Runsmall

Remote git with XCode 4

Remote SVN has been working great for me as a combination of version control and backup.
Here is how I set it up. But I like the Git architecture of multiple distributed archives better, so I decided to switch.

I wanted a really simple setup:

• A project archive locally for each project.
• A remote archive for each project that I could push changes to periodically.

For the remote archive I wanted to use my Linux server running Centos. It's in my office connected by a fast cable system (>5 mbps upload) to the Internet.

Setting this up should have been simple, but it wasn't. XCode behaved strangely and frequently crashed. I got strange error messages; for example when I successfully connected to an empty remote repository, push wouldn't work, claiming that there were changes in the remote that needed to be pulled first. I finally succeeded.

Here are the steps I followed to finally get things working. I got some of the ideas from here.

1. First of all, do all command line work with XCode closed. You shouldn't have to, but everything seemed to work better and some errors only went away when I started doing this.
2. Set up your .gitignore. It is a file named ".gitignore" that you put in the root of your project, and it tells git which files not to track. You must do this before running any git commands. Here's mine, copied in part from several sources, with additions. Most of the entries are things in the XCode project that reflect UI state and can change rapidly, making it impossible to achieve a cleanly checked-in archive.

$ cat .gitignore 
xcuserdata
*.xcuserstate
build/
*.pbxuser
*.perspective
*.perspectivev3
*.old

3.  Follow the process documented in the XCode Organizer Documentation tab to initialize a local project:

git init
git add .
git commit -m "initial commit"

4. At this point you can start XCode and check that it finds your local repository, and allows you to commit updates. Quit XCode again for the next step.
5. Create a bare copy of your local repository to upload to the server. Note the --bare option on the git clone command.

cd ..
git clone --bare test
Cloning into bare repository test.git...
done.
tar cfvz test.tgz test.git

6. Upload the compressed tar (the .tgz) to your server, and login to the server.
7. On the server, uncompress the tgz in the directory where your remote will live. I created a user named git and placed mine under /home/git.

   tar xvfz test.tgz
   

8. Start XCode and add the remote. Three key points:
   • The remote is added under the "Remotes" folder of your local archive, not as a new archive. This is very confusing at first, as you can't find anything in the interface to tell you where or how to configure a remote until you click the magic icon.
   • The remote needs to be specified with its complete path, e.g.

     ssh+git://git@myserver.com/home/git/test.git
     

   • XCode is very picky about remote names. It will let you enter the name and the complete URL, only to reject the name and erase your URL, forcing you to start from scratch. The one rule I have detected is no spaces; there may be others.
9. You should now be done. Try doing a push to see if it works.

The configuration needed on my Centos server was minimal. I'm assuming this is all working through Apache WebDAV, but I am surprised and pleased not to have found myself in a permission-setting, daemon-configuring fire drill. 

Overal, this procedure seems needlessly long and complex. It should be possible to create this simple configuration using only git and XCode. But it works well now. If anyone knows a better way I'd love to hear it.

Windows 7 - An error has occurred. Not all the features were successfully changed.

Solved this one thanks to an entry at sevenforums.com.

If you're getting this error, there's a good chance you've moved your ProgramData or Program Files directory to a different partition because you have an SSD boot drive. Foe at sevenforums says he moved Program Files; I just moved ProgramData.

Ignore the Windows solutions, and move the $$PendingFiles directory to the partition that contains ProgramData.

Use the Advanced option of the Security tab of File Properties to take ownership of $$PendingFiles and each of the files in it, change permissions for Administrators to Full Control, and delete them. 

Rename or delete $$PendingFiles, create a new $$PendingFiles on the partition that has ProgramData, and (assuming the new partition is d:) do

mklink -j c:\$$PendingFiles d:\$$pendingFiles

Problem solved!

Gestures in iPhone programming

I believe in source code. UI design is best in an interactive environment, but everything else about software can and should be expressed in human-readable text.

In iOS programming, you can't do everything in source code. Some of your work has to be done with mouse gestures in Interface Builder. You cannot link controls to events or variables in source code; you must use mouse gestures. 

To read a control's data, set its text, or change its appearance programmatically, the control must be connected to a variable of the correct type declared with the IBOutlet modifier.

To respond to an event in a control, a procedure declared with the IBAction modifier must be connected to the event.

The only way to accomplish either of these things is to use a mouse to draw a line between the "File's owner" icon and the event or control of interest. "Fle's owner" represents your view controller.

To link an event to an action: drag from event (e.g. Touch Up Inside) to File's Owner and select IBAction.

