Re: Win2000 has 63,000 'defects'

[Gautier <gautier.demontmollin@maths.unine.ch>]

Hyman Rosen:

> > Someone to sell or install them poor an Ada compiler ? ;o) G.

> Why do you think the defects have anything to do with the language
> used to develop Windows 2000? Do you know that the defects are
> coding errors, as opposed to being design errors or unimplemented
> features?

That's not so simple... Some coding errors may be catched at compile time
in one language and appear as an bad crash on an user's screen months
later when coded in another one. As complexity grows, uncatched (coding
or design) bugs tend to combine with other. In some way the "shape"
of a bug disappears rapidly with the complexity and the time elapsed
after the coding.

Some concrete examples. In the code I'm developing (a comparison of
finite element methods for multiple partial differential equations)
there are *lots* of coding issues that have been catched by the compiler
because I'm using Ada's rich typing: simple things like exchanging
the equation and node index, and so on.

You forget that design features help avoiding other *bigger lots* of coding bugs.
Back to the example (a tiny excerpt)...

...
procedure FT_elem(gelem: t_element) is

  nsd:      constant positive:= spatial_dimension(gelem);
  iel:      constant positive:= nodes_per_element(gelem);
  nb_faces: constant positive:= faces_per_element(gelem);

  subtype sp_vector  is vector(1..nsd);         -- vecteur spatial
  subtype sp_matrix  is matrix(1..nsd,1..nsd);  -- matrice spatiale (codecond)
  subtype el_array   is vector(1..iel);         -- tableau sur un element
  subtype el_iarray  is ivector(1..iel);        -- idem, composantes entieres
  subtype el_matrix  is matrix(1..iel,1..iel);  -- matrice elementaire
  subtype el_grad    is matrix(1..iel,1..nsd);  -- gradients sur l'element
...

Needless to comment!

An extreme but true example: a program (some environmental predictor) had
a strange crash recently. There was a line in the 15-year old Fortran code that
"almost" never occured, where a variable name had 2 letters exchanged!...

2 things I'm sure of:
 - every mean is good to track a bug, and I prefer to have a compile-time error
   than a day of Sherlock-Homesque bug tracking with a debugger, after a week of
   disfunctionment
 - the bugfest tends to grow exponentially or factorially or something like that
   (specialists know that better) with code length (and its ratio of bugs...).

BTW you can make a naive probabilistic calculation. Say you have 1000-line
modules. Each of them has a probability of 0.00001 of crashing per hour because of 
an uncatched bug. Now you have a 30,000 module program (say Windows 2000) that
uses all of them during an hour. Each module can crash or not during the hour
independently of each other (a generous hypothesis since bugs normally help
each other). Then, the probability of a crash for the whole program during an hour
is 1.0 - (1.0-0.00001) ** 30_000 = 0.259 (26%).  8-|

Gautier

_____\\________________\_______\_________
http://members.xoom.com/gdemont/gsoft.htm