Embedded Muse 165 Copyright 2008 TGG September 15, 2008

EDITOR: Jack Ganssle, jack@ganssle.com

CONTENTS:
- Editor’s Notes
- Inheritance
- Funny Datasheets
- Code Inspections
- Comm Monitors
- Jobs!
- Joke for the Week
- About The Embedded Muse

****************************************************************
This issue of The Embedded Muse is sponsored by Netrino.

The job of writing reliable and maintainable embedded software is not rocket science. So why are there so many bugs in your company's firmware? Part of the problem has been the lack of quality training. Netrino has solved this by packing valuable lessons on the few dozen coding best practices that most reduce firmware bugs (in programs with and without an RTOS) into a 4-1/2 day hands-on Embedded Software Boot Camp. The next public session runs October 6-10, 2008.

Visit http://www.NetrinoInstitute.com/Boot-Camp-Muse for full details.

****************************************************************

Editor’s Notes

Are you in the San Jose, Phoenix or Austin areas? I’ll present a public version of the Better Firmware Faster class in San Jose on December 8, Phoenix on December 10 and Austin on December 12. Registration and other info here: https://www.ganssle.com/classes.htm . You’ll earn 0.7 Continuing Education Units, learn a lot, and have more than a little fun. Sign up before November 8 and receive a $50.00 discount.

The Renesas Developer Conference will be in San Diego October 13-15. It's a heavily technical, hands-on, conference with sessions on CAN/LIN, Zigbee, RTOS, Ethernet, USB, Flash Technology, etc. They're also going to have the Mythbusters there for a special session. See http://devconrenesas.com .

Venture Development Corporation (VDC) has completed its 2008 embedded systems engineering survey. To download a copy of the highlights, see: http://www.vdc-corp.com/misc/08_esdt_eu_survey_highlights.pdf .

Inheritance

Bird "is a" Animal -- Inheritance
Bird "has a" Rock -- Containment
Bird "has a" Beak -- Containment

Circuit Board "is a" Insulating Material
Circuit Board "has a" Resistor
Circuit Board "has a" Battery"

But there are other things that people tend to screw up beyond containment/inheritance. It's possible to get that part right and still have bad decomposition of objects. The second set of examples is deliberately constructed to have some additional potential problems

For one thing, what is a Circuit Board? Is it the blank PWB or is it the stuffed board? If it's poorly named, we might make bad assumptions (and worse we might mix our assumptions without realizing it). If I assume that Circuit Board is a blank PWB then inheriting from Insulating Material might make sense but containment of the resistor doesn't make sense. But if you assume that it's the stuffed board, then inheriting from Insulating Material wouldn't be right but the fact that it contains a resistor would make sense. (It might be better to say that Stuffed Board "has a" Blank PWB which "is a" Insulating Material.)

James Grenning wrote: The notion that OO's goal is to model the real world is not really up to date with current OO thinking. Some do use that way, and it does not usually lead to better designed software. The main goal of OO is dependency management, information hiding through encapsulation and the ability to define and program to interfaces. The typical problem of long dependency chains can be very effectively broken by inserting an interface. In C++ an interface is a base class with all pure virtual implementation. Essentially you get to define what an object is supposed to do, without revealing how it will do it. When a compile time dependency is broken by inserting an interface, essentially the top down dependency from client to server is inverted, making both the client and the server depend on the interface, without knowledge of the other.

Being able to model things form the problem domain is a bonus on top of dependency management.

Funny Datasheets

Code Inspections

The code reviews cover a period of one and a half year, ending mid august 2008. The code in question is C code for an embedded 8051, controlling the radio part of the chip and implementing the lower layers of the protocol stack, developed by a team of four to five persons.

(Jack notes: It’s hard to put the data into this text-formatted newsletter, so it’s on-line at https://www.ganssle.com/misc/nordic_reviews.xls . There were 23 reviews of a total of 66 files which uncovered 399 errors, or about 6 per file on average.)

The rule is that before a review, the code shall compile cleanly and give no Lint warnings. The issues and errors found during the review are classified as major, minor, code that is correct but still should be improved, specification violations and code standard violations. In addition, some issues have not been classified. Both the number of issues and their classifications should be taken with a grain or two of salt, though. (Reviews have been done at different times, with different reviewers, some issues are difficult to classify, ...) Some issues are given multiple classifications, but they are only counted once in the "Total" column.

A few more things:
The code reviews are most often done by four persons (moderator, recorder, reader, author) who meet after preparations and read through the code, paraphrasing it. Another variant used include two reviewers reading the code thoroughly on their own, and then a short meeting with the author where the reviewers' give their comments to the code.

On average, most of three to four man days are spent on a review. For the reviews in this collection, a grand total of around eighty man days is a quite good estimate of the time used.

In an email exchange Asbjorn answered some questions for me:

> - About how many lines of code were reviewed?

That we have not tracked, actually. (But I can see that it might be a good idea to do so.) A rough estimate would be some hundred lines of code (not including comments and white space), maybe a bit more, per review on average.

> - Do you know how many bugs remained and were fixed after the reviews?

Not per reviewed module. But we have found around 45 bugs in total on this part of the project, that is on all the reviewed code together. These 45 bugs are typically functional errors (system does not work, does not work as intended, does not follow specification), often related to the interworkings of various software and hardware modules.

> - Is the team satisfied with the results?

Opinions are a bit divided, I think. Some feel that reviews are taking much time, others seem quite satisfied. This overview of the reviews was intended as input into this discussion, to see how many bugs we actually did catch and how useful the reviews were.

We have been trying various approaches to the reviews, from the full reviews (four persons in meeting, paraphrase all of code) via somewhat lighter reviews (two other persons read the code on their own, making comments on errors and things they do not understand) to a few very informal ones on smaller pieces ("Hey, could you have a look at this?"). I think the discussion on, and search for, the "best" way of doing reviews for us will continue for some time yet.

Comm Monitors

Tjark van Dijk wrote: I followed the subject about protocol analyzing tools and maybe this product (indeed, one of my own) may be of interest to your audience. http://www.tildesign.nl/rs232_ttl_converter-20107101P2N .It's an RS232 to TTL converter, just Rx and Tx. The reason to develop a product like this was that I became soo tired making again an RS232-TTL converter on a breadboard because the controller board I had to work with just output TTL. And off the shelve solutions cost around $ 50 or more and were large.

If a Muse reader wants to order and tells me MUSEDISCOUNT in the order they can buy the unit for 11 Euros plus shipping in any quantity.

Jobs!

Joke for the Week