Testing your code, pretend you are an end user

What is one of the best ways to test your code? Pretend you are an end user. Run it like you bought it instead of like you are testing it.

Near the end of this release I add export to CSV for the scheduler. People want all sorts of reports so why not let them export the data and do whatever they fancy in Excel? Works like a charm. I was doing some other testing and though "Hey, I will just export the data so I can sort it and see it in Excel to fix this bug!". Whammo, I hit a bug right away in the Export dialog.

I just wanted to export the appointment ID, start time and duration so I deleted all the other fields in the list box but then I could not add any back. The list box was reporting a selected index of 4 (the original count of items in the list) even though there were no items in the list. I changed the code to ask the backing data model for the row count so I could insert at the proper place and one bug fixed.

Next I wanted to sort the exported data. The problem is recurring appointments exported with the start date of the recurrence not the data of the grid which means they sorted in front of non-recurring appointments. When I export data I want the date to be the same for all appointments and just for the time to change. I fixed that issue too in the fixes branch and trunk.

Here I am using the software like an end user and finding bugs. Most developers are afraid to test like an end user. We are scared of finding bugs. We tend to test in a little vacuum, testing small bits of code in very isolated conditions. No user runs your program like that. They dive in and hit every menu item and button trying to make it do what they want. They multi-select, drag and drop, double click, press disabled controls and have all sorts of fun. You need to do that too. Beat on your program like it owes you money.

For the days leading up the release I just ran the software and found a number of usability issues. Why can I set default cell width for open / closed business hours but not overall row height? Why is the pivot set per view and not just once for grid? Either you like Time as a row or a column and don't want that to change if you change the other axis to Category, Provider or Facility. Yes it annoyed the technical writer that I changed this late in the game but the code and writing took us an hour and this is something that would annoy the user every day. It annoyed me after playing with the program for an hour.

You can't depend on your QA staff to be an end user. A lot of the time they are running through regression testing scripts following a very narrow path through your code. They don't mind finding bugs so they will beat on things in odd ways but it is still a view with blinders on. Those test tend to skip over usability and look at functionality. Yes, you can follow these 10 steps to get from Y to Z but why are there 10 steps? Can we do it in 5? Export was two layers deep in a menu but is something that many will use everyday to print out a reminder call list. I put it as a button on the main toolbar. Once I did that I made a note to make the toolbar layout configurable so those who don't use that button don't need to see it. I plan on implementing that in the next release.

It is so easy to just keep adding features figuring that will make clients happy but remember they are still using the existing features on a daily basis and those should be easy to use with as few steps as possible. I save off the settings they used the last time the exported data. Who wants to reconfigure that everyday? Maybe you do three types of exports. I allow you save as many named export as you want. Where do you write the file? I default to the user home directory. What if you give that export definition to another person? They don't have a directory called c:\users\john.doe so I write out {home.dir} instead of c:\users\john.doe and do replacement text when you load the export definition. Don't annoy your users, computers are there to eliminate steps and avoid the mundane.

Allow multiple ways to do things. Some people are mouse / menu folks and others use hot keys. If you use hot keys use them all over otherwise heavy keyboard users will be annoyed. Some like toolbars and will use tooltips to see what they do if the icon is not obvious. Others are text oriented and want to find what they want in the menu. Menus need to be logically laid out with descriptive text. I hate a menu that has "Preferences", "Options" and "Tools" - those words mean the same thing to me and I always pick the wrong one when I want to change a setting. Microsoft I am looking at you here with nearly every Office application. Use you tech writing team or someone you consider to be a wordsmith to help here.

Take the time to nurture your application. Don't just keep throwing more stuff at it watching it bloat up. Consolidate smaller dialogs when possible. No one likes to bring up 10 dialogs to get their initial preferences in place or to have to remember where to find some obscure setting.

You want them to be able to quick find and use every feature you have. Why waste time coding it if no ones knows it is there?

