Forem: Junichi Kajiwara

Multi dimensional array in rust

Junichi Kajiwara — Thu, 19 Jul 2018 09:39:59 +0000

How to build plv8 2.3.0 for Raspberry Pi 3

Junichi Kajiwara — Thu, 22 Mar 2018 11:04:16 +0000

What is plv8

see:

https://pgxn.org/dist/plv8/doc/plv8.html

Building V8

I have done this by cross compiling.

You have to do following steps on Linux box(x86_64).

Installing depot_tools

cd ~
mkdir -p local/src
cd local
mkdir build
cd build
git clone https://chromium.googlesource.com/chromium/tools/depot_tools.git
export PATH=$PATH:~/local/depot_tools

fetching v8 source

cd src
fetch v8
cd v8
git checkout 6.4.388.40

6.4.388.40 : plv8-2.3.0 uses this version.

You may add follwing line to ../.gclient .

target_os = ['unix']

then, you call gclient sync.

gclient sync

Cross compileing for Raspberry Pi

tools/dev/v8gen.py arm.release -- \
is_component_build=false v8_static_library=true v8_use_snapshot=true \ 
v8_use_external_startup_data=false v8_enable_i18n_support=false

gn args out.gn/arm.release

then,edit parameters following:

is_debug = false
target_cpu = "arm"
v8_target_cpu = "arm"

# Additional command-line args:
is_component_build=false
v8_static_library=true
v8_use_snapshot=true
v8_use_external_startup_data=false
v8_enable_i18n_support=false

building d8 and libv8*.a

ninja -C out.gn/arm.release d8

When your build success,you'll have these files.

ls out.gn/arm.release/libv8*.a
libv8_base.a
libv8_libbase.a
libv8_libplatform.a
libv8_libsampler.a
libv8_snapshot.a

Copying files to raspberry pi

ssh pi@raspberrry.local mkdir $HOME/local/v8
scp out.gn/arm.release/libv8*.a pi@raspberrypi.local:/home/pi/local/v8
scp include pi@raspberrypi.local:/home/pi/local/v8

after this step,you may work on your raspberry pi.

Installing libc++

before build plv8,you'll need libc++.

sudo apt install libc++-dev libunwind-dev

Installing PostgreSQL

Of course you'll need PostgreSQL.Because plv8 is an extension of the PostgreSQL.

sudo apt install postgresql-server-dev-9.6 postgresql-9.6

Building plv8 itself on Raspberry Pi

mkdir -p ~/local/src
cd ~/local/src
curl https://github.com/plv8/plv8/archive/v2.3.0.zip
unzip v2.3.0.zip
cd plv8-2.3.0

copy v8 headers and libraries to plv8's source tree

mkdir -p build/v8/out.gn/x64.release/obj
cp -r ~/local/v8/include build/v8
cp ~/local/v8/lib*a build/v8/out.gn/x64.release/obj

Edit Makefile

Now, you have built v8 libraries.you don't need build v8 again.
So, you can edit Makefile following:

AUTOV8_DIR = build/v8
AUTOV8_OUT = build/v8/out.gn/x64.release/obj
- AUTOV8_DEPOT_TOOLS = build/depot_tools
AUTOV8_LIB = $(AUTOV8_OUT)/libv8_snapshot.a
AUTOV8_STATIC_LIBS = -lv8_base -lv8_snapshot -lv8_libplatform -lv8_libbase -lv8_libsampler
- export PATH := $(abspath $(AUTOV8_DEPOT_TOOLS)):$(PATH)

SHLIB_LINK += -L$(AUTOV8_OUT) -L$(AUTOV8_OUT)/third_party/icu $(AUTOV8_STATIC_LIBS)
-- V8_OPTIONS = is_component_build=false v8_static_library=true v8_use_snapshot=true v8_use_external_startup_data=false

- ifndef USE_ICU
-   V8_OPTIONS += v8_enable_i18n_support=false
- endif

- all: v8

# For some reason, this solves parallel make dependency.
- plv8_config.h plv8.so: v8

- $(AUTOV8_DEPOT_TOOLS):
-   mkdir -p build
-   cd build; git clone https://chromium.googlesource.com/chromium/tools/depot_tools.git

- $(AUTOV8_DIR): $(AUTOV8_DEPOT_TOOLS)
-   cd build; fetch v8; cd v8; git checkout $(AUTOV8_VERSION); gclient sync ; tools/dev/v8gen.py x64.release -- $(V8_OPTIONS)

- $(AUTOV8_OUT)/third_party/icu/common/icudtb.dat:

- $(AUTOV8_OUT)/third_party/icu/common/icudtl.dat:

- v8: $(AUTOV8_DIR)
-   cd $(AUTOV8_DIR) ; env CXXFLAGS=-fPIC CFLAGS=-fPIC ninja -C out.gn/x64.release d8

include Makefile.shared

CCFLAGS += -I$(AUTOV8_DIR)/include -I$(AUTOV8_DIR)
# We're gonna build static link.  Rip it out after include Makefile
SHLIB_LINK := $(filter-out -lv8, $(SHLIB_LINK))

ifeq ($(OS),Windows_NT)
    # noop for now
else
    SHLIB_LINK += -L$(AUTOV8_OUT)
    UNAME_S := $(shell uname -s)
    ifeq ($(UNAME_S),Darwin)
        CCFLAGS += -stdlib=libc++ -std=c++11
        SHLIB_LINK += -stdlib=libc++
    endif
    ifeq ($(UNAME_S),Linux)
        CCFLAGS += -std=c++11
-       SHLIB_LINK += -lrt -std=c++11 -lc++
+       SHLIB_LINK += -lrt -std=c++11 -lc++ -lunwind
    endif
endif

Building plv8

make

Installing postgres extension

sudo make install

run

sudo su - postgres
psql

CREATE EXTENSION plv8;
SELECT plv8_version();

Reading the message written in binary for Ruby 25th Anniversary using by mruby.

Junichi Kajiwara — Thu, 22 Feb 2018 05:20:07 +0000

Hello,I found this tweet.

// Detect dark theme var iframe = document.getElementById('tweet-966421801705267200-549'); if (document.body.className.includes('dark-theme')) { iframe.src = "https://platform.twitter.com/embed/Tweet.html?id=966421801705267200&theme=dark" }

I wonder what is written and I tried to read using by mruby.

Code

It seemed that the binary code is ascii code.

I wrote the decode binary code in mruby.

rawData = "01001001 00100000 01001100 01101111 01110110 01100101 00100000 01010010 01110101 01100010 01111001 00101110"
puts rawData.split(" ").map{|c| c.to_i(2)}.map{|c| c.chr("UTF-8")}.join

Result

bin/mirb 
mirb - Embeddable Interactive Ruby Shell

> rawData = "01001001 00100000 01001100 01101111 01110110 01100101 00100000 01010010 01110101 01100010 01111001 00101110"
 => "01001001 00100000 01001100 01101111 01110110 01100101 00100000 01010010 01110101 01100010 01111001 00101110"
> puts rawData.split(" ").map{|c| c.to_i(2)}.map{|c| c.chr("UTF-8")}.join
I Love Ruby.
 => nil
>

I noticed that Ruby has putc, but mruby didn't have.

epilogue

to tell the truth,I first time I wrote decode code in JavaScript.
but I thought this is a chanse to learn mruby,so I wrote this in mruby.

Link

https://github.com/mruby/mruby

require('./opencv.js') is not enough for using OpenCV.js

Junichi Kajiwara — Tue, 28 Nov 2017 03:27:08 +0000

Begin

I noticed that I can use official OpenCV with JavaScript(WASM).

https://docs.opencv.org/3.3.1/d4/da1/tutorial_js_setup.html

The problem

When I wrote code following:

const cv = require('./opencv.js');

let mat = new cv.Mat();
console.log(mat.size());
mat.delete();

Unfortunately I've got only error messages like this:

TypeError: cv.Mat is not a constructor
    at Object.<anonymous> (/Volumes/EXTHDD3/local/opencv/build_wasm/bin/dame.js:3:11)
    at Module._compile (module.js:641:30)
    at Object.Module._extensions..js (module.js:652:10)
    at Module.load (module.js:560:32)
    at tryModuleLoad (module.js:503:12)
    at Function.Module._load (module.js:495:3)
    at Function.Module.runMain (module.js:682:10)
    at startup (bootstrap_node.js:191:16)
    at bootstrap_node.js:613:3

Solution 1

I think I need some wait, and I rewrite the code following:

const cv = require('./opencv.js');

setTimeout(()=>{
        (()=>{
                let mat = new cv.Mat();
                console.log(mat.size());
                mat.delete();
        })();
},10000);

This time,there is no error message.

{ width: 0, height: 0 }

Solution 2

Solution 1 is not goot at looking, and depends on CPU's speed.

I found that Emscipten will call onRuntimeInitialized when it's ready to call main function.

const cv = require("./opencv.js");
cv['onRuntimeInitialized']=()=>{
          let mat = new cv.Mat();
          console.log(mat.size());
          mat.delete();
};

{ width: 0, height: 0 }

Called the library created by Go from mruby on Windows and displayed the image of the web camera at the command prompt

Junichi Kajiwara — Thu, 16 Nov 2017 01:32:41 +0000

background

I created a library in Go language specifying files and URLs at the Windows command prompt and displaying images at the command prompt

Preparation

The Go side inputs jpeg and png as the input source, as it is, it will call from mruby via the file as it is, worrying about the burden on the SSD w. Of course speed too.