To link a control to an outlet: control-drag from File's Owner to the control and select IBOutlet.

Here are examples:

To link an event (Touch Up Inside) to an action, drag from the event in the inspector to File's Owner:

When the mouse button is released you'll see a list of IBActions declared in the view controller. Pick one to be called when the event occurs.

To link a control to an outlet, control-drag from File's Owner to the control:

When you release the mouse button you'll see a list of IBOutlets in the view controller. Pick one to use to access this control.

Overall, this is a reasonably convenient way to create these connections, at least for simple applications. But there really should be a human-readable file in a simple, easy to edit format (not XML) to express these connections and make maintenance easier. Am I missing something? I can't find anything like that in the application build directory.

Remote SVN with XCode

I want to store the source for my iPhone development projects on a remote Linux host using XCode's integrated source management. This gives me remote backup along with source management. XCode supports PerForce, CVS and Subversion, so git wasn't an option. 

I want this to be a private repository, accessible through SSH, and I want to use the same repository for multiple projects.

There are plenty of posts and tutorials about XCode and Subversion, but none i found give comprehensive coverage to the configuration I want. Here's the procedure I eventually came up with:

1. Create a subversion repository (named svn) in my home directory on the Linux host.

[david@myhost ~]$ svnadmin create svn

2. Configure the empty repository in XCode

    select SCM:Configure SCM Repositories

    click + under the repository list on the left

    give it a name and select Subversion as the SCM system

    enter the URL:

    svn+ssh://david@myhost.mydomain.com/home/david/svn

   XCode will parse the URL, do not interfere.

    Enter a user and password 

3. Setup project directories 

I want to use the same repository for multiple projects, so I created a directory for the first project. All the tutorials say to add branch, tags, and trunk, and keep the main line under trunk.  As a lone developer whose repository will be non-public I ignored this and will store the main line directly under the root for this first project. Should the team need to grow beyond me, I'll just move it to a new directory with the conventional structure. 

 4. Import the project

Select the directory to import into. Click the Import icon at the top left of the Repositories dialog, and select the project directory to import. Do not double click! while navigating to the project directory, or XCode will accept the command with no confirmation, and import the entire directory tree. There is no way to cancel this command, except by quitting XCode (and ignoring the dire warnings it strews in your path).

 5. Check the project out

In order to use XCode's SCM, the project has to be in a Subversion working directory. So close the project, then in the SCM:Repositories dialog, pick the checked-in project and click "Check Out". Select or create a new directory for your project and a new working copy will be created. Once you are sure this new copy is OK, delete the original project. From now on you will work in the new working directory. 

That's it! 

 

What Apple should make instead of the Apple Tablet

The concept of an Apple tablet is pretty underwhelming to me. Unless there's some big surprise (an innovative combo E-Ink/LCD screen, for example), I expect to be uninterested. Here are some products that would get me more excited. 

1. An Apple netbook

Steve Jobs doesn't like netbooks, but netbooks are products with a future. They have proven their value as secondary computers for travel or carrying around, or as inexpensive computers for kids or students. 

Apple needs to make a netbook that isn't cheap and crappy like most netbooks are. It should be something like the smart little Sony Vaio W - small, lightweight, and stylish with a 10.1" WXGA screen. The Vaio W easily justifies its $450 base price. A similar Apple netbook priced at, say, $549 would fly off the shelves. It wouldn't cannibalize any current Mac products other than the similarly-priced Mac Mini.

Lots of people install OS X on the $279 Dell Mini-10V to create hackintoshes, so the Apple netbook already exists, but in the cheesy, low-quality format that Steve Jobs detests. Apple is leaving money on the table and denying its customers a product they want.

Specs of the Apple Netbook:

• Intel Atom, preferably a dual-core Mount Olive to save power 
• 1 GB RAM, maximum 2GB
• 160GB hard disk, optional 60GB SSD
• 1366x768 LED-backlit display
• VGA-out as well as mini-DisplayPort, and at least three USB ports
• Price: $549

Why won't Apple build it? Because of fears, unjustified I believe, that it will take sales from the Macbook Air. It won't - the Apple Netbook should be practical and utilitarian for light duty, like a Toyota pickup truck, while the Macbook Air is refined to near impracticality, like a Lotus Elise.

2. A Core i3/i5/i7 Mac Mini

The Mac Mini is stuck in a concept that it needs to be a low-end computer for people who aren't really serious. The problem with that idea is Moore's Law. Even a low-end computer today can do practically anything you want to do, if it's designed and put together right.

