uninitialized typedef in value_sub on pointers

[Kevin Nomura]

Sometimes the second operand of a pointer subtraction has
incomplete type info loaded, resulting in a bogus but
self-repairing error as demonstrated by:

(gdb)  p (sk_Proc *) fm_givebackPid-sk_proc_table
First argument of `-' is a pointer and second argument is neither
an integer nor a pointer of the same type.
>>> this next line was retrived with the uparrow, and a single space
>>added at the end, which seems to make a difference:
(gdb)  p (sk_Proc *) fm_givebackPid-sk_proc_table 
$1 = 0


Reproduces with gdb 4.18 and gdb 5.0, but not with the small
testcase I tried to craft.

I made the following patch to gdb-5.0/valarith.c:value_sub() to
get around this problem:

113,114c113,114
<              && TYPE_LENGTH (check_typedef (TYPE_TARGET_TYPE (type1)))
<              == TYPE_LENGTH (check_typedef (TYPE_TARGET_TYPE (type2))))
---
>              && TYPE_LENGTH (TYPE_TARGET_TYPE (type1))
>              == TYPE_LENGTH (TYPE_TARGET_TYPE (type2)))


Perhaps someone with more knowledge of the way gdb loads symbol
info can divine the true nature of this bug and/or whether this
is an appropriate fix.

Kevin