Here is how the latest iTunes 10 application under Windows 7 on a computer that has the DPI setting set to 150% (“Larger”) also known as 144ppi.

Steve, what kind of QA is the iTunes team doing? As you can see this is not a minor/cosmetic issue anymore.

Limited support for ignoring files or directories

If you used Subversion, Git or others probably you are already used to mark some files or directories to be ignored. On these systems ignore patterns are stored on the server and they do apply for all the users. This does make perfectly sense for me, if you want to stop your SCM from looking for .pyc files or temp directories you add the rule and everybody will benefit from it.

Now if you try to obtain the same behavior in Perforce you will be surprised that what their support for ignoring files is not what you would expect.

First, you can define the ignoring patterns in your clientspec, so they will be available only to this clientspec, meaning that you will have to multiply them in all your clientspecs. Sharing them with others? … yep copy/paste.

Now lets try to have a deeper look on how they really work:

• partially supported only in GUI (p4v) – for example they will continue to show in file tree and even when you use the Reconcile Offline Files. In fact the only place where they are working is on Advanced Reconcile – a place that you probably do not need/want to use.
• forking? …. you just have to recreate the ignore pattern in the view.

Now imagine that I do work a lot with Python and that entire source tree is full with temporary .pyc files.

Mapping the same repository file/directory to multiple locations on disk

Nope, this is not something you can do – you can try but it will fail to work properly. Trust me, it will save you precious time.

Having workspaces mapped to the same directory

If you do not want troubles, do not forget to keep each workspace (clientspec) in its own directory and be sure that these directories do not overlap, even partially.

Running as root under OS X, not quite

The login is not working well on OS X if you run perforce as root, this is happening because the system does not have HOME environment variable for this user. Do you what what was the response from support: why do you run Perforce as root? WTF!

I don’t want to imply that perforce support in not good, in fact I have to confess that they are very well prepared and give well documented answers and really fast. In fact the above case was the only mistake they made.

Workspaces (clientspecs) that are too hard to maintain

Do you have a big projects with tens of developers if not hundreds, and you change the project structure – now each developer will have to update all his workspaces to match the new one.

It was so hard for Perforce to add something like an @import to the clientspec, so it could just base a workspace on another one?

Now another “cool”, feature is that it is not possible just to copy and paste the view definition from one workspace to another because you will have to do some find-n-replace on it to replace the original clientspec name from each line with your name.

It’s true that there is an option to clone a clientspec, but this will create a new clientspec and you will still be required to move your customizations to it.

Personal conclusions

First, I want to say that even with the problems presented above, Perforce is a very reliable tool and I am not aware about any open-source solution that would be able to deal with repositories of hundreds of gigabytes and a changelist history of 6 digits.

If you know other workarounds for Perforce let me know.

Installing Tomcat

Here is the quick guide:

• install macports
  • suport port upgrade
  • sudo port upgrade outdated.
• sudo port install tomcat6, or if you want other version check port list|grep tomcat
• config is now at: /opt/local/share/java/tomcat6/conf
• startup script: /opt/local/share/java/tomcat6/bin/tomcatctl

cp /opt/local/share/java/tomcat6/conf/tomcat-users.xml.sample /opt/local/share/java/tomcat6/conf/tomcat-users.xml
nano /opt/local/share/java/tomcat6/conf/tomcat-users.xml

Activate at least the tomcat user and make it a manager.

For running tomcat as a daemon on port 8080 but enable also port 80 by using a firewall redirection edit /opt/local/share/java/tomcat6/conf/server.xml and add proxyport="80" URIEncoding="UTF-8" inside <Connector .../>.

For forwarding port 80 to 8080 run this line and add it do /bin/catalina.sh:

ipfw add 100 fwd 127.0.0.1,8080 tcp from any to any 80 in

Set proper memory limits for Java machine running tomcat or you may be in trouble later. Inside/opt/local/share/java/tomcat6/conf/local.env

export JAVA_JVM_VERSION=CurrentJDK
export JAVA_OPTS="-Xmx3000M -Xms3000M -Djava.awt.headless=true -Duser.timezone=UTC"

In my example, I allocated ~3Gb or RAM but you can adapt this, anyway don’t put less than 1GB if you are running hudson inside tomcat.

Running as a service (daemon)

Run nano /Library/LaunchDaemons/org.apache.tomcat.plist and paste the code below:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple Computer//DTD PLIST 1.0//EN"     "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
    <key>Disabled</key>
    <false/>
    <key>Label</key>
    <string>org.apache.tomcat</string>
    <key>ProgramArguments</key>
    <array>
            <string>/opt/local/share/java/tomcat6/bin/catalina.sh</string>
            <string>run</string>
    </array>
    <key>RunAtLoad</key>
    <true/>
</dict>
</plist>

Check that launchd detected your new daemon, if not reboot

launchctl list|grep tomcat

Start tomcat manually.

launchctl start org.apache.tomcat

If the status is something else than -, you have a problem and you should investigate it: launchctl log level debug and check /var/log/system.log.

