Giuseppe "Cowo" Corbelli weblog

DISCLAIMER:The following is tested ONLY on win32 systems (Windows 7/Windows XP).
Suppose you are developing two C/Python modules using boost::python library. Let's say you have two modules A and B with interfaces like

//Module A
typedef struct opaque_ *opaque;
BOOST_PYTHON_OPAQUE_SPECIALIZED_TYPE_ID(opaque_)
opaque get();

//Module B
typedef struct opaque_ *opaque;
BOOST_PYTHON_OPAQUE_SPECIALIZED_TYPE_ID(opaque_)
void use(opaque);

You'd expect to interchange opaque objects between the modules, i.e. develop in python somehow like this

ptr = A.get()
B.use(ptr)

The typeid specialization instantiated by BOOST_PYTHON_OPAQUE_SPECIALIZED_TYPE_ID looks like to be stored in a static map allocated by boost::python. Thus, if you link boost::python statically you'll end up having two separate maps and you won't be able to use cross-module opaque pointers because the type opaque will look different in the two modules.
If you link boost::python dinamically the map will be shared (only one copy of the DLL mapped in the calling process address space, of course) and everything will work as expected.
In short: IF YOU WANT TO USE CROSS-MODULE OPAQUE POINTERS ALWAYS LINK BOOST::PYTHON DINAMICALLY

For some reason some people don't like having consoles on the screen. Someone complained of a program not working after it had been shut down by closing its console. Curious behaviour, indeed...
Rather than explaining to the dumbasses what they are doing or creating a GUI app to be minimized in tray I searched for a different solution.
For sure you can use a launcher with a crafted STARTUPINFO but what if you cannot/don't want to? The key is the FreeConsole function (http://msdn.microsoft.com/en-us/library/ms683150(v=vs.85).aspx). Once you call this function the console will be destroyed (i.e. the black screen will go away) and you won't be able to access it anymore, so make sure you're not using the stdio system afterwards.
An example in python would be:

try:
    #Only works on win32 with win32 modules installed
    import win32console
except ImportError:
    pass
else:
    win32console.FreeConsole()

as simple as that!
In a GUI subsystem program it SHOULD have no effect besides returning FALSE but I never tried that.

Recently I had to decode a barcode from a color image got from an industrial linear camera. The image was very high resolution (4000 x 4000) but of course the barcode part was just a small subset, so first I started by cropping the original image to obtain just the barcode strip.

Tools

• Python as the programming language, however C/C++ is also an option
• ZBar as the barcode decoding library
• OpenCV as the image manipulation library

The base image

As you can see the base image is an 8bit depth BMP image. Barcode decoding however works on black&white images only. So the first step is image thresholding, converting the color image to a BW one. The goal can be achieved with the following code:

#!/usr/bin/python
import cv

#Input image
im = cv.LoadImageM("barcodecrop.bmp")

#Work buffer
tmp = cv.CreateImage(cv.GetSize(im), cv.IPL_DEPTH_8U, 1)

#Output image
oim = cv.CreateImage(cv.GetSize(im), cv.IPL_DEPTH_8U, 1)

#R,G,B channels
r = cv.CreateImage( cv.GetSize(im), cv.IPL_DEPTH_8U, 1 )
g = cv.CreateImage( cv.GetSize(im), cv.IPL_DEPTH_8U, 1 )
b = cv.CreateImage( cv.GetSize(im), cv.IPL_DEPTH_8U, 1 )

cv.Split( im, r, g, b, None )

#Create the work buffer
cv.AddWeighted( r, 1./3., g, 1./3., 0.0, tmp )
cv.AddWeighted( tmp, 2./3., b, 1./3., 0.0, tmp )

#Threshold and save into the output buffer
cv.Threshold( tmp, oim, 43, 255, cv.CV_THRESH_BINARY )
#Alternate thresholding options
#cv.AdaptiveThreshold( tmp, oim, 255, cv.CV_ADAPTIVE_THRESH_MEAN_C, cv.CV_THRESH_BINARY, 71, 12)
#CV_ADAPTIVE_THRESH_GAUSSIAN_C
#CV_ADAPTIVE_THRESH_MEAN_C

#Save the output image
cv.SaveImage("fixed.bmp", oim)

Now you have a thresholded BW image saved in barcodeth.bmp.

