I tried to run a few of my scripts on a linux machine, and kept getting weird errors like

“: No such file or directory”

I finally figured out that the files were in DOS formatted line separation (\n\r) which was confusing the shell. A quick

dos2unix *.py

fixed things.

I had a very annoying bug in my wxPython application that I couldn’t figure out. In a GUI application that has a status bar and a context menu, any call to self.PopupMenu(mnu) would result in the following:

wx._core.PyAssertionError: C++ assertion “(number>=0) && (number <m_nFields)” failed at ../src/generic/statusbr.cpp(142) in SetStatusText(): invalid status bar field index

I use Boa Constructor to build my wxPython app, and in one of the generated code was the following:

 def _init_ctrls(self, prnt):
        # generated method, don't edit
        wx.Frame.__init__(self, id=wxID_FRAME1, name='', parent=prnt,
              pos=wx.Point(503, 180), size=wx.Size(988, 660),
              style=wx.DEFAULT_FRAME_STYLE, title=u'')
        self._init_utils()
        ...
        self.SetStatusBarPane(3)

setting SetStatusBarPane(-1) disables automatic help text printing to the status bar, but also stops this very annoying bug.

I started programming in python from the days of 1.5.2, and have not kept up to date with some of the new features. Two features that caught me by surprise:

enumerate()  # builtin function

"...".format() # String method

enumerate() takes a list an returns an interator tuple (index,value) as it goes along the sequence. Very handy and obviates the need for the very messy:

for i in range(len(L)):

    print "I need the index {0} and the item {1} at the same time".format(i,L[i])

This illustrates the second new feature I just realized, the .format() method. This apparently replaces the ‘%’ operator and allows formatted strings to include positional replacement strings using {idx} notation.

The title is quite a mouthful, but I wanted to record what it takes to get debugging working in the Eclipse IDE for the ARM Cortex M3, specifically the Luminary Micro LM3S811 evaluation board.

Unfortunately, the state of Cortex M3/LM3S811 support by OpenOCD is not yet at the point where it is smooth. I think the underlying problem is that resetting the LM3S811 is not happening as expected.

1. Go to the YAGARTO website and download/install the components. Note that the website itself states that YAGARTO does not support cortex m3 yet, we are using this site to get most of the other components, including Eclipse and the Zylin CDT extension.
2. Go to the CodeSourcery website and download the Lite (free) version of their GCC compiler (EABI platform). It is a set of command line tools, but we will be integrating them into Eclipse.
3. Go to this site and get the “software-package Version 20080409”. This is the version of OpenOCD that actually works for the LM3S811. The OpenOCD version on the YAGARTO site, although newer, did not work in the following steps. This site is actually very informative and remains the best site for Cortex M3/GNU ARM toolchain info. Unfortunately, it does not talk about Eclipse integration.
4. Configuring Eclipse is complicated by the fact that several sources of information (YAGARTO, etc.) all say slightly different things. I will eventually list out the exact steps, but for the most part follow the HOWTO’s on the YAGARTO site.
5. […more to follow on Eclipse configuration…]
6. I am assuming at this step you can successfully program the main.bin file. The debug step is a bit more complicated, as it is a two part process
   • Run OpenOCD from the External Tools menu, using the .cfg script that is specifically for debugging (as opposed to programming.
   • Run GDB from the Debug menu. Note this was the most confusing part.
     1. Create a configuration under the “Zylin Embedded debug (Native)” option.
        • Main Tab: The C/C++ Application is the main.bin file that was created (from an .elf file) from the build process using a Makefile.
        • Debugger Tab: Browse for the CodeSourcery GDB file. The default location is C:\Program Files\CodeSourcery\Sourcery G++ Lite\bin\arm-none-eabi-gdb.exe
        • Commands Tab: Here is the key part. Use the following commands, taken from a variety of sources. Again, since resets don’t work properly for the LM3S811, most lines other than the first and last are useless. Note the symbol-file command should be to the .elf file, not the .bin or .sym file.
          • Initialize: target remote localhost:3333
            monitor reset
            monitor sleep 500
            monitor soft_reset_halt
            symbol-file “C:\\PATH\\main.elf”
          • Run: hbreak main.c:38
        • What is going on in the hbreak command is that a hardware breakpoint is being set at a particular line number in main.c. Hardware because software breakpoints don’t seem to work on embedded targets (I could be wrong here). This ugly manual setting is because the reset/halt/init mechanism doesnt work, so without manually setting a break point “Debugging” programs actually lets the program run without stopping.
        • Note this means you need a separate Debug configuration for every subproject/program. This is definitely a pain, but as far as I know the only way you can start debugging at an arbitrary point in the source code. Note that single stepping works well, if slowly.