Sync one minute Java timer to the current time

I am working on a patient scheduler and I need to draw a line across the screen indicating the current time. Helps them not create appointments in the past. We also have a clock on screen via the Jide status bar. It was driving me crazy that the line moved down or across screen but not in sync with clock. That means the line might move at the 20 second mark of the clock all depending on when you started up the program. Turns out it was adding one line of code to get them within one second of being in sync which made me happy.

timer = new Timer(60000, this);
// Force initial delay to sync within a second of current clock
Calendar cal = Calendar.getInstance();
timer.setInitialDelay((60 - cal.get(Calendar.SECOND)) * 1000);
timer.setRepeats(true);
timer.start();

It is the one line of code setting the initial delay that solves the problem. After that the timer triggers every 60 seconds and away we go. Just makes things look so much cleaner.

Programming then and now

Is programming getting harder? I am not sure if the actual art of programming is getting harder but user expectations are growing so quickly it is much harder for a single programmer to keep up with all areas of expertise.

Let’s set the way back machine to the land of terminals. Your user got a color choice of white, green or orange and maybe they could toggle the foreground and background colors. Basic operation was CRUD – Create, Review, Update and Delete. You picked from a menu and acted on one item at a time. Screens appeared and you typed in data probably in all uppercase. Everything was done in 80x25 all text. If you needed more room you added another entry screen.

Next up we tapped into the PC as a smart terminal. That gave us more colors. Everybody stuck with 80x25 for a bit but some ventured out to 80x50. You could color menu items; required items could have a label in red etc. Still looked like a terminal in general.

The GUI came into play. Everyone tried to make each dialog box look like a terminal window at first. Then things started to change. You could have a list box of data and scroll through it. Not everything had to fit the physical screen. Not only did you have colors but you had images. The mouse drove things but most still used it just to click on buttons or to quickly get to a certain control on the form. Next tooltips, right button mouse menus and lots of custom controls started to appear. People also went nuts with fonts doing bold, italic, 20pt, etc. to make things nearly unreadable. Various mouse clicks came into play. Double click to add from this list box to that one. Drag and drop within an app became required and users started to demand drag and drop between apps. The clipboard was a cool place to put things to save on typing.

Once people got used to the GUI they wanted it to do everything for them and if one program did it they expected all programs to do it. I want spell checking anyplace I can type words. Drag and drop should always work. I can copy anything I want to the clipboard. If you want a patient name here I better be able to create it right here by bringing up the patient create dialog. When I search on something I should not just have to scroll through a huge list but I want to start typing and have it narrow down my selection. What you only narrow by things that start with what I type? I need it to find the text anywhere within the data! I want to export all my data, wait expect for that column. Why don’t you remember the last thing I did every time I come into this dialog box? I want to import an image and have it automatically do OCR and pull out the text. Hey, all I did was reorder fields on this form and now the OCR is doing it all wrong! How come I can’t see my contacts here? What about my web browser favorites?

Our current platform is the smart phone and tablet. Now they expect to hardly type anything. It is a data consumption device. But they want to display the data in a very pretty manner with all sorts of swiping gestures that should be natural. What happens on single tap, double tap, swipe left or swipe up? The screen is back to tiny, really back to the 80x25 days in a lot of cases. But the screen can be bigger. A tablet is a PC screen again so just porting the smart phone version annoys that group of folks. Touch screen devices have cameras and people want to use them. Just when we were elated with tons of RAM to store everything so they can search on it willy nilly we have been told you barely have any memory to play with again.

When you write a program you might have to do four versions of it.

#1 A full blow PC app with every bell and whistle allowing you to manage all the data with ease. Might need to run on a Mac too.

#2 Slightly trimmed down Web version available under every browser under every OS for data access on the go.

#3 Smart phone version that has access to all the data, is super pretty, supports the camera, speech to text and all sorts of touch processing

#4 A tablet version that does everything the smartphone does also supports shrinking / expanding sub-screens so you can not only see where you are but where you were. Still fit all of this into as tight as memory as you can.

As you can see a single programmer has to know and do a lot more than ever. Where once everything was text now there are text and graphics. May the user could hit a predefined function key but now you have a master menu item, right mouse button menu item, accelerator keystroke, mouse swipe and finger swipe to access each event. You must support data in as many formats as possible for Drag and Drop, export and “hey someone deal with this” such as tapping on a web or email link. You need to interact with other programs and their data that may or may not be on the device.

