Embedded Systems Engineering Standards Column vol 13.6 September 2006 Snake oil . By Chris Hills

 

These are my own personal views and not those of my company Phaedrus Systems. www.phaedsys.com which is where the full version of this column resides under the Technical Papers button.

Having had a break from the column, life has been busy, I thought I would cover something controversial but first my annual suggestion that everyone takes a break from the pressures of work. See the other columns on the web site about that. Summer 2005 column, summer 2004 Column

However, it bears repeating, that the mind as much as the body needs time to recharge. The head master at my grammar school always used to end his summer holiday address to the school with “…read a good book”. The other adage, “a change is as good as a rest”, reminds me that a change of environment is good too. So a day at ESS in October at the NEC Birmingham will also do you the world of good. The bookings are way up for this point in the schedule and the show looks like being bigger than last year. Well worth a day out, just to get ideas.

I did some seminars earlier this year (With ARROW on ARM) and it was surprising how many of the Engineers, some quite experienced, did not know they did not know (ref1) some of the options in the way of MCU, tools and software such as RTOS, other sw components and methods. So a trip to ESS will be a day well spent as you don’t know what you don’t know. I will have a stand there and of course I will have decent coffee again, bring your own doughnuts, come and have a chat but please visit all the other stands as well. I got in to trouble last year as one or two people thought my requests to come to the show and stop at my stand was giving me an unfair commercial advantage… I did not think that people would come to the show just to see my stand but would also wander round all of the stands. It is worth doing that anyway as you never know who or what might be there.

Recently I have managed to get into a bit of an argument over the GCC compilers. The aficionados of GCC tell me that it is better for several reasons, though performance isn’t one of them. One is that as you have or can get the source “of any version” you can re-build any old version of the compiler. Whilst this is true unless it is re-compiled with an identical binary of the original compiler used to build it the result is not going to be the same. The new “old version” will be a different beast and behave differently. If this was not the case people would not need older versions of compilers for maintenance on legacy project’s in the first place.

The chances of having an already built version of the original compiler used to compile the old version of GCC are quite low. So it will be almost impossible to recreate an exact copy of an old version of a GCC compiler. The argument used was that you couldn’t get older versions of commercial compilers. This is not true. At least it is not true for all the commercial compilers I know. More to the point the commercial companies have all their sw under [version] control and keep binaries of all versions they produced so they can issue an exact version. Also the issue of dongles etc is normally not a problem on the old versions of compilers despite the claims of the open-source camp.

The next argument was that with the open-source you could fix bugs… This is extremely unlikely for several reasons. Firstly compilers are complex things, though I am informed from several sources that GCC core is only about 5 years behind the average mainstream commercial compilers it is still a complex beast. The chances of a person not involved in the development fixing a bug without introducing others or having other side effects is quite low. The more important point it that, after many years of answering support calls for compilers I have found that large number of “compiler bugs” I have had to deal with were in fact not bugs just a miss understanding of correct C behaviour. Also given the arguments in the C language reflectors it is unlikely that anyone not in the C standards loop is going to have a the correct information to fix compilers. Yes, some of the GCC core maintainers are on the C standards panels. Though it has to be said there are several companies that maintain their own versions of the core.

Finally, If you fix “bugs” or rebuild an old version from source you will need to fully retest the compiler you have built. Commercial compilers are usually rigorously tested against industry standard test suites such as Perennial and Plum-Hall and other systems such as Paranoia that test specific areas of the libraries as well as rigorous in house test suites. As far as I know GCC is not tested against these industry standards as the commercial test suites cost a lot of money.

You could use the GCC test suite but this only really confirms it has built correctly and conforms to GCC C, not to ISO-C. You also need the correct version of the test suit for the compiler and be able to show that the test suite used is a valid file and not an edited one. The other point is that even if you did test a version of the GCC it would only apply to that particular binary, or copies of the binary not the source. As soon as you did any changes to the source you need to retest the whole compiler.

Part of the problem here is one of liability. If you use a compiler you have modified or built yourself you are going to have more of a problem showing due diligence than using a built and tested commercial compiler. This is because you will have to show that the compiler you have built is fit for purpose.

Interestingly there is a GCC that is available for safety critical work… It is supplied as a binary and it is maintained and tested by the company that supplies it as part of a package of other software. It is not inexpensive either. If you use any gcc bar the one they supply then all the guarantees are invalidated. Actually they are not alone. Several companies supply gcc packages though not for safety critical work, at a cost. Technically the cost is for the, IDE, the support tools, the installation system or the support. In effect gcc is becoming commercial and should be judged against the other commercial tools as such. Well, I have dug myself a hole and will probably get buried by the disciples of GCC.

In a similar vein I saw an article in Computing (August 10th issue) about cryptographic software. The author was complaining that consumers of crypto SW had no real idea what was good and what was bad crypto implementations. The trouble was most of the poor quality systems were being sold at a much lower price than the high quality systems. The problem as the author saw it more people were using low priced, low quality crypto software whilst at the same time reducing the market for the more expensive yet quality software by either lower sales or drastically reducing their prices. The net result is the spread of lower quality crypto and the reduction of the high quality vendors in the market.

The IT industry is the only one I have come across where it’s practitioners delight in using low quality and cheap tools. I find it very strange. There are parallels with the building trade where carpenters, electricians etc buy the best tools they can but the cowboys and DIYers buy the cheaper tools. If we all followed the same trend into our private lives we would all be driving Trabant’s.

On the standards front I hope to have some MISRA-C news shortly re the TC and the example suite. The MISRA C++ team are well into their work as are the MISRA-Autocode team. However the MISRA-UML team are still working out their frame of reference. If anyone is interested in working on MISRA-UML please email me.

I can report that due to a minor spat between two members of the BSI C panel on the panel email reflector one got suspended which lead to more exchanges which lead to the convener suspending all of the panel and then resigning after 8 months in the job... So there is currently no BSI C panel. Hey Ho! I put it down to the weather.

Finally according to The Register the humble PC is 25 years old last month. Can you remember where you were when IBM launched it in summer of 1981? The PC had a character based monochrome screen, no colour let alone graphics, 640K RAM and ran at 4.7mHz! Incidentally it also had a cassette tape interface and no hard drive as I recall. Apparently PC World has one on display in their branch at Staples Corner in London. Whilst almost every part of the PC has changed most are still beige. The thing that made the PC so popular was the fact that IBM forgot to patent it. This meant that others could produce them too. IBM did patent the MCA, OS2 and PS2 which did not go far. Come to that neither did any non-standard PC's. The answer is standard interfaces and if you must do a new one you need an industry group to back it to give it critical mass.

(1)There are known knowns, there are known unknowns and there are unknown unknowns.

That is that is there are things we know that we know. There are things we know that we don’t know. There are also things we don’t know that we don’t know.

However, I would also add that there are things we know but we don’t know that we know them…

 

 

Author Details and contact

 

Eur Ing Chris Hills BSc CEng MIET MBCS MIEEE  FRGS   FRSA is a Technical Specialist and can be reached at This Contact

 

Copyright Chris A Hills  2003 -2008
The right of Chris A Hills to be identified as the author of this work has been asserted by him in accordance with the Copyright, Designs and Patents Act 1988