Intel's new Arrandale mobile processors are dual core CPUs with integrated graphics, memory controller, and PCIe. They have a lot of promise for a computer with the Mini's footprint and power profile. With integrated graphics and a memory controller on the processor package, they outperform equivalent Core 2 Duos while using less power overall. The ability of the Core I5 and I7 to boost a single core above the CPU's base clock frequency means great performance for most common computing tasks. Something like the Core i7-620M at 2.66 GHz base with 3.33GHz Turbo would wipe the floor with the P8800 Core 2 Duo that is the  current Mini's top CPU, but I am almost as tempted by the Core i7-640LM. At a mere 2.13GHz base it doesn't sound impressive, but Turbo Boost to 2.93 GHz would make it scream for single-threaded tasks (read: most of what most people actually do).  Its TDP of 25 watts only matches the P8800, but includes graphics, a memory hub, and PCIe.

Meanwhile, why not offer the Mini with an SSD? If you have enough DRAM, hard drive speed is the bottleneck for most desktop computers. A Mac Mini using Arrandale processors with the equivalent of the 80GB Intel X-25M would run circles around the current design while using less power.

In the meantime, the base price of the Mini ought to come down. You can get a useful Windows 7 desktop for $249. There's no reason why the Mini should cost more than twice that much.

Specs of the Arrandale Mini:

• A Core i3-350M version for $499 with 2GB DRAM and 160GB, 5400RPM hard drive
• A Core i5-540M version for $599 with 4GB DRAM and 320GB, 7200 RPM hard drive
• A Core i7-620M version for $799 with 8GB DRAM and 320GB, 7200 RPM hard drive
• 80 GB SSD upgrade for $250
• 160 GB SSD upgrade for $500

Why won't Apple build it? They will, but probably not until Arrandale is a year old.

3. A few tweaks for the Macbook Air

The Macbook Air is a beautiful concept. Thin, elegant, and minimalist, it has spawned lots of admiration but few imitators. The same design discipline that created its unsullied sleekness also made it totally unfitted for business. 

As you can see from the picture, the Macbook Air has practically no way to connect to anything, other than by wireless networking. The Mini DisplayPort may be the wave of the future, but for the next few years it means you'll need a dongle to connect to a projector or to most non-Apple displays.

I have been in the boardroom more than once, scheduled to present to the CEO or the SVP. At best you have five minutes to make your laptop work with the VGA port of their projector, and you have to schmooze your immediate client - usually a VP or director, in my business - while you're making it work. And it almost never works right the first time. Little extra pieces of crap to connect my laptop to the projector are not on the program. Period.  

A single USB port is also ridiculously inadequate. If you are using a mouse, you can't attach a data key. 

It's possible to make an elegant, thin, light notebook that still can work as a business computer. I hate to keep going back to Sony, but the Sony Viao X, in the same price range as the MacBook Air, is thinner (0.55" to 0.76"), lighter (1.6 pounds to 3.0) and has up to 14 hours battery life (albeit with its included extended battery, at a weight closer to the Air's). It's a beautifully slim and minimal computer that still manages to have a standard VGA port, two USB ports, and even a wired Ethernet jack that folds open when in use.

This wouldn't have nearly the star power of launching an Apple Tablet, but if Apple reworked the MacBook Air to have a VGA port in addition to the Mini DisplayPort, and two or three USB ports, its potential audience would increase dramatically. 

Specs of the tweaked MacBook Air: same as the current one.

That's my list. Apple won't do the netbook because they don't think they could make their margins with one. Apple won't tweak the MacBook Air because that wouldn't be purist enough. It will make an Arrandale Mini, but I don't expect to see it until late 2010.

HP LX195 hard drive upgrade

The Hewlett Packard LX195 comes with a 640GB hard drive – a Western Digital Scorpio Blue, to be exact. It's adequate for backing up a few normal office PC's, but it doesn't have the kind of capacity to handle lots of massive media files, and I don't want to find myself running out of space in a year or two.

So I decided to upgrade the hard drive. I found a great deal on a a 1.5TB Seagate Barracuda LP. The Barracuda LP is an odd duck – it spins at 5900 RPM, designed more to save power than for high performance. That works for me, since this drive will mostly do nightly backups while I'm sleeping.

The LX195 case is a really nice piece of design and integration. I wish I could buy a case like this – most small cases for Mini-ITX boards will only accept a 2.5 inch laptop drive, but 3.5 inch drives have better performance and cost less. I followed the excellent instructions at Home Server Hacks to open the case:

The hard drive is mounted with vibration-reducing elastomeric grommets. It took less than five minutes to take out the 640GB Scorpio Blue and put in the Barracuda LP.