BTW, the errors involved in resetting and halting the LM3S811 are shown below


Open On-Chip Debugger 1.0 (2008-04-09-14:00) svn:
$URL: http://svn.berlios.de/svnroot/repos/openocd/trunk/src/openocd.c $
Error: jtag.c:927 jtag_add_reset(): BUG: transition of req_tlr_or_trst and req_srst in the same jtag_add_reset() call is undefined
Warning: jtag.c:1197 jtag_check_value(): value captured during scan didn't pass the requested check: captured: 0x0f check_value: 0x01 check_mask: 0x0f
Warning: jtag.c:1157 jtag_read_buffer(): in_handler reported a failed check
Error: jtag.c:927 jtag_add_reset(): BUG: transition of req_tlr_or_trst and req_srst in the same jtag_add_reset() call is undefined
Warning: jtag.c:1197 jtag_check_value(): value captured during scan didn't pass the requested check: captured: 0x0f check_value: 0x01 check_mask: 0x0f
Warning: jtag.c:1157 jtag_read_buffer(): in_handler reported a failed check
User: target.c:382 target_process_reset(): Timed out waiting for halt after reset

I have finally gotten a reverse proxy app to work on the Windows Server 2003 machine.

The solution:

1. Install cygwin. This by far was the biggest stumbling block, as I didn’t want to install cygin on a production server. However, after scouring the web (and google) for free/open source reverse proxy apps for Windows, I came to the conclusion that the only ones available require cygwin.
2. Get Pound. I’ve looked at a variety of reverse proxys (nginx, lighttpd, apache, etc.) and each one seemed to have some disadvantage that I couldn’t ignore. Not to mention many of the alternatives weren’t dedicated reverse proxys but web servers. More details are here.
   • One caution is that the most recent version of pound (2.4.2) would not compile, as it required some IPv6 support in the headers. I chose to download the previous version (2.3.2) and it did compile smoothly.
   • ./configure –without-ssl –disable-log –disable-dynscale
   • Strangely enough, even with the –without-ssl option the openssl library is linked into the pound.exe executable.
   • make
   • Copy the pound.exe and the associated DLLs ( cygssl-0.9.8.dll, cygcrypto-0.9.8.dll, cygrunsrv.exe, cygwin1.dll, cygpcre-0.dll, cygpcreposix-0.dll) to a new directory (e.g. c:\apps\pound)
3. Create a pound.cfg file in the c:\apps\pound directory. Note that the sample config files shown in this site do not work. I am assuming they use commands for an older version of pound that have since been removed
   • A sample config file that works for me is:
   • 
     # Set pound to run like a normal app, so that cygrunsrv can daemonize it
     Daemon 0
     ListenHTTP
     Address 1.2.3.4
     Port 80# Prevent any access to the ZMI from outside
     Service
     URL ".*/manage"
     End# If you have multiple domains, repeat the Service block
     Service
     HeadRequire "Host: .*A.edu.*"
     BackEnd
     Address 127.0.0.1
     Port 8123
     End
     End
     End

   • Note that for this to make sense, the web server (Zope in this case) should be configured to listen only on the localhost ip address (127.0.0.1), and the firewall to block port 8123, otherwise the intent of having only pound exposed to the internet will not be realized.
   • By the way, adding multiple BackEnd directives allows for load balancing, even with prioritization. Really simple and cool!
4. Once you have pound running and have tested it on the command line (pound -f pound.cfg), you want to install it as a service so that it will be automatically started on startup.
   • cygrunsrv –install Pound –path C:\Apps\Pound\pound.exe –args “-f
     C:\Apps\Pound\pound.cfg” –stdout C:\Apps\Pound\pound.log –stderr
     C:\Apps\Pound\pound.log
5. I haven’t configured it yet, but I will be adding additional URL filters to reverse proxy to the IIS server.

I think I have finally settled down on the embedded system that will be used for the robotics course. At the sensor processing/integration level, the PSOC CY8C29466, either with the evaluation board that Cypress sells, or my own custom board. At the machine learning/interface to MS Robotics Studio level, the Luminary Micro LM3S811 ARM Cortex M-3 Microcontroller, with their evaluation board.

I am still debating whether to move to C compiler for the PSOC (up to now I have been teaching the M8C assembly language). With the addition of the LM3S811, with its own set of (free) C compiler tools, I am tempted to standardize on C for both microcontrollers. On the other hand, teaching assembly language is worthwhile I believe, for future roboticists.