Adding hudson to tomcat
This is the easy part, just do:

cd /opt/local/share/java/tomcat6/webapps
wget http://hudson-ci.org/latest/hudson.war
chown _www:_www hudson.war

Now go to http://localhost/manager/ and start hudson.

I created two test Win32/Unicode native applications one using MFC and one without MFC – the issue appear on both. I didn’t bothered to test .NET because I’m not interested in it.

Take a look at my screenshots below. It’s really simple to reproduce it yourself: just use the MSDEV Wizard to create a new application and paste some Unicode text into one of the dialogs.

Some interesting thing to consider:

• I didn’t change the font in the dialog – it’s using the default value: “MS Shell Dlg”
• If you modify the language of dialog from the default “English (United States)” to a language that does contains the testing characters they will display – but this is insane! It does mean that is still using codepages?! and worse that you cannot create a dialog containing characters from different scripts?
• Isn’t Windows supposed to do font face substitution?
• I tested only on Windows XP SP3.

Sad but this does look like a What You See is What you WON’T Get.

Later, same … day

Let’s see the list of bugs I’ve just discovered on Microsoft Connect using it’s “very sad” search feature

• By default the RC Files are not Unicode, even on Unicode projects, duh!
• Resource editor doesn’t tell you to save the files as Unicode when you have Unicode characters on it. So you loose them without knowing it!
• Resource editor doesn’t support UTF-8 – sad!
• Resource compiler doesn’t support UTF-8 – sad!

Bugs found on Microsoft Connect: 176612, 100833, 162872. Some of them being submitted 3 years ago for Visual Studio 2005: None of them solved even in newer versions like 2008.

And the DIY formula

Open the RC file using text editor and save it using File > Advanced Save Options and select Unicode – Codepage 1200. It took me quite some time to figure it out just because several people considered that Unicode related bugs are not so important. Maybe we can suggest to rename the Unicode option from the Application Wizard to “Partially Unicode”.

First we should identify the source of language related data we could use for making a decision:

1. HTTP_ACCEPT_LANGUAGE – included in any HTTP request. Something like:

   en-ca,en-gb;q=0.7,en;q=0.3
   "q" parametter indicates the quality (higher is better).

2. user preference: a cookie or user preference if the user is already logged in
   
3. your list of supported languages
4. (optional) Your list of language fallback preferences, in case you do not have one we can consider that it is “en”
5. (optional) HTTP_ACCEPT_CHARSET.
6. (optional) IP address – this can be converted to a country or language.
   

Basic rules

1. If user had a preference (cookie/profile) and it’s a valid we use it – return
2. We try to match all language variants existing in HTTP_ACCEPT_LANGUAGE with our list of supported languages and select first match – return
3. If we have language fallback list we choose the first language from it – return
4. We’ll choose ‘en’ if it is present in the supported language list – return.
5. We’ll choose first available language from the supported language list – return.

Remarks

• I ignore the requested charset because I prefer to use UTF-8 everywhere. I don’t want to convince you it’s the necessary best choice but from my point of view it is pretty good. I’m aware that there are still mobile devices on the market without UTF-8 support.
• I ignore the IP information because I consider it not so relevant – there are many people traveling and you’ll get his current location, also it’s harder to implement.
• Possible future improvements: create a small list of languages user may know (probability). Let’s say I have an website in 50 languages and I want to show to the users only the first 2-3 languages and a selection control (combobox) for the entire list. Here it could be very useful to use the information provided by the IP address.

Questions still to be answered

Could we figure a way of displaying combined content based on user preference? It’s high likely that your website may not contain all articles in all languages and that some users will want to real articles in several languages.

How can we display this option to the user? Remember that you can’t display a huge list of checkboxes to the user – it should be something small and intuitive.

Also check the list every time you add a language because it’s another common mistake to assume that two letter domain names or country codes are the same as the language codes.

An even a better idea could be to not use images at all

You may find additional resources at: sil.org or at W3C – using lang

Keyboard shortcuts in green and mnemonics in yellow.

The best keyboard related definition I’ve found for mnemonic it was on Sun Java glossary:

mnemonic – An underlined alphanumeric character, typically in a menu title, menu item, or the text of a button or component. A mnemonic shows the user which key to press (in conjunction with the Alt key) to activate a command or navigate to a component.

If you had Win32 experience you may call them “menu accelerators”. They define the letter used for triggering them by using Ampersand (&) inside the menu text. I would not recommend the usage of “accelerator” term because looking over the net I’ve found that in many cases it does refer to keyboard shortcuts.

