Luke Kanies wrote:

I fully plan to make dangerous tools, tools that allow you to be as stupid or as smart as you like. If you're stupid, you'll break your network, and your boss will fire you; hopefully he'll be smart enough to hire someone in your place who isn't dumb enough to use tools in stupid ways.

There seems to be a misunderstanding here. Safety is not about having tools that are limited, but instead about having tools that are aware to some extent of the effects of their actions and can predict the results of changes well enough to protect their own infrastructure and avoid getting the configuration management infrastructure into irrecoverable
states. If you insist upon making a tool that -- by nature -- can easily
get the system into an irrecoverable state, then I don't see why people
should use it. That's power without responsibility.  Not being able to
blow away the management infrastructure seems a reasonable "limit".

This is not to say that this limit is built in. It is instead a policy.

In fact, I have a really good example of limits made powerful.
My tool "slink", described in my 1996 paper, allowed one to
specify a modification policy that the tool wouldn't violate.
This allowed us to collaborate for years on maintaining shared
software repositories with *no* danger of corrupting system files
via a misconfiguration. This saved our tails many times. The only
reason we stopped using it was to move to cfengine, and in doing so,
we *lost* a safety factor; we can no longer protect the system files
from configuration mistakes (except through mounting them without
root!).

In other words, safety is not about "making languages limited", but
instead about "being able to specify policies that assure safety
and limit scope of action".

--
Dr. Alva L. Couch
Associate Professor of Computer Science
Tufts University
Medford, MA 02155
http://www.cs.tufts.edu/~couch

_______________________________________________
lssconf-discuss mailing list
[email protected]
http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss

Reply via email to