What Have We Lost

In the last 40+ years that I have been developing software there has been a massive increase in computing capability, both in terms of performance and the speed and ease of which new software can be created.

If we look back , the level of benefit to the user hasn’t kept pace with either of the above metrics. 40+ years ago I would have used a word processor to write this abstract (albeit without some of the fancy formatting) on a computer that had a single 8 bit processor and ran at Mhz speeds (ie: Commodore 64). On my modern computer with 16x 64 bit processors each running at Ghz speed writing this abstract is much the same experience.

Let’s take a moment to look back at how things were done in the past, particularly at techniques which I do not find in common practice, to see how we can do better. This is not to say that 40+ years of progress should be thrown out the window, this is to say some 40+ year old techniques still have value today and should not be forgotten.

Transcript

1. Lloyd Moore, President Lloyd@CyberData-Robotics.com www.CyberData-Robotics.com What Have We Lost
2. Agenda: As technology and software development techniques have moved forward we have left behind some simpler techniques that are still useful. Let’s review some of them….
3. Link Time Overlays Description:  A linker option allowing individual functions to be replaced / mocked in an executable. The technique is somewhat related to function overlays of the distant past when memory was VERY scarce. Modern Day Approach:  Dynamic Linked Libraries (to some degree) How To:  Set the linker command line option allowing multiple definitions --allow-multiple-definition for GCC  Identify individual functions to be replaced  Create duplicate function definitions in different compilation units/object files/libraries  Based on the build type include or don’t include the object the object with the alternate definition
4. Link Time Overlays Advantages:  Allows for a known common code based to be modified for testing, while GUARANTEEING the integrity of the rest of the program.  Particularly useful when software needs to pass a formal qualification and the exact code used in production must be modified.  Linkage is static and minimal modifications are needed to the program.  Program configuration is controlled and guaranteed by the build system. Disadvantages:  Technique is obscure and not obvious to many folks these days.  Duplicate function names are not expected and will mislead folks.  Generally frowned upon if used where not absolutely required.
5. CSV Files, Simple Formats Description:  Using only the simplest file format needed to represent the information. Modern Day Approach:  Markup languages such as JSON and XML How To:  For a CSV file write out an optional header line with the field titles comma separated: Field1,Field2,Field3  Write out each set of values, comma separated as a single line: 1,2,3  If size and performance are really critical can also use a binary format such as: STX 0x01 0x02 0x03 ETX
6. CSV Files, Simple Formats Advantages:  For simple, regular data the format is easy to read and write in code. (fprintf, fscanf, strtok or similar)  Removes the need to pull in a serialization/de-serialization library.  Files are typically MUCH smaller in size as redundant field names and delimiters are removed.  Can easily interop with a spreadsheet.  Binary files can be even smaller and faster. Disadvantages:  Not as human readable.  Generally not usable where the data structure varies significantly, and if it does much of the simplicity is lost.  Depending on the application may not be as resistant to error detection.  Needs custom code to parse into internal data structures (JSON).
7. Simple Serial Protocols Description:  Basically the same ideas we just discussed but instead of storing to disk the information is sent over a serial link (cabled or wireless). Modern Day Approach:  More complex formats such as JSON or XML embedded inside a transmission protocol such as TCP/IP. How To:  Most beneficial here will be the binary protocols in cases where efficiency is critically important.  Generally will want to have a packet format with the following properties:  A clear start of packet/message indicator for framing  A clear end of packet/message indicator for framing  Some type of error detection and/or correction (1+2 = 3) STX 0x01 0x02 0x03 ETX
8. Simple Serial Protocols Advantages:  Can be optimized and tuned for the specific application:  Send x,y coords from [0,99]  Errors handled by repeating packet  Can do this with 2 bytes by flagging one coordinate having high bit set  A typical JSON packet would be about 20 bytes (depending on exact white space): { "x": 10, "y": 20 }  Allows for the use of simpler lower speed physical layer, sending more data, and/or faster update rate. Disadvantages:  Not easily human readable.  Not parsed by protocol analyzers.  Have to write custom code on both sides to match to data contents.  Complex or variable data becomes much harder to work with.
9. Simple Serial Ports Description:  RS-232/422/485 are simple 2 to 9 wire serial ports that have been around for many decades. They used to be standard everywhere but now are only seen in special use cases. Modern Day Approach:  Ethernet most common for long distances, USB most common for short distances. How To:  PCI cards are still available that support RS-232/422/485  Can also use USB to RS-232/422/485  FTDI chipset is the most common and widely supported on all major operating systems  Ethernet to RS-232/422/485 adapters are available  In code simply open the serial port as a file and start reading and writing to it.  Will need to configure the communication parameters (baud rate, start bits, stop bits and parity bit) in the OS.
10. Simple Serial Ports Advantages:  Simple to use and can have fine grained control  Can be extremely robust to environmental influences  Chipsets can be “rad hard”  Can easily connect to very simple microcontrollers and hardware devices  Latency is well defined Disadvantages:  Will have to code any “networking stack” that you need yourself  Low baud rate by today’s standards  Limited range by today’s standards  Some cabling can be heavy by today's standards
11. “Stream” File Processing Description:  Treat the file like a set of information packets, reading one “packet” at a time and processing it. Modern Day Approach:  Read the whole file into a large array in memory and process the array. How To:  Open the file just as you would normally  Instead of reading the whole file into memory at once read just as much as you need for a single “work unit” of processing  Keep reading the file and processing “work units” until you reach the end of the file
