Simone 'Wiz' Tellini - Software Development

I do hate MySQL

nospam@example.com (Simone) — Fri, 12 Jun 2009 10:25:43 -0700

I can't stand it, really.

I'm using it at the moment only because I'm forced to. I can't help considering it as a toy DB, especially comparing it to PostgreSQL or SQL Server or any other serious RDBMS.

The last thing that hit my nerve is this bug. Come on... can't it even handle foreign keys in a decent way?

And I'm trying not to think that it allows you to define a table with foreign keys and silently ignore them.

</rant>

FeedbackReporter and Mantis integration

nospam@example.com (Simone) — Sat, 02 May 2009 14:23:30 -0700

I've just written a small script to store reports generated by FeedbackReporter on a Mantis system.

It should come handy to keep track of your application crashes (crash? uh? My apps never crash! )

You can download it here.

Simply extract it on your webserver and edit config.php to adjust your settings.

It can work in two ways: if you're installing it on the same server running mantis, set MANTIS_LOCAL to true and MANTIS_PATH to the root of your mantis installation.

Otherwise, you can even install it on a different server and let it communicate with your mantis via SOAP. In order to do this, set MANTIS_LOCAL to false and edit MANTIS_URL. This setup needs the SOAP extension of PHP5, so make sure it's available.

Feedback is welcome, of course.

Connection without a listening socket?

nospam@example.com (Simone) — Tue, 10 Feb 2009 12:25:00 -0800

I've encountered a network problem that left me a bit puzzled.

Usually, when you want to setup a TCP connection between two sockets (let's called them S1 and S2), you need to perform these steps on the server:

create S1

bind S1 to a port (or to a port/address, if you need to)

set S1 to listen for connections

on the client:

create S2

tell S2 to connect to S1

and finally, the server receives the connection attempt and accepts it to remove it from the backlog of the listening socket.

Well, apparently there's another way, which I suspect few know about.

Check out this simple application:

using System;
using System.Net;
using System.Net.Sockets;
using System.Threading;

namespace SocketTest
{
class Program
{
private static IPEndPoint a = new IPEndPoint( IPAddress.Loopback, 18000 );
private static IPEndPoint b = new IPEndPoint( IPAddress.Loopback, 18001 );

static void Main(string[] args)
{
Thread threadA = new Thread( ThreadA );
Thread threadB = new Thread( ThreadB );

threadB.Start();
threadA.Start();
}

static void ThreadA()
{
for(;;) {
try {
Socket sock = new Socket( a.AddressFamily, SocketType.Stream, ProtocolType.Tcp );

sock.Bind( b );
sock.Connect( a );

Console.WriteLine( "Connected from " + b + " to " + a );

Thread.Sleep( Timeout.Infinite );
}
catch {
}
}
}

static void ThreadB()
{
for(;;) {
try {
Socket sock = new Socket( b.AddressFamily, SocketType.Stream, ProtocolType.Tcp );

sock.Bind( a );
sock.Connect( b );

Console.WriteLine( "Connected from " + a + " to " + b );

Thread.Sleep( Timeout.Infinite );
}
catch {
}
}
}
}
}

Let's see what happens compiling and running it:

Yes, you get a connection between the two sockets. It's something I was never told about and I would have never expected to see. But it's correct as per RFC 793.

I've found out about it when a couple of clients on a test system started connecting between themselves, while they were supposed to connect to 2 different servers on localhost. Both clients were periodically trying to connect to their server, which was down. They were creating a new socket for every attempt, without binding it, thus Windows' TCP/IP stack assigned a sequential port number to them. Eventually, the two processes reached the configuration required for the "Simultaneous Connection Synchronization" as described in the RFC.

Interesting. I guess you live and you learn.

Quote of the Day

nospam@example.com (Simone) — Mon, 04 Aug 2008 22:53:09 -0700

[...] a good programmer cultivates the virtue of laziness. (But not just any laziness: you must be aggressively, proactively lazy!)

-- Chris Pine, dealing with the DRY rule

Quote of the Day

nospam@example.com (Simone) — Sat, 12 Jul 2008 12:15:14 -0700

Programming today is a race between software engineers striving to build bigger and better idiot-proof programs, and the Universe trying to produce bigger and better idiots. So far, the Universe is winning.

-- Richard Cook, read on FSM

Quote of the Day

nospam@example.com (Simone) — Sun, 06 Jul 2008 21:21:04 -0700

[...] as we all know, premature optimization is the root of all evil.

-- from a cocoa-dev email

amen!

