Re: [RFC] Don't allow setting register in non-innermost frame

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Yao" == Yao Qi <yao@codesourcery.com> writes:

Yao> I checked the difference of value of register and value of variable
Yao> (saved in register), TYPE_OBJFILE_OWNED (value_type (toval)) is the
Yao> only difference between them.  IIUC, type of variable is owned by an
Yao> objfile, while type of register is not (alloc_type vs. alloc_type_arch),
Yao> so looks we can use it to differentiate values in register and values
Yao> in variables.

I would rather not use this approach.
My reason is that there is no obvious connection between
TYPE_OBJFILE_OWNED and register-ness -- and it is the sort of invariant
that is very difficult to ensure will remain true over time.

If lval_register can't work, then another choice would be a new flag on
struct value.  This would be somewhat ugly but, I think, more robust.

As Jan points out, it does seem strange to warn about the set rather
than query.  A warning comes too late.

After re-reading the whole thread I wonder whether pressing forward with
this patch is best.  IIUC this didn't arise from any user complaint but
rather just thinking about the problem.  And, while I agree it could be
confusing in some situations, in practice it seems that perhaps it has
never actually bitten anybody.

Tom