12. “Stream” File Processing Advantages:  In most cases code ends up being no more complex than working from memory – but it is different.  The resulting code can typically be used on any type of “stream” including a network socket.  Get some parallelism for free:  The OS will buffer and cache access to the file for you automatically  Can process the “work unit” while the OS works in the background on the file  Can handle files of any size, including live streams that run forever! Disadvantages:  Random access to the data structure is harder in some cases.  Some parsing operations will benefit from a state machine implementation which may be harder for some.
13. Blinking LED & Debug GPIO Description:  Configure a simple GPIO line as an output and toggle it from within your code. This is most effective during “board bring up”. Modern Day Approach:  Using the JTAG connected debugger and/or debug serial port of the device. How To:  Start by constructing an absolute minimum program needed, or use an existing example program.  Configure a single line (or port whichever is simpler) to be an output.  Set the output high and then low, possibly repeating at some interval.  Monitor the state of the output line with a scope, logic analyzer or simply connect a low power LED to the line (exact circuit will depend on the hardware being used).
14. Blinking LED & Debug GPIO Advantages:  Extremely simple to get working compared to other approaches.  Extremely low timing latency which can be critical for debugging some issues.  Multiple lines can be used to convey system state in real time.  Can be used to debug issues where the processor will not stay running (think watch dog).  Used to be the “standard” first test running when bringing up a new board.  Can work when the JTAG (or other debugging link) is unstable. Disadvantages:  Requires external hardware of some type to monitor.  Requires some basic electronics knowledge to set up.  Limited information can be sent.
15. “Bare Metal” Coding Description:  Creating your application code directly “against” the hardware without using a framework or operating system. Modern Day Approach:  Use a “small” ARM processor and put Linux on it. How To:  Create a description of what you are trying to do (requirements).  If the application is simple, or has “hard real time” requirements consider working directly “against” the hardware.  Exact details will vary greatly depending on what you are trying to do and could be several additional talks.
16. “Bare Metal” Coding Advantages:  Can implement more “hard real time” applications.  Can be simpler and faster development in some cases.  Can use lower cost hardware.  Can improve power consumption.  Can improve boot time.  Fine grained access and control of all available hardware.  Can use interrupts and DMA engines to simulate a multi- threaded application to improve determinism and latency. Disadvantages:  May require more coding to implement “device drivers”.  Development and debugging tools more limited.  You will be doing “embedded systems” development by it’s vary nature.
17. Hand Optimized Functions Description:  Creating a dedicated function that does only what you need it to do in place of using a library function. Modern Day Approach:  I need to do something – where is the library that does this for me? How To:  First consider what is REALLY special about what you need to do!  In most cases just go get the off the shelf library and use it!  Profile and evaluate the available libraries and be sure they cannot be made to work for you.  Profile and evaluate the available libraries again and be sure they cannot be made to work for you.  Ask a friend/co-worker to get a second, third and forth opinion.  Make sure you know WHY the existing solutions won’t work in your case and have some idea of how to improve the situation.  Ok if you made it to here and you still think there is something special about what you need to do – code it up and make sure it works better for your use case!
18. Hand Optimized Functions Advantages:  Will be completely tailored to exactly what you need to do, and nothing else saving overhead.  You might come up with a way better way of doing something and get some great “kudos” for it! Disadvantages:  Will take considerably more time to develop, debug and maintain.  You might spend a lot of time attempting to develop the solution only to find out your solution is no better than the available solution.  Coworkers are going to make you justify why you did it yourself!
19. Simply Installed Programs Description:  Creating a statically linked program that consists of ONE file only containing everything the application needs to run. Modern Day Approach:  Create a large set of files that need to be installed and then either create an installer application or wrap everything into a Docker container. How To:  Evaluate if a single file is really appropriate for your application.  Put development processes in place that control the addition of dynamic linking, individual configuration files, registry settings and such. (Should have these anyway!)  There are many way to embed binary data into an executable file if needed, and some frameworks support this directly.  “The Power of Compile Time Resources”, Jason Turner https://www.youtube.com/watch?v=3aRZZxpJ_fc  On the “first run” the single executable can setup registry settings and other needed configurations in place of having an installer.
20. Simply Installed Programs Advantages:  Application is easily moved around.  No installer needed.  No dependency issues, or DLL hell.  Application development can be simpler in some cases.  Application performance can be faster in some cases. Disadvantages:  Architecture of the application will be limited.  When there is an update the whole application will need to be replaced.  Not all dependencies that you may need to use can be packaged into an executable.
21. Cardiac Cardboard Computer Contributed by James Newton A computer made from cardboard that was used for training. Simulated all the basic operations of a very simple CPU by sliding cardboard members.
22. Cardiac Cardboard Computer Developed by James Newton Has been “reborn” as a website developed by James Newton: https://jamesnewton.github.io/cardiac/
23. Open Discussion & Q & A I’m sure there are other techniques that I have not covered, any stories from the audience….