Getting rid of the symbol_ranges and aliases stuff

Getting rid of the symbol_ranges and aliases stuff

[Daniel Berlin]

If I implement the LOC_COMPUTED and LOC_COMPUTED_ARG, does anyone object 
to me removing SYMBOL_RANGES and SYMBOL_ALIASES in the process?

Reasons:
1. The only thing that generates this is redhat internal, and I don't 
think that gcc even has it enabled by default.
2. It wastes 8 bytes in every symbol (because it's implemented at the 
wrong level).
3. It's STABS specific.
4. It's subsumed by LOC_COMPUTED.
5. It's probably not been tested in a *long* time, since only redhat 
internal gcc can generate the info, and it's not done by default.


This is the live range splitting extensions to stabs stuff.  It works by 
having multiple symbols that are aliases to a given symbol at a given PC. 
In effect, it uses multiple symbols to simulate DWARF2 location lists.
It could be reimplemented in terms of location functions, cleanly.

But since nothing publicly available generates the info, I can't do that 
as part of a LOC_COMPUTED and LOC_COMPUTED_ARG implementation.
Nor do I think I should be required to.


--Dan

Re: Getting rid of the symbol_ranges and aliases stuff

[Jim Blandy]

Daniel Berlin <dan@dberlin.org> writes:
> This is the live range splitting extensions to stabs stuff.  It works by 
> having multiple symbols that are aliases to a given symbol at a given PC. 
> In effect, it uses multiple symbols to simulate DWARF2 location lists.
> It could be reimplemented in terms of location functions, cleanly.
> 
> But since nothing publicly available generates the info, I can't do that 
> as part of a LOC_COMPUTED and LOC_COMPUTED_ARG implementation.
> Nor do I think I should be required to.

You should just ignore the old LRS stuff.  The Dwarf 2 location lists
should be single symbols with the LOC_COMPUTED address class with a
`read_variable' method which use the PC from the frame it recieves to
choose the right location expression.

I guess I don't see how the old LRS stuff is in your way.

Re: Getting rid of the symbol_ranges and aliases stuff

[Daniel Berlin]

On 5 Apr 2002, Jim Blandy wrote:

> 
> Daniel Berlin <dan@dberlin.org> writes:
> > This is the live range splitting extensions to stabs stuff.  It works by 
> > having multiple symbols that are aliases to a given symbol at a given PC. 
> > In effect, it uses multiple symbols to simulate DWARF2 location lists.
> > It could be reimplemented in terms of location functions, cleanly.
> > 
> > But since nothing publicly available generates the info, I can't do that 
> > as part of a LOC_COMPUTED and LOC_COMPUTED_ARG implementation.
> > Nor do I think I should be required to.
> 
> You should just ignore the old LRS stuff.  The Dwarf 2 location lists
> should be single symbols with the LOC_COMPUTED address class with a
> `read_variable' method which use the PC from the frame it recieves to
> choose the right location expression.
> 
Yes, I know.

> I guess I don't see how the old LRS stuff is in your way.

Um, if I remove everything but LOC_COMPUTED and LOC_COMPUTED_ARG, and 
replace them with sets of functions, it means the live range stuff 
needs to be dealt with.

It's not in the way for the first patch (which just adds LOC_COMPUTED and 
LOC_COMPUTED_ARG), but it'll be in the way soon after, as things are 
converted to use them.

If you look, I said I can't convert them as part of replacing 
the existing classes with LOC_COMPUTED and LOC_COMPUTED_ARG, and i don't 
think i should be required to.

From a practical perspective would be easier to remove them now, than 
remove them as part of a followup patch to the LOC_COMPUTED introduction 
that does the conversion (which is when they'd be in the way).
It's cleaner.