The decoding phase

Barcode decoding is handled by ZBar. I just used one of the provided examples as a reference. Far from being optimal (everything is file based, there's an unneeded conversion...) but it gets to the point. That's it.

I have a system with root over iSCSI on a Debian system. To maximize network transfers I have increased the length of Tx and Rx buffers to 4096 using ethtool:

ethtool -G eth0 rx 4096 tx 4096

However this means the adapter has to be resetted after boot. To avoid this on an Intel 82578DM (e1000e driver) you can recompile the driver increasing the default ring buffer size.

1. Open /usr/src/linux/drivers/net/e1000e/e1000.h
2. Search for

   #define E1000_DEFAULT_TXD

3. Modify the line to

   #define E1000_DEFAULT_TXD 4096

4. Search for

   #define E1000_DEFAULT_RXD

5. Modify the line to

   #define E1000_DEFAULT_RXD 4096

6. Save & recompile

Jumbo frames can be modified with the DEFAULT_JUMBO but the switch does not support them so I have no way to check.

I'm by no means an Apache expert but I had to setup an authentication system for Apache 2.2 capable of querying a Microsoft Active Domain server using LDAP and falling back to plain old htpasswd files.
LDAP is quite complex stuff and, again, I'm by no means an expert but I was able to gather up some infos here and there.
The key concept is using Multiple authentication providers within the AuthBasicProvider Apache directive.
To issue queries to the LDAP database you may be able to use anonymous access. Mine was not the case: since I don't know how to enable anonymous queries (and neither does the administrator) the quick'n'dirty solution was to use my own account.
I am also using Trac on this machine so some references you'll find are related to the Trac installation.

Loading the right modules

Module name	Why?	Directive
authn_file	To enable the htpasswd file based auth provider	LoadModule authn_file_module /usr/lib/apache2/modules/mod_authn_file.so
authnz_ldap	To enable the ldap based auth provider	LoadModule authnz_ldap_module /usr/lib/apache2/modules/mod_authnz_ldap.so
ldap	Required by authnz_ldap	LoadModule ldap_module /usr/lib/apache2/modules/mod_ldap.so

The VirtualHost configuration

<VirtualHost *:80>
    ServerAdmin webmaster@somewhere
    ServerName myserver.mydomain
    DocumentRoot /var/www

    WSGIScriptAlias /Copan /var/www/Trac/Trac.wsgi

   <Location "/Trac">
        WSGIApplicationGroup %{GLOBAL}
        SVNPath /var/svn/Svn

        AuthType Basic
        AuthName "My Trac installation"
        AuthBasicProvider ldap file

        #You may be able to use anonymous queries, thus without
        #using the Binding
        AuthLDAPBindDN "MYDOMAIN\myusername"
        AuthLDAPBindPassword "mypassword"
        #This also works for me, but does not restrict to "user accounts"
        #AuthLDAPURL "ldap://domain_controller:389/DC=mydomain?sAMAccountName?sub?(objectClass=*)" NONE
        AuthLDAPURL "ldap://domain_controller:389/DC=mydomain?sAMAccountName?sub?(objectClass=organizationalPerson)" NONE
        AuthzLDAPAuthoritative Off

        #Fallback on standard htpasswd file
        AuthUserFile /var/svn/passwd

        AuthzSVNAccessFile /var/svn/authz

        Require valid-user

    </Location>
</VirtualHost>

First of all, THIS IS NOT A COMPLETE TUTORIAL. If you want to start from the beginning, read this excellent article on http://diotavelli.net.