To treat the start address passed from C as a slice of Go

I found this link:

https://github.com/golang/go/issues/13656

Apparently,

https://github.com/golang/go/wiki/cgo#Turning_C_arrays_into_Go_slices

It seems to be nice to write (there seems to be a limitation on the capacity up to 2 GB ..)

func termPutImage(imageData *C.uchar, width, height C.int) {
    len := 3 * width * height
    slice := (*[1 << 30]C.uchar)(unsafe.Pointer(imageData))[:len:len]

To convert to the Image type of Go language

As shown below, we were able to convert the array (P6 format of PPM) received from C side to Image type.

img := image.NewRGBA(image.Rect(0, 0, int(width), int(height)))
for h := 0; h < int(height); h++ {
    for w := 0; w < int(width); w++ {
        r := uint8(slice[3*(h*int(width)+w)])
        g := uint8(slice[3*(h*int(width)+w)+1])
        b := uint8(slice[3*(h*int(width)+w)+2])
        c := color.RGBA{r, g, b, 0}
        img.SetRGBA(w, h, c)
    }
}

To make libhogehoge.a format as Go

On Windows, shared libraries can not be created, but output of archive format libraries is supported.

go build -buildmode = c-archive -o libimgtype.a main.go

This creates a library in .a format. For MinGW's gcc environment, you can use it immediately.

Make MSVC mruby available create a lib file and link

The mrbgem named mruby-webcam we are about to use this time uses the MSVC version of OpenCV, and according to it, we will use MSVC.

Therefore, prepare a def file and generate a. Dll format from this and the .a(archive) format library of the previous. Also, generate the necessary .lib file from .def when linking in MSVC processing system. It will be necessary to do some troublesome work.

How to make a dll

The DLL went with the MinGW version of gcc (gcc specified with cgo).

gcc -m64 -shared -o imgtype.dll imgtype.def libimgtype.a -Wl,--allow-multiple-definition -static -lwinmm -lntdll -lWs2_32

How to create a lib file

The. lib file is done with the lib command attached to MSVC.

lib /machine:x64 /def:imgtype.def

Now I am ready to integrate into MSVC mruby.

https://github.com/kjunichi/libimgtype

Calling Go from mruby

If you come here, you only write mrbgem. Write the created. Lib file in mrbgem.rake as follows:

spec.linker.flags_before_libraries << "imgtype.lib"

You can pass the path to the imgtype.lib file to the environment variable LIB, and mruby.exe and mirb.exe can be created by incorporating Go library with rake.

To run

This mrb gem is required below:

conf.gembox 'default'
conf.gem '../mruby-webcam'
conf.gem '../mruby-imgtype'
conf.gem :github => 'matsumoto-r/mruby-sleep'

Below is a mruby script that displays the webcam image at the command prompt or Vim: terminal.

Imgtype.init

cam = Webcam.new
cam.setFmt "ppm"
cam.capture {|img|
  Imgtype.imgtype img
}
while true
    cam.snap
    Sleep::usleep(1000)
    if Imgtype.get_key == "ESC" then
        break
    end
end
Imgtype.close

Result

Imgtype.get_key reads Windows API directly and responds. I was surprised to get key input without polling.

https://t.co/gtYXg0tOA3

How to build plv8 on Windows

Junichi Kajiwara — Wed, 15 Nov 2017 08:17:36 +0000

This article is obsolete.Now officially plv8 supports for Windows!

TL;DR;

https://github.com/plv8/plv8/pull/239

Let's try

You need following:

vs2015/vs2017
Postgres9.6
git(and patch.exe)
cmake
nuget.exe

$ git clone https://github.com/kjunichi/plv8.git
$ cd plv8
$ git checkout support-msvc

you have to patch to PostgreSQL's header file.

copy "c:\Program Files\PostgreSQL\9.6\include\server\port\atomics\generic-msvc.h" .
set path=%path%;c:\Program Files\Git\usr\bin
patch --binary -p1 generic-msvc.h < generic-msvc.h

then,you copy patched generic-msvc.h to PostgreSQL's header file(c:\Program Files\PostgreSQL\9.6\include\server\port\atomics\generic-msvc.h).

$ bootstrap.bat
$ cmake . -G "Visual Studio 14 2015 Win64" -DCMAKE_INSTALL_PREFIX="C:\Program Files\PostgreSQL\9.6" -DPOSTGRESQL_VERSION=9.6
$ cmake --build . --config Release --target Package

when building is success,you'll get this.

plv8-2.1.0-postgresql-9.6-x64.zip

Unzip it, and copy to PostgreSQL directories.