In my first attempt at upgrading, I used a PC to clone the 640GB drive onto the 1.5TB drive. This didn't work. Windows Server binds itself to the specific drive it's installed on. Windows Server booted, but it would not use the Seagate drive for backups or open the backups that were on it. This kind of hardware dependency is a bad thing.

The second, less desirable option was to use the server restore disk that came with the LX195 to create a fresh install of WHS on the Seagate drive. This is a fairly tedious procedure:

• Uninstall all WHS software from a PC.
• Connect that PC directly to the LX195 with a network cable.
• Run the server recovery program from the disk on that PC connected to the network the LX195.
• Use a paper clip to press a recessed switch on the LX195 at a certain point in its boot cycle.

This puts the LX195 into recovery mode. It should be able to reset itself to factory configuration using the information from the disk in the PC's DVD drive.

Problem is, that didn't work for me. The PC software would attempt to connect to the LX195 and fail, after about four minutes. It gave an error message so truly useless I didn't even bother to write it down – something like "step 3 failed".

After head-banging several times, time to call Hewlett Packard support. This was the excellent experience I'm becoming used to getting from HP support. No time wasted on brain-dead diagnostics or other nonsense. The tech had a simple solution: give the PC a specific static IP address, and start the recovery process again. This magic number is 169.37.37.42. It worked flawlessly, and a short time later I had a factory-fresh server with adequate storage.

Windows Home Server

Windows Home Server is misnamed. Few people really want a server for their home. It should be Windows SOHO Server, and Microsoft should market it to micro-businesses like mine - businesses with fewer than ten employees and no IT help. WHS has the features of Windows Server that are most important to these businesses: backup and remote access; and it makes those features dead easy to use.

The backup situation at Seal Rock Research has been adequate but never truly satisfactory. For the last eight years, we've backed up to servers running Samba on various flavors of Unix or Linux. Each PC has a simple batch file that incrementally copies changed or new files from our document directories to the server. Too much of this is manual. Everyone has to remember to launch the batch file, and if Outlook is running the Outlook files don't get backed up. Applications and OS are not backed up, and I have had to rebuild machines a few times. We have never lost a document, but I wanted to improve this situation.

I installed the trial version of Windows Small Business Server on an old PC to try it out. It wanted to take control of my gateway router so I went no farther with that. SBS seems like overkill to me anyway; running my own Exchange server is not on my list of must-dos.

The trial version of Windows Home Server was much more successful. In an hour or so I installed it on an old PC, connected one of our client PC's to it and got a complete image backup. 

So When HP started blowing their HP LX 195 out the doors for $199 I was ready. I bought one the first day that price was available at NewEgg. That deal is gone but good deals are still available at Amazon. 

The LX 195 is a tiny box with an Atom 230 CPU, 1GB of memory, a 640GB hard drive, and Windows Home Server. It is headless by design, with no video or PS/2 ports on its back panel. In two hours I had it up and running, and installed the connector software on all our PC's and laptops. I left everything turned on, and the next morning our entire network was backed up. This, really, is magic.

I want to setup remote access but there are wrinkles. I have a dual-NAT network. The gateway router has a 10.1.10 internal network that connects to my LAN/wireless router. The LAN router has an internal 192.168.1 network, which all PC's and laptops connect to. Here's a picture:

 

If the WHS server is on the 10.1.10 network, it can do remote access but not backup; on the 192.168.1 network, it can do backup but not remote access. For now backup is the key application, so I'll leave it on the 192.168.1 network, but this makes me want to flatten my network topology. There are probably lots of ways to do it, but I'm tempted to just burn a public IP so the Linksys can be the only NAT router on my network:

 

Another wrinkle is storage. We aren't a very data-intensive business, but the initial backup used up over half of the ~600GB the LX195 makes available. The LX195 has only one internal drive bay. Expansion is by adding external drives to the four USB ports on the back. This is OK, but I'd rather not have two pieces of gear where one would do. I don't want to find myself in a space crunch a year or two from now, so I've ordered a 1.5TB drive as a replacement.

The Nexus One sucks for business

Julie got her Nexus One yesterday. It may well go back to Google next week.

As everyone else is saying, it's a beautiful phone with a superb screen and a snappy processor. Android supports Exchange server, and it only took a few minutes to get her email and contacts synced with our hosted Exchange provider. The problem is, the Nexus One won't sync with the Exchange calendar.