Single instance per session application with argument passing

nospam@example.com (Simone) — Wed, 04 Jun 2008 12:44:00 -0700

Suppose you want to create an application which you want to limit to a single running instance, like some image viewers do for example. Also, you need different users on a Terminal Server not to conflict with each other. Oh, and you're coding in C#.

A possible solution to the problem is the one described in this Flawless Code post.

What I didn't like is, it requires the 3.5 version of the framework, while I preferred to be able to run on machines with only 2.0 installed. So I've replaced the NamedPipeClientStream/NamedPipeServerStream with a P/Invoke-based implementation based on this example.

Also, I create a different pipe per user session, since I haven't found a way to create a "local" pipe. An article at Dr Dobb's suggests that it's possible, while MSDN doesn't mention this possibility...

I've zipped up my test solution in case someone else has the same needs.

How much would it cost to rewrite it?

nospam@example.com (Simone) — Mon, 02 Jun 2008 17:13:44 -0700

I've just added LogMiner to Ohloh, a site that offers an interesting feature: it can analyse a project source code and estimate how much it would cost to hire a development team to recreate the project from scratch.

I think that it's a simple way to estimate the effort you put over the time in open source projects.

Here's the resulting figure for LogMiner:

If at first you don't succeed... try, try again...

nospam@example.com (Simone) — Mon, 19 May 2008 16:57:02 -0700

Here's another interesting piece of code I've just digged up in a C# application I'm reviewing.

If you can come up with a bright idea about what those try/catch blocks are supposed to do, you've got more imagination than me...

class Foo
{
private int bar;

// ...snip...

public int GetNewBar()
{
int ret = 0;

lock (this)
{
try {
bar++;
ret = bar;
}
catch {
try {
bar++;
ret = bar;
}
catch {}
}
}

return ret;
}

// ...snip...
}

Don't code like this at home, kids

nospam@example.com (Simone) — Wed, 14 May 2008 10:25:13 -0700

Sometimes when you're reviewing someone else's code, you find interesting pieces of "art". Like the following loop construct in a C# application:

int i = anArray.Length - 1;

while (true)
{
if (i < 0)
break;
else
{
...
i--;
}
}

Doh.

You might be tempted to rewrite it as:

for( int i = anArray.Length - 1; i >= 0; i-- ) {
...
}

but it would be too simple, too boring, wouldn't it?

Quote of the Day

nospam@example.com (Simone) — Fri, 18 Apr 2008 12:07:59 -0700

I always assumed the designers of the Objective-C language and Cocoa frameworks trust me to know what I'm doing while I find very elegant ways to shoot myself in the foot.

-- from a mail sent to Cocoa-dev

PHP templates

nospam@example.com (Simone) — Thu, 24 Jan 2008 22:21:31 -0800

When you design a complex application, it's always useful to separate the presentation layer from the business logic. Usually, when dealing with web apps, this involves handing templates which are "filled" with data by your controller.

The common rationale is that the template should contain as little logic as possible and that it should be easily editable by non-programmers.

There are several PHP packages (Smarty, PHPTal, etc...) that aim at providing a flexible and neat template framework. They usually try to force you not to put code in the template by providing a limited set of directives that can be used, such as placeholders, foreach-style control blocks, conditional blocks... but in the end, every framework I've seen also includes a directive that allows you to put arbitrary PHP code inside the template.

Thus I've always wondered: what's the point? Why don't you just use PHP as a templating engine instead of inventing a new language?

So, let me introduce my super-duper template class!

<?php
/*
* Copyright ©2007-2008 by Simone Tellini
* http://tellini.info/
*/

class Template
{
private static $commonVars = array();
private $tpl;
private $vars;

public function __construct( $template )
{
$this->tpl = $template;
$this->vars = self::$commonVars;
}

// you might set some variables before instantiating your templates when
// you want to have some data available in all the templates you need to
// build a complex page
public static function setCommon( $key, $value )
{
self::$commonVars[ $key ] = $value;
}

public function set( $key, $value )
{
$this->vars[ $key ] = $value;
}

public function execute()
{
$T = $this->vars;

ob_start();

include( $this->tpl );

$html = ob_get_contents();

ob_end_clean();

return( $html );
}
}

?>

Ta-da!

A sample template may look like:

<html>
<body>

Hello, <?= $T[ 'user' ] ?>!