 My first step is to get the two microcontrollers talking to teach other, most likely using I2C. Will post more once that is accomplished.

This year I am revamping both Web development and robotics courses to use Microsoft technologies: ASP.Net 3.5, IronPython/Dynamic Language Runtime, Robotics Studio.

 As a first step I am trying to port over the course website, or at least some sample code, to IIS. Apache has its headaches with long config files, but the IIS setup is even more confusing, since not everything is accessible in one file.

1.  The first stumbling block was I couldnt get a simple Default.aspx file recognized by IIS to serve when I went to http://localhost.
   • Solution: register ASP.Net 2.0 (3.5) with IIS
   • In the c:\Windows\Microsoft.NET\framework\v2.0.50727 directory run the command aspnet_regiis -i

Note that ASP.Net 3.5 is really ASP.Net 2.0 with some extensions.

2. Second step was to add SSL (Secure Socket Layer) support, ie allow https:// access. This was surprisingly easy from the IIS perspective! The only hard part was to generate the Certificate Authority and Server certificates.
   • solution: Look at this page
   • Be sure to open up the firewall with port 443!
   • Install the server certificate in the IIS manager
   • To remove the annoying warnings on the client browser, install the certificate authority certificate on the client machine.
3. NFS mounting. The students will be given accounts on the unix (really linux) systems with their own directories. We want to allow students to log onto the Windows 2003 server (cygwin sshd) using their unix username/password (ActiveDirectory/LDAP, to be discussed below) and mount their NFS directories. This will aid in “uploading” web site code to the server.
   • This site has very good information on setting up Windows Services for Unix (SFU) for Windows Server 2003

This year I am revamping both web development and robotics courses I teach during the summer.

• Web Development
  • Up until now I have been using Plone/Zope/Python for both the course website and teaching web development. Lately though I have been examining Visual Web Developer, ASP.Net, and C#. While I think Plone/Zope has its advantages, clearly ASP.Net is more popular, and its feature set (especially with 3.5) is getting to be pretty comprehensive. What has finally convinced me to migrate is the fact that ASP.Net pages can now be written in IronPython, the .Net implementation of CPython.
  • IIS seems to have its issues, however, so I am looking into reverse proxies that will hide IIS and Plone/Zope behind it.
• Robotics
  • I have been using Parallax BASIC Stamps, Sumobots, and Cypress Semiconductor PSOCs. While the Sumobot is a very nice hobby robot, the computational power of the BASIC Stamp that is at its core leaves a lot to be desired.
  • Microsoft has the Robotics Studio, a .Net platform for robotics work. It has gained sufficient support that I think it is a viable platform on which to teach my course. It requires an interface on the robot that is too much for the Sumobot (although there is sample code for the BOEBot, which is similar). I have been looking at the ARM series of microcontrollers for some time now, and have decided to have a hybrid solution, where a PSOC handles all the low-level sensor/motor interfaces, and the new ARM Cortex M3 handles the AI/machine learning as well as the interface to the Robotics Studio.
  • I purchased the Olimex STM32-P103 prototyping board, based on the ST STM32F103RBT6 Cortex 32-bit microcontroller. It has the following features, from the Olimex website.
  • ARM 32 bit CORTEX M3â„¢ with 128K Bytes Program Flash, 20K Bytes RAM, USB, CAN, x2 I2C, x2 ADC 12 bit, x3 UART, x2 SPI, x3 TIMERS, up to 72Mhz operation.
  • I think I have gotten spoiled by the rich set of peripherals available on the PSOC, because while the set of peripherals the Cortex has is reasonable, by itself it will not be able to manage all the sensors and hardware of a robot.
  • I am thinking of using a new chassis for the robots, I need to do a full engineering preliminary design before I can commit to that course of action.

After much reading and searching I have finally managed to burn, run and debug a sample LED blinking program on the STM32-P103. The beauty of open source development tools is obviously the cost and the frequent updates, but the disadvantage is the lack of good documentation or manual. I have been able to piece together bits and pieces through a lot of trial and error, and will be posting everything I have learned so far soon, with sample code.

The UN has made data it collects from its member nations available here. Very interesting, and I am looking forward to mining some of the data to prepare for the web course this summer, as well as try my hand at some economic analyses.

It is no secret that IE has some bugs in rendering CSS styles, most notably with respect to the box model. This wikipedia entry has a good summary of one of the bugs, which is really a misinterpretation of the CSS standard. It doesn’t help that for some web developers, Microsoft’s interpretation makes more sense than the “correct” standard!