[Dwarf-discuss] DW_OP_piece limits and assumptions

[Daniel Jacobowitz]

On Wed, Apr 27, 2005 at 06:06:29PM -0500, jeff nelson wrote:
> If you follow the strict interpretation of DW_OP_piece, the expression
> ought to result in TWO values, not one. That is, when you reach the
> second DW_OP_piece, what definition or rule do you use to combine it's
> value with the results of the first DW_OP_piece?
> 
> To resolve this issue, I propose the following addition to the
> description of DW_OP_piece:
> 
>     DW_OP_piece is a context-sensitive operator. For the first occurance
>     of DW_OP_piece in a DWARF expression, the operator result is pushed
>     on the DWARF expression stack. For all remaining occurances, the
>     operator result is appended to the value at the top of the stack. In
>     this way, the results of several consecutive DW_OP_piece operations
>     are combined into a single value.
> 
> Note that:
> a) "appended" is undefined.
> 
> b) I make the simplifying assumption that the the operator result is
> saved on the stack (i.e., leaving questions 1 and 2 above unresolved for
> now).

This is obviously a result of the same disagreements about how location
expressions are handled that came up in the earlier discussion :-)
Could you explain why you believe the result of DW_OP_piece belongs on
the stack?  Here's my interpretation:

Consider this location expression:
	DW_OP_fbreg 20

The resulting object can be of any size, and is never put onto the
stack.  Then consider this one:

	DW_OP_fbreg 20 DW_OP_piece 32 DW_OP_fbreg 70 DW_OP_piece 4

The resulting object is 36 bytes.  Conventionally (though I think this
is not explicit in the standard) it is built up in memory order.  When
we reach the first DW_OP_piece, fp+20 is on the stack; DW_OP_piece
consumes that.  It does not need to place a new value on the stack. 
Instead it records some information for the evaluator of the location
description.  There is no way for further stack operations to get at
the value.

-- 
Daniel Jacobowitz
CodeSourcery, LLC