It seems similar to the games industry. At one time a single developer or maybe two could write a game it now takes a team. Now you have sound, graphics, core game engine and menu system folks. We are seeing a similar need for UI specialist, data export folks, core OS interaction and many others.

On my current projects I have worn every hat in the list. I wrote both the iOS and Android versions. I did all the graphics and designed all the screens and touch interactions. I wrote a custom control for a menu system that appears when you tap on certain items and did all the DBA work creating 25 tables under SQLLite on each device. I am at a very small company so this is pretty much expected. At my previous job I had a User Experience team to handle the screen layouts and all the graphics. I had a design team to decide on how things in general should work. If I had a question about how something should look or operate I pretty much knew who could answer that question.

For the main Java application we wanted to add export functionality. By the time I was done with that it was over 1,000 lines of code that I wrote in a just two days. Why so much code? The export part was easy, just take the data and write to disk in CVS format. But that is not what a user wants. They need to pick every field they want exported, the name of the file, should I append the date of the data to the file name, where should the required appointment ID go, how to write out multiple lines for multiple associated physicians. What if they export different data for different needs? I allow you to name your format and save as many named formats as you like. I need to load / save this to the server. Then we needed permissions for creating named filters but if you don’t have that permission you need to be able to do ad hoc exports and save that locally and not on the server. It is all up and running but there was a ton to do and think about. We want to release on Monday so I had to get it all coded up and ready for QA to give it a solid once over. This is what you do working for small company on a monthly basis.

Hammering away at Jide upgrades

Still hammering away at upgrading to the latest version of Jide. So many areas of the code to touch as we also want the user to be able to set their font size used by all the dialogs / panels. I have found so many areas of the code doing "new object" that don't need to especially in the table cell rendering area. I am replacing that with a single object that is just returned as needed. Silly to have that much object thrash.

Some of the other areas I am making more generic. Too many specialized derivations of the table when one generic implementation with column definitions would fit the bill.

I was not around during the initial code creation. Much of it was done under huge time constraints. As we open files to edit them it is a good time to clean them up. We do delete a lot of code which could throw a lot of code metrics systems off course but we enjoy seeing the functionality stay the same and the code size shrink. Less code means less potential bugs.

Java Graphics.translate is very slow

To get around an issue I was having with a JIDE CellSpanTable in a JideScrollPane hosted by a JXLayer I was using graphics.translate(x, y) but it was killing the speed of my code. I don't know what the issue happens to be with JXLayer when you call setViewPosition a Viewport of a table. It gets out of sync with JXLayer but only in the Y axis. X seems to be fine no matter what. I need to set the view port position to open of business when the grid first starts up and then at other times as I reload data from the server.

To get around the bug I used a hack of sending the new Y offset into my renderer so I could call translate and all would paint fine. It painted just as expected but the speed was horrendous. I am using the JXLayer so I can drag an appointment around on the scheduler. I fixed the problem by adjusting just the two rectangles I use for painting instead of doing a translate on the entire graphics canvas. Paints just as fast as it did before now.

It has always been slow on the Mac with or without the translate. In fact the translate call seemed to make no difference on the Mac as far as speed goes. On my i7 with 6g of RAM running Win7 64bit the speed was unusable with the translate in place.

Since this is all happening in drag and drop code trying to put a profiler on it was going to be near impossible. I knew it was running fast before so I commented out things I had recently added until I got my speed back. I knew I had made the rendering code more flexible so I thought maybe a loop in there was causing the issues. Luckily for me I had some old test code with the old rendering code and it was running slow too so I figured it had to be the recently add translate call.

It still feels like my solution is a hack and I would love to get rid of it. I tried all sorts of SwingUtilities.invokeLater variations along with setting view port position, getting horizontal and vertical scrollbar positions, resetting view port and then setting the scrollbar positions but it still just screwed up. I don' think this is a JIDE issue. I think it would screw up with a standard JTable and JScrollPane. Somehow JXLayer is just getting out of kilter.

As far as the scheduler goes it is working nicely. I have been making things faster, easier to use and cleaning up dialog box clutter. I have probably removed more lines than I have added and the whole thing is a lot easier to understand with data in objects instead of in maps with list of other maps. There are still new features to be added and it has not been seen over in QA yet.