</body>
</html>

The $T array is the only variable visible to the template (except for $this, which points to the Template instance and the super-variables $_GET, $_POST, etc...), since it's the only variable defined in the scope where the template has been include()'d

The controller might look like this:

<?
require_once( 'classes/template.php' );

$tpl = new Template( 'templates/helloWorld.php' );

$tpl->set( 'user', 'World' );

print( $tpl->execute() );
?>

I've been using a class like that for a couple of years now, without any regret. It's been used in sites whose content and graphics are now constantly edited by their owners: even though they aren't programmers, they never had any problem whatsoever in editing or creating their own templates. I truly like it beacuse:

it's simple

it surely isn't bloated

it's faster than many (most?) of the other templating engines

it's as flexible as you need

Of course, the fact that you're using PHP to create your templates shouldn't tempt you to start placing business logic inside them!

Attributes, reflection and PHP

nospam@example.com (Simone) — Sun, 13 Jan 2008 12:16:00 -0800

After a long exposure to .NET, when I get back to PHP I sometimes find myself missing some cool functionality (and a good IDE too... none of those I've tried so far can compete with the comfort of VisualStudio + ReSharper).

Among these missing features, attributes are what I needed in order to solve a certain problem in an elegant way.

Looking at what the reflection API was offering, I came up with a simple way to decorate my properties/methods/classes with attributes: through the documentation comments!

Let's see an example:

<pre>
<?

class A
{
/**
* @attr
* @attr2
*/
protected $foo;

protected function GetPropAttrs( $property )
{
$class = new ReflectionClass( get_class( $this ));
$prop = $class->getProperty( $property );
$doc = $prop->getDocComment();
$attrs = array();
$matches = array();

preg_match_all( '/@([a-zA-Z_][a-zA-Z_0-9]*)\s/m',
$doc, $matches, PREG_SET_ORDER );

foreach( $matches as $m )
$attrs[] = $m[ 1 ];

return( $attrs );
}

public function test()
{
$attrs = $this->GetPropAttrs( 'foo' );

print_r( $attrs );
}
}

class B
extends A
{
/**
* @attr3
*/
protected $foo;
}

$obj = new A();

print( 'Test A:<br>' );
$obj->test();

$obj = new B();

print( '<p>Test B:<br>' );
$obj->test();

?>
</pre>

Running the above script you'll see the following output, confirming that everything works as desired:



Test A:

Array

(

    [0] => attr

    [1] => attr2

)



Test B:

Array ( [0] => attr3 )

If you're using the /** */ comments for real documentation, you might want to use another escape character rather than @, to avoid problems.

Happy coding

Is MySQL bad for (open source) software quality?

nospam@example.com (Simone) — Fri, 04 Jan 2008 15:54:00 -0800

I think so.

Why? Because it's basically a toy DBMS with too many non-standard behaviours.

A lot of open source projects choose MySQL mainly because it's widely available among hosting providers, while better RDBMS like PostgreSQL¹ are hard to find. The real problem is, a lot of open source developer know very little about SQL and databases: they make things up as they proceed in their project, almost often caring only that "it works" for them². The result is that they learn wrong habits and write poor DB-related code, a trait common to many projects, sadly.

For instance, the inspiration for this rant comes from this bug report for Flyspray, a bug reporting system written in PHP. The report is about a pretty lame error in the SQL query used to fetch the task list, as detailed in the 9th comment. PostgreSQL is not "incredibly annoying regarding GROUP BY" as the Flyspray maintainer believes: PostgreSQL simply does the right thing to do in that case, just like every other serious³ RDBMS on the market. I was recently hired to port Flyspray to Oracle: needless to say that I had to fix this problem as well as several other queries.

Other examples of DB abuse include Mantis, another PHP bugtracker. If you had a chance to use it, probably you've seen the statistics at the bottom of the page, something like:



43 total queries executed.

30 unique queries executed.

I beg your pardon? 43 queries to display a 10 rows table? and why 13 of them have been performed twice?

So, please... kids don't use MySQL at home.

I'm not saying that you should never use MySQL (there would be plenty of reasons to avoid it anyway ) or that projects based on it suck (I like mantis, for one). It's simply a bad choice for a beginner wishing to learn how to do things properly, IMHO.

¹ my favorite open source RDBMS. If it seems like I'm doing some advocacy, well... maybe I am
² I've got nothing against this attitude when it comes to free software: if you want things to get done as you wish, you should be prepared to do them yourself or pay for them, instead of expecting other people to do the dirty work for free
³ yes, that excludes MySQL

The meaning of const in C

nospam@example.com (Simone) — Sat, 08 Dec 2007 14:06:27 -0800

John Engelhart posted to Cocoa-dev a detailed explanation of what exactly const means in C programs.

Keep in mind that he's talking about C, not C++ (where const has a stronger meaning and even a slightly different semantic when used to declare constant values).

Still, it's worth reading.