on your browser.

Then implement fib sequence, a utility function, and the controller/view. This allows you to get the Fibonacci sequence number just by URL parameter.

Below is the main routine code from the repository.

{-# LANGUAGE OverloadedStrings #-}
module Main where

import           Control.Applicative
import           Snap.Types
import           Snap.Util.FileServe
import           Text.Templating.Heist
import           Text.Templating.Heist.TemplateDirectory

import           Glue
import           Server
import           Data.ByteString.Char8 (pack, unpack)

fibs :: [Integer]
fibs = 0 : 1 : zipWith (+) fibs (tail fibs)

fibInString :: Int -> String
fibInString n = "fib(" ++ show n ++ ") = " ++ show (fibs !! n)

main :: IO ()
main = do
    td <- newTemplateDirectory' "templates" emptyTemplateState
    quickServer $ templateHandler td defaultReloadHandler $ \ts ->
        ifTop (writeBS "hello world") <|>
        route [ ("foo", writeBS "bar")
              , ("fib/:n", fibHandler)
              ] <|>
        templateServe ts <|>
        dir "static" (fileServe ".")


fibHandler :: Snap ()
fibHandler = do
    param <- getParam "n"
    maybe (writeBS "must specify fib/n in URL")
          (writeBS . pack . fibInString . read . unpack) param

(writeBS . pack . fibInString . read . unpack) param is converting param :: Maybe ByteString into Int, getting a Fibonacci sequence number in String, converting into ByteString again, and passing it to writeBS function which is defined in Snap Framework.

It was not difficult or complicated to implement such a simple web service in Haskell as long as you have basic Haskell knowledges like Maybe Monad or String manipulations. The problem was, in my opinion, it took long time to build the web app. Every time you fix your code, you have to wait for the compilation before you access your web service.

Investing The Methods Of An Object On Unite

How often do you use Object#methods? That's convenient particularly on IRB.

methods on irb

This is also available on Unite.vim. Write the following code on your ~/.vimrc

let s:unite_source = {
      \ 'name': 'evalruby',
      \ 'is_volatile': 1,
      \ 'required_pattern_length': 1,
      \ 'max_candidates': 30,
      \ }

function! s:unite_source.gather_candidates(args, context)
  if a:context.input[-1:] == '.'
    let methods = split(
          \ unite#util#system(printf('ruby -e "puts %s.methods"', a:context.input[:-2])),
          \ "\n")
    call map(methods, printf("'%s' . v:val", a:context.input))
  else
    let methods = [a:context.input]
  endif
  return map(methods, '{
        \ "word": v:val,
        \ "source": "evalruby",
        \ "kind": "command",
        \ "action__command": printf("!ruby -e \"p %s\"", v:val),
        \ }')
endfunction

call unite#define_source(s:unite_source)

Execute :Unite evalruby on your Vim, and write an expression.

unite

Ruby Advent Calendar jp-en: 2010

This entry is for the event Ruby Advent Calendar jp-en: 2010. The previous post was from yhara and the next post will be from authorNari. I'm sure he will write an entry about GC.

I also had written an entry of the event last year http://ujihisa.blogspot.com/2009/11/write-implementation-and-spec-on-same.html. Time flees away without delay.