There are times you just have to punt

I have run a very simple file, print, backup and SVN server at home for a number of years. It was built using old parts laying about the house. A few weeks back the fan on the video card and the chipset fan on the motherboard died. I had another old AGP video card so I slapped that in and it kept running until this past Thursday.

My wife needed data off the HD so I had here take the external USB drive I use for Monday / Wednesday / Friday backup cycle and had her plug it directly into her machine. Made me a bit nervous but she needed her files. I just asked her not to delete anything.

Thursday night I pieced together a new machine to be the server. I plugged in the original HD and went to work. This was my son's box and it had been acting flaky so I had some doubts. I got WinXP running again but when I tried to the the motherboard USB 2.0 drivers to work things went to hell quickly. Devices were not recognized and unplugging or plugging one in could cause a BSOD and reboot.

I spent way too many hours on this before I gave up. I hit the web to see what the cheapest Win7 based machine I could get at Microcenter happened to be. I was sick of WinXP on the server and all other machines in the house are running Win7. They had an open box eMachine for $215. I know, it is an eMachine but I have bought them in the past for family members on budgets and if you just run what it has as the base configuration they keep running. I have not had one fail yet. My big worry is their small power supplies.

This is a typical machine with AMD 2.8g processor, 2g of RAM, 500 mb HD, nVidia chipset and video card in a slim line case with plenty of USB 2.0 ports. Fired right up and I spent the next bunch of hours uninstalling crap, installing patches and restoring a previous backup from the external USB drive. The proper area of the drive is shared and the printer is shared so the rest of the family is happy. I have installed Apache and SVN but have not fully configured them yet. I was sick of computer stuff so I played some games and did work on a side Android project.

In the end of should have given up much earlier and just bought a new machine. Trying to get old suspect hardware running just to be cheap is not a good solution. We have proven over the years that the server is critical to running our house. I have plans to get more services running on the box - SVN, some flavor of SQL, etc. so it needs to be slightly on the mission critical side of things.

Win 7 Professional always freezes at 15% during patch installs

Various machines around the office - all Win7 64bit Professional - freeze during patch install shut down at 15% complete. We hit the reset button, tell it to boot normally when prompted and the patches install just fine. This has happened every time over the past year that we have installed patches. I don't have this issue at home where I am running Win7 64bit Ultimate and Home. Annoying but does not seem to be a show stopper.

I have completed my work on the JIDE update. There are some outstanding issues but others are fixing them while I get back hot and heavy on the MigCalendar scheduler replacement. I have some open Mac issues with Jidesoft. They don't appear to do much testing on the Mac which is a shame. You can basically use the main OS look and feel on the Mac but the other ones are useless. Colors that are not readable, tabs that are not clipped so they paint on other tabs etc.

I was doing various bits of editing of the code under IntelliJ 11 on the Mac and PC and ended up getting SVN all confused. I finally just extracted a fresh copy of the code and renamed the directory to get the Mac to play nice again. I was doing weird stuff like double editing files on my home PC and the Mac laptop then trying to revert on one and check in on the other. Classes were moved to new packages etc. which I am sure caused a great deal of grief. All fixed now, just some of the fun of being a multi-platform developer.

The MigCalendar conversion is going nicely. The new JIDE based grid is so much faster at everything you can hardly believe it is functional, it feels like something should be missing otherwise how can it be this fast? MigCalendar has a lot of wonky ways of doing things and used a lot of static collections to do it. Glad to be rid of it. I did a build yesterday with no MigCalendar references in it at all. I deleted a ton of code and old dialog boxes in the process. Plus the new scheduler is so much more flexible and configurable. The clients are going to be elated when we ship.

I plan on writing an email to the MigCalendar folks letting them know why we are discontinuing the use of their product. I figure developers rarely are notified as to why they have lost a sale so it seems only fair. We just found it to be to bulky, not flexible and the documentation to be poor to non-existent. I was never able to find any useful references on the web either an they only allow you to see their forums if you are paid up on your license. Not a way to run a successful business in our book.