[SystemSafety] A small taste of what we're up against

[Les Chambers]

RE: Martyn's unassailable truth

"The economic incentives to conform to standards and to go far beyond them are very strong in avionics, which is 
perhaps one reason why the take-up of rigourous software engineering has been stronger there than in most other 
areas of software development."

To support and amplify:

The most potent drivers of safe software are:
1. The skills and motivation of the people writing the code
2. The processes they use to cooperate

For me, the language they use is a hygiene factor somewhere further down on the list. 

In 1980 my software was controlling a very large chemical reactor with tasks written in assembler. The operating 
system had no file or memory management. To load a task you typed in a disk track, sector, word count and the 
required memory location. The control panel was 30 feet long. On it were 10 indicators that would give you an 
early warning of problems. Peripheral vision expanded into wide angle mode, you hit return and endured the 
eternal one second delay before the operating system picked up your new task. Do not try this at home.
I was not special. There were at least 30 engineers doing exactly the same all over the world. I was not aware of 
any problems in the three years I worked with that technology. My only near miss was due to my failure to review 
every stick of code loaded by others on that control computer.

It's interesting to observe the level of passion unleashed on this list over the C language (this is tech passion). 
Would that it be applied to preventing problems before they arise (software engineer education) and catching them 
as close as possible to their injection (intelligent coaches, code analysis/inspection). Could it be our profession 
needs more techs with a philosophical bent? You need philosophers if you want your profession to have GUTs 
(grand unified theories).

I recently listened to a Google robot scheduling a hair appointment. The hairdresser seemed unaware she was 
talking to a robot. The conversation was complex. The robot sounded creepily human (no uncanny valley 
detectable). Would that the raw brainpower applied to achieving this be applied to monitoring code as as it is 
entered. Machine learning should be a snap. We have the massive datasets of good and bad code. Compared to 
conducting a human conversation this should be a simpler problem. Where are the intelligent agents? Has anyone 
seen one? When will our profession stop pandering to the trivia demanded by the mass market (booking hair 
appointments???) and engage with the real toil of making software safe for everyone - not just avionics users.  
When will we stop writing bodice-rippers and start producing literature across the board?

Meanwhile Martyn, I support your short prayer before take off. Don't forget to match it with a prayer of thanks on 
landing. I do. 

Les



--
Les Chambers
les at chambers.com.au
+61 (0)412 648 992