[Fwd: Re: Proposal: offset based member name (or type) lookup]

[Christoph Jaeschke]

> Daniel Jacobowitz wrote:
>> On Sun, Jan 04, 2004 at 12:25:37PM -0500, Daniel Jacobowitz wrote:
>> 
>>>On Sun, Jan 04, 2004 at 06:56:05PM +0100, Christoph Jaeschke wrote:
>>>
>>>>Hi,
>>>>
>>>>I would like to propose a offset based member name lookup of
>>>>structs/classes. Assume:
>>>>
>>>>    struct X {
>>>>        int a;
>>>>        int b;
>>>>    };
>>>>
>>>>    struct Y {
>>>>        struct X x[4];
>>>>        int c;
>>>>    };
>>>>
>>>>    struct Z {
>>>>        Y  y;
>>>>    };
>>>>
>>>>If you apply a patch for gdb 6.0 I've prepared, you can ask gdb by
>>>>
>>>>    ptype Z + 28
>>>>
>>>>about Z's relative name at offset 28
>>>>
>>>>    .y.x[3].b
>>>>
>>>>using
>>>>
>>>>    ptype Z+28, typechain
>>>>
>>>>gdb will append also the typechain
>>>>
>>>>    .y.x[3].b, typechain = ::Y::X[4]::int
>>>>
>>>>
>>>>There may be a better suited command than 'ptype' for it, it was choosen 
>>>>just because it was the simplest way to add it.
>>>>
>>>>The patch contains also a caching mechanism, speeding up a lot if you do 
>>>>many lookups in big nested structures. It has a minimal memory 
>>>>requirement and will not hurt for single lookups.
>>>>
>>>>The patch can be send if you are interested in.
>>>
>>>Well, I don't like the syntax, but I think this is a wonderful idea.
>>>The other feature I've been meaning to add since forever is a variant
>>>of ptype which shows byte offsets for every field.
>>>
>>>The problem with using ptype is that "Z+28" already has a meaning; gdb
>>>will try to add the two, or call an overloaded + operator in C++, et
>>>cetera.  This should probably be a new command; that'll be more useful
>>>for MI anyway.
>> 
>> 
>> Oh, and I forgot the obligatory followup to this: if you are
>> interested in contributing this patch, and I hope you are, then please
>> file an FSF copyright assignment.  Jim or Andrew can send you the
>> appropriate forms to get the process started.
>> 
> 
> Sure, I'm interested in contributing. But, because I'm not very familiar 
> with the command interface (that's why ptype was re-used), I would be 
> happy if someone else create a new command and just call the processing 
> func.
> 
> If Jim or Andrew send me the forms, you will get the patch.
> 

BTW, assuming also

	...

	struct Z {
	    Y  y;
	};

	Z z;

someone may enhance the 'info symbol' command (using 'whatis' and the 
patch) to fully resolve the name at some address, for example

	(gdb) info symbol 0x8049530
	z + 16 in section .bss
	(gdb) whatis z
	type = Z
	(gdb) ptype Z+16
	.y.x[2].a
	(gdb)

concatenating to:

	z.y.x[2].a