If you are wondering why it’s important to clearly distinguish between them: it’s because the mnemonics are changing when the UI language is localized but the keyboard shortcuts are not supposed to be localized.
.

Also if you encounter pages in other encodings you should convert them into UTF-8.

What you should know about UTF-8:

• Any character can be encoded in UTF-8, converting any other encoding to UTF-8 can be done without loosing data
• UTF-8 is the de facto standard on all internet related protocols.
• Any ASCII text is a valid UTF-8 text
• UTF-8 is the most simple Unicode encoding and it’s the only one that is not dependent of the byte-ordering.
• It’s best to mark UTF-8 as the default encoding for any page you create.

The only real disadvantage I’ve discovered when using UTF-8 encoding is that the encoded text is larger than for some languages like the Asian ones. Still this should be no problem if you enable HTTP compression.

When the usage of UTF-8 can break things:

• If the text is going to be used on ancient devices that are able to use only 8-bit characters. Like: TV related equipments, or old mobile phones.

I’ve did some research to see how different companies are dealing with the shortcuts when people are using different languages and different keyboard layouts. Here are a short list of my results:

• If possible the shortcut is not changed
• If a key is not available on a specific layout we should try to use the same physical position
• There should be no link between application language and keyboard layouts. It’s a mistake to consider that application language will match the keyboard layout. I’ll give you just an example: in Romania over 95% of the computers are using US keyboard layout and the other 5% are using one of the 4-5 different Romanian keyboard layouts.
• some people are using multiple keyboard layouts – usually they are power users and we can’t ignore them
• keyboard layouts can be switched anytime
• characters are safe to be used only if they are Latin

Physical position rules

I think that the physical position of a key is more important than the letter printed on it. Here are my arguments:

• human brain is learning the shortcuts by hand movements (see Pavlov’s dog experiment)
• it’s compatible with the usage of multiple keyboard layouts

But it’s not always so simple: we can’t expect that the French users would swap the usage of Ctrl-Q with Ctrl-A ( Quit vs Select All) just because they are using different layouts. So a basic rule would be to stick to the layout for letters and numbers BUT try to use the physical position for other keys.

Anyway is high likely that people using multiple layouts would choose alike layouts (derived from the same root).

One-to-one key matching

Because the number of keys is the same we have to find a way of remapping the keys so all shortcuts working on US to be usable on other layouts. When I’m referring to keys, I’m referring to the hardware keys.

Case study

Question: How do I make OS X tell me that the key pressed by the user was ‘q’ when current layout is Arabic?

Rules:

• we are not allowed to assume it’s a ‘q’ just because the key code is the same, maybe it’s a Arabic DVORAK keyboard!.
• the shortcuts cannot be hardcoded (user can edit them)

On Windows it does work 98% using virtual key codes and you can use them to extract this info, but on OS X the key codes (even Apple call them sometimes virtual key codes) they are some kind of scan codes. For the rest of 2% you have to do some small hacks with the OEMs.

Current code


case kEventRawKeyDown:
{
unsigned char macChar;
unsigned int keycode;
unsigned int modifiers;
unsigned int keyboardtype;

// kEventParamKeyMacCharCodes
// for Latin characters it does behaves like Windows
// for !Latin it does not return consistent data: on Arabic it does return the same as on US but on French it does return French specific characters for the OEM keys.
...


• VK_A code is present on any keyboard layout even if the character produced is not A

• usually is on the same position, it does move only on some layouts like French AZERTY, and it’s logical to be moved – if the keyboard has an “A” printed on it user is supposed to press this button – we can’t request French users to switch Ctrl-Q and Ctrl-A just because they have a different keyboard layout.

• we can’t rely on the characters because many layouts do not produce Latin characters or some keys are not even producing a character (dead keys).

• it’s not safe to use the scan codes because they will break on AZERTY, DVORAK and other layouts.

I think I do not have to explain more why virtual keys are the way we must store and process keyboard shortcuts.
And now let’s try the same thing on Apple OS X platform. It does seem that apple has only 2 properties for keyboard events: key codes (kEventParamKeyCode) and characters. They consider key codes to be some sort of virtual keys, but in fact they are just physical key positions and this does mean that they are some sort of hardware independent scan codes.
So now I’m stuck with this big question:

How do I detect that the user pressed Command+A under OS X in a way that will work with any keyboard layout?

I should add that the user can have multiple keyboard layouts installed.

External links

• Carbon keyboard events
• Virtual keys on Apple
• http://rentzsch.com/macosx/virtualKeyCodes – Article about Virtual Key Codes on OS X
• Carbon Event Manager Reference
• Text Input Event Handling for Applications With Carbon
• Losing the character code when using the control key