It's back, the negative dimension problem - explanation, of sorts

[Toon Moene]

[ egcs-1.0.3a rejects DIMENSION AAP(-1:*) ]

>  Yes, it has been solved (again) in the most recent
>  snapshot.There is no negative dim problem in ...531.

I think what basically happened was this:

1. Originally, egcs before the 1.0 release (and any combination
   of g77 + gcc) had terrible code generation for loops over
   arrays on Alpha's.

2. Richard Henderson repaired this by "promoting" index variables
   in the Fortran Frontend to the same size as pointers.

3. After the first release it turned out that this mistreated
   arrays with negative bounds.  This was repaired, both in the
   main development trunk of CVS as well as the release branch.

4. In the mean time, in the main trunk (not the 1.0.x release
   branch) a problem with undefined upper bounds was repaired.

5. Subsequently, the final (hah ;-) problem with negative array
   bounds and index expressions was resolved by doing all
   arithmetic in a signed "type" the same size as the (unsigned)
   pointers.

6. This solution (5.) was backstitched into the release branch
   but nobody detected in time that that solution didn't play
   nice with the way undefined array bounds were handled
   previous of (4.).

HTH (Richard will correct me if this is too revisionist).
Toon.