Hi all,

I've been working on getting debugging (dynamic) arrays working with fpc
(free pascal compiler)

Attached is the patch I have so far, including some tests. All
pascal-arrays I tested work now, but it could very well be that I broke
something else. I'll try to explain that I did... (patch is based on the
vla-branch)

Some changes are quite fundamental. Most important change is that the
length of a variable is not longer only stored in the type struct, but
also in the value struct. 

This is done because with Dwarf-debuginfo, the size of a variable can be
dependent on some dwarf-blocks that ask for the evaluation of some data.
So the size of a type-struct is not defined. You always need a
data-address with it. That's the case in the value-struct.

Problem is that there are all sort of hacks to work around this problem,
and I tried to keep the code working with these hacks. One of the
problems is that in some cases, the memory for the variable is already
allocated before the object-address is set. In that case the
'old'-behaviour is used: the length of the struct-type. This length is
also cached, so that when the value-length is asked again, the same
value is returned as the first time. This is necessary because when
memory of the given size is allocated, the given size can not change
anymore offcourse.

Further array-bounds are now also part of the value struct. And
check_typedef is splitted into two parts, one part that searches for the
real type, not the references. And the second part that evaluates and
copies the data to a new struct. This is because I need the first part
without the second on a few places.

How do you think about these changes?

Joost