Here's how Exchange syncing works with the BlackBerry, the iPhone, and even hapless Windows Mobile:

Simple and sweet. Everything is stored in the cloud and all devices are always in sync.

The Nexus One works that way with email and contacts, but the Nexus One calendar app will only sync with Google's calendar. We can use Google calendar sync to sync events with Outlook on the PC. That produces this kludge:

Calendar sync uses a polling model, and won't update more often than every ten minutes. Julie will need to be careful not to turn the PC off after making changes in the calendar. The notebook won't participate unless I install calendar sync on it too. Changes made using Outlook Web Access won't sync until a PC or notebook with Google Calendar Sync does the job. This is the kind of mess decent modern software and devices haven't required for years.

There's an app called Touchdown that apparently does the job, but you have to use its interface to access your calendar. It also costs $20.

I can't believe Google shipped this phone with such a basic component missing. The Nexus One is not a business phone. Maybe we need to wait for Nexus Six.

Chrome OS shows Google is the new AOL

Google's Chrome OS is more evidence that Google aims to become the new walled garden of the Internet, a la America Online. For a long time, AOL was a destination where subscribers to its connectivity services became targets for ads, enriching the company twice - once when they paid for their dial-up service, and again when they responded to pitches on the AOL home screen.

Chrome OS follows what is now Google's standard three-fold strategy: offer something that is (1) really good, (2) free, and makes it (3) dead easy to access Google's services. The Chrome browser, Android OS, and now Chrome OS are examples.

No one is forced to use Google's products or, using them, to use Google's applications and services. Like Microsoft, which triumphed for a while by placing its browser icon on the desktop, Google plans to win by being the most convenient option. People like being coddled, even if it means they're being herded into Google's pastures.

Like Microsoft, too, Google will no doubt be surprised when the antitrust vultures swoop in to punish it for for treating customers so well they don't want to patronize anyone else. So it goes.

Why Macs are like Birkins and BMWs

What does a Macintosh computer have in common with a BMW car or a Hermes Birkin handbag? 

If you said exceptional trend-setting design and a high price to go with it, you're half right. 

There is one other thing: like BMWs and Hermes handbags, Macs have superior resale value.

I bought a Mac Mini when they first came out in early 2004. It was the $499 base model – 1.25 GHz G4 processor, 256MB RAM, and a 40GB hard drive. I bought it for one reason: to learn the OS X interface so I could do market research on Mac software. For that purpose it worked well enough, but it was much too sluggish to use as a general purpose computer.

Sometime later I paid $20 for 1GB of RAM and installed it. The sluggish Mini became a smooth, responsive PC that could easily handle day-to-day computing tasks – even photo editing. Its tiny form factor and near-silent operation made it ideal for my home office.

I was quite happy with the Mac Mini and still am, but in the summer of 2009 I started using Microsoft Windows 7. I run it on a home-made desktop with a 2.4 GHz Athlon II. Windows 7 really is much better than Vista or even XP – it's about as good as Mac OS X. I ended up using the PC more and the Mac less, partly because the PC lets me use Outlook 2007 with my hosted Exchange account. Outlook is beastly software, and I'd love to try Snow Leopard's Exchange compatibility, but Snow Leopard won't run on G4 Macs.

So the Mini was surplus. Checking eBay, I was amazed to learn that Minis like mine went for up to two hundred dollars. Not wanting to go through the anxieties and unknowns of an eBay sale, I listed my Mini on Craig's List for $200 and sold it in a few days for $175. I threw in a keyboard and mouse, but they had no value to me and I don't think they influenced the price the buyer paid. So including the cost of the memory, but ignoring the keyboard and mouse, the actual cost of my Mini was $344 – darn cheap for a desktop computer in 2004, and even today. The selling experience was quick and congenial.

Why do Macs have resale value? Part of it is that there are only a few models. Buyers know exactly what they are getting. This, more than anything else, reduces the friction associated with a transaction.

Part of it is industrial design. No one ever lost style points for using a Mac.

And part of it is something else. Macintosh computers appeal to interesting people. At least, the two people who wanted to buy my Mini were more than commonly interesting. One is an educator who uses Macs in community-oriented educational events and programs. He was too far away, and didn't want to make the trip to San Francisco. No problem. The actual buyer was a German programmer for Pixar who plays with Amiga software in his spare time. It turns out there's an Amiga port for the G4 Mini, and he wanted to run Amiga OS on a Mini (and now he does: he emailed me later to let me know it had worked).

Not only did I get a third of the cost of my Mac Mini back; I had two encounters with interesting people I would not have met otherwise. That's a pretty compelling case for buying another Mac.