This post addresses a feature not described in the article above. What if I want to EXTEND a QT widget and retain all its designer features? For example, what if I want to extend QStackedWidget so that I get features like the "Insert page" menu? It can be done by manipulating the XML returned by the plugin.

Head on to the example plugin code:

#!/usr/bin/env python
# -*- coding: utf-8 -*-
import sys

from PyQt4 import QtGui, QtDesigner
from PyQt4.QtDesigner import QPyDesignerCustomWidgetPlugin

from some.other.package import CSlidingPanel

class CSlidingPanel_plugin(QPyDesignerCustomWidgetPlugin):
    def __init__(self, parent=None):
        QPyDesignerCustomWidgetPlugin.__init__(self)
        self.initialized = False

    def initialize(self, QDesignerFormEditorInterface):
        if self.initialized:
            return
        self.initialized = True

    def isInitialized(self):
        return self.initialized
    
    def createWidget(self, parent):
      return CSlidingPanel(parent)
    
    def name(self):
      return "CSlidingPanel"
    
    def toolTip(self):
        return "Animated sliding panel"
    
    def whatsThis(self):
        return "Animated sliding panel"
    
    def includeFile(self):
      return "some.other.package"
    
    def group(self):
        return "Some group name"
    
    def isContainer(self):
        return True
    
    def domXml(self):
        return ("""
            <ui language="c++">
                <widget class="CSlidingPanel" name="SlidingPanel">
                    <property name="toolTip">
                        <string>{0}</string>
                    </property>
                    <property name="whatsThis">
                        <string>{1}</string>
                    </property>
                    <property name="geometry">
                        <rect>
                            <x>0</x>
                            <y>0</y>
                            <width>500</width>
                            <height>200</height>
                        </rect>
                    </property>
                    <property name="minimumSize">
                        <rect>
                            <width>500</width>
                            <height>500</height>
                        </rect>
                    </property>
                    <property name="styleSheet">
                        <string>background-color: rgb(184, 184, 184);</string>
                    </property>
                </widget>
                <customwidgets>
                    <customwidget>
                        <class>CSlidingPanel</class>
                        <extends>QStackedWidget</extends>
                    </customwidget>
               </customwidgets>
            </ui>""".format(self.toolTip(), self.whatsThis())
        )

Pay attention to:

• The includeFile method which should return the full module name (packages included) for the widget being exported
• The language="c++" attribute of the ui XML element. I don't know why but it seems to be required
• The <customwidgets> section and in particular the <extends> element which defines the inheritance as far as QT Designer is concerned

These days I'm building a C++ Win32 client for a .NET based SOAP web service with basicHttpBinding, using Apache AXIS2/C 1.6 as client library and its WSDL2C code generator.
I don't have any in-depth knowledge of this stuff so for me it's mainly a research over the Internet and the documentation trying to put something together; IMHO SOAP is a somewhat complex technology without any real benefit, I'd prefer the much simpler and just as effective XMLRPC.
Enough rambling; the following code is based on the code created by AXIS2/C tool WSDL2C. I am sending a custom Cookie header to cope with some ASP auth scheme.

const char headerName[] = "Cookie";
const char headerValue[] = ".ASPXFORMSAUTH=6131A0.......";

//Initialize the environment somewhere
//Client stub to the SOAP service
axis2_stub_t* stub = axis2_stub_create_SecureWCF(environmentPointer, axisHomeDir, NULL);
//Get the options for the client stub
axis2_options_t* options = axis2_stub_get_options (stub, environmentPointer);
//Create a list of headers to be sent along with the HTTP request and add them to the stub
axutil_array_list_t* headers = axutil_array_list_create (environmentPointer, 1);
axis2_http_header_t* http_header = axis2_http_header_create(environmentPointer, headerName, headerValue);
axutil_array_list_add (headers, environmentPointer, http_header);
axis2_options_set_http_headers(options, environmentPointer, headers);

