Re: Linux kdb (was Re: first thread)

["Matthias Urlichs" <smurf@noris.de>]

Hi,

John Alvord:
> The main reason it hasn't been done is that Linus is absolutely opposed.
> He believes that fixes developed from crash dumps typically cure symptoms
> rather than correct the underlying problem. That seems a bit weird to me,
> but has been expressed several times.
> 
A crash dump exposes too much of the static situation at the time the crash
happened, but not the dynamic events which led to it. Therefore it leads to
static fixes ("dereferenced NULL? Slap 'if(!ptr)return;' onto it!") without
examining the underlying reason ("The thing is NULL? It's not supposed to!"
Scratch head, come up with a way to reproduce the problem, insert a couple
of debugging printf() calls, eventually come up with the _right_ solution).

I dimly remember a few bug fixes during 2.1 development (I usually browse
the patches) where that is exactly what happened and where the obvious
solution, like the above if(!ptr), would quite obviously (in hindsight ;-) 
have caused an even bigger mess in the long term.

> A secondary reason is that people are worried that the attempt to write
> out the crash data will destroy user data, since the system is unstable.
> 
Doing that safely is _not_ easy.

-- 
Matthias Urlichs  |  noris network GmbH   |   smurf@noris.de  |  ICQ: 20193661
The quote was selected randomly. Really.    |      http://www.noris.de/~smurf/
-- 
I have read your book and much like it.
	-- Moses Hadas (1900-1966)
-
Linux-future: thinking about the future of the Linux kernel
http://humbolt.nl.linux.org/lists/