> Ah, this version is much beter.
> 
> The results are:
> 
>                         PASS  FAIL
>   gcc 2.95.3 -gdwarf-2  290   3
>   gcc 2.95.3 -gstabs+   290   3
>   gcc 3.3.2  -gdwarf-2  290   3
>   gcc 3.3.2  -gstabs+   290   3
> 
> The three FAIL results are:
> 
>   p/c fun1()^M
>   $1 = {a = 0x08044004c400000000000000}^M
>   (gdb) FAIL: gdb.base/structs.exp: p/c fun1() for call-tld 1
> 
>   p/c L1^M
>   $3 = {a = 0x00004004c400000000000000}^M
>   (gdb) FAIL: gdb.base/structs.exp: p/c L1 for call-tld 1
> 
>   p/c L1^M
>   $1 = {a = 0 '\0'}^M
>   (gdb) FAIL: gdb.base/structs.exp: p/c L1 for return-tc
> 
> All the test names are unique.

Here's today's version (...).  The big change is that I've managed to 
reduce the number of different programs being built - each variant is 
now only built/run once - should help with remote testing.

As for how xfail/kfail are specified, it isn't pretty but it appears to 
work :-(  See if I can think of a simplification.

Andrew