Voilà, the new header will be sent with the next op() call.
Another option for the C++ guys is the WSO2 Web Services Framework. Here's a snip that does the same job as above

std::string wso2Home("somewhere");

org_tempuri::SecureWCFStub* stub = new org_tempuri::SecureWCFStub(wso2Home);
org_datacontract_schemas__2004__07_webapplication1::GetTestObjectSVC* svc = 
    new org_datacontract_schemas__2004__07_webapplication1::GetTestObjectSVC();

axis2_options_t* options = stub->getOptions()->getAxis2Options();
const axutil_env_t* env = wso2wsf::Environment::getEnv();

axutil_array_list_t* headers = axutil_array_list_create (env, 1);
axis2_http_header_t* http_header = axis2_http_header_create(env, headerName, headerValue);
axutil_array_list_add(headers, env, http_header);
axis2_options_set_http_headers(options, env, headers);

Suppose you want to statically link some libraries and some others dynamically when building your executable with the GCC/G++ toolchain. For sure you can pass the static library with its full path along with the other objects, something like:

gcc -o executable anObject.o /usr/lib/someLibrary.a

But what if I don't want to search library full path? What if I want to use standard library search path? In short: how the hell do I tell the linker to manage this stuff?

Here are the relevant options:

• Option -Wl,-Bstatic tells the linker to link all the subsequent libraries as STATIC
• Option -Wl,-Bdynamic tells the linker to link all the subsequent libraries as DYNAMIC

WARNING: linking order in the static chain matters!

An example:

gcc -o executable anObject.o -Wl,-Bstatic -lstaticLib -Wl,-Bdynamic -ldl -lpthread -ldynamicLib

libstaticLib.a will be used, even if a libstaticLib.so is available.

I had to face this problem: copy the already read email from IMAP accounts to local mailboxes, for backup purposes. I did't find any software which could handle the job by itself. Mutt has everything that's needed, but there's no way to tell "do nothing and quit if there are no already read messages" and besides it's designed to be an interactive tool. So I had to wrap up a simple script; here's what you need:

• bash
• Maildrop
• IMAPfilter
• Fetchmail

IMAPfilter is used to select already read messages in the mailbox (additional conditions can be specified here), fetchmail does the download and maildrop performs the local delivery. The already read mail is first moved to a /backup folder, then downloaded and flushed. So the FIRST THING TO DO is create a /backup subfolder. Mail is downloaded locally to a gzip-compressed mbox, one for each month.

DATE=$(date +"%b-%Y")

TMPFILE=$(mktemp /tmp/mailbackup.XXXXXXXXXXX)

echo "----> Backup mail for myUser"
MAILBOX="/backups/mail/myUser-$DATE.gz"
gzip -qdc "$MAILBOX" > $TMPFILE
imapfilter -e "account=IMAP{server = 'imap.server',username = 'myUser',password = 'passwd'} \
msg = account.INBOX:is_seen() account.INBOX:move_messages(account.backup, msg)"
fetchmail -a -r backup -m "/usr/bin/maildrop ~/.maildroprc $TMPFILE" -f ~/.fetchmailrc-myUser
gzip -c -9 $TMPFILE > $MAILBOX

Commands should be on a single line, so join the lines split by a '\'
Here's the .maildroprc

DATE=`date +"%b-%Y"`
VERBOSE=2
#Mailbox name is the first user argument
MAILBOX="$1"

to $MAILBOX

The .fetchmailrc-myUser

poll imap.server protocol IMAP user "myUser" password "passwd"

Sometimes I use PyInstaller to build standalone (AKA frozen) executable distributions of Python sofware. As a package builder I've been happily using EPM for a while.
You may have seen an error like the following when you try to execute the program from EPM-built package:

Cannot open self ./myProg or archive ./myProg.pkg

This is all because of executable stripping. By default EPM strips executables, and frozen python exes don't work anymore. So, in your EPM list file, remember to use the nostrip() directive, like here:

f 755 user group ${prefix}/myProg ${srcdir}/myProg nostrip()

If you want to deploy stripped executables, use strip=True as arguments in your EXE and COLLECT objects in the